DETAILED ACTION
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Status of the Application
2. Claims 1-20 have been examined in this application. This communication is the first action on the merits.
Claim Rejections - 35 USC § 101
3. 35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
4. Claims 1-20 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1: Claims 1-20 are focused to a statutory category namely a “method” or a “process” (Claims 1-13), a “system” or an “apparatus” (Claims 14-18) and a “non-transitory computer readable medium” or an “article of manufacture” (Claims 19-20).
Step 2A Prong One: Independent Claims 1, 14 and 19 recite limitations that set forth the abstract idea(s), namely (see in bold except via strikethrough):
“” (see Independent Claim 14);
“” (see Independent Claim 14);
“
“obtaining, , data associated with one or more business units ” (see Independent Claim 1);
“pre-processing, , the data to generate pre-processed data” (see Independent Claim 1);
“determining, , one or more features associated with one or more financial information for one or more horizons based on the pre-processed data using a plurality of input models” (see Independent Claim 1);
“generating, , one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons” (see Independent Claim 1);
“generating, one or more forecasts for each horizon for a pre-determined time interval, using a model comprising one or more forecasting models, wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon” (see Independent Claim 1);
“providing, the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users associated with the one or more users” (see Independent Claim 1);
“ obtain data associated with one or more business units ” (see Independent Claim 14);
“pre-process the data to generate pre-processed data” (see Independent Claim 14);
“determine one or more features associated with one or more financial information for one or more horizons based on the pre-processed data using a plurality of input models” (see Independent Claim 14);
“generate one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons” (see Independent Claim 14);
“generate one or more forecasts for each horizon for a pre-determined time interval, using a model comprising one or more forecasting models, wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon” (see Independent Claim 14);
“provide the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users associated with the one or more users” (see Independent Claim 14);
“obtaining data associated with one or more business units ” (see Independent Claim 19);
“pre-processing the data to generate pre-processed data” (see Independent Claim 19);
“determining one or more features associated with one or more financial information for one or more horizons based on the pre-processed data using a plurality of input models” (see Independent Claim 19);
“generating one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons” (see Independent Claim 19);
“generating one or more forecasts for each horizon for a pre-determined time interval, using a model comprising one or more forecasting models, wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon” (see Independent Claim 19);
“providing the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users associated with the one or more users” (see Independent Claim 19).
Here, for Independent Claims 1, 14 and 19, the claims recite an abstract idea of collecting, analyzing, modeling, and outputting financial cash flow data and forecasts for business units. This abstract idea falls squarely under the judicial exception categories of Mental Processes (evaluating/judging financial metrics via algorithms) and Certain Methods of Organizing Human Activity (managing business operations, commercial interactions, and financial tracking).
For example; (1) the step of “Obtaining data associated with business units from data sources”: Receiving, gathering, or obtaining data is a basic information-gathering step that falls under Mental Processes and Methods of Organizing Human Activities. A human can mentally or manually request and read business records. (2) The step of “Pre-processing the data to generate pre-processed data”: Data cleaning, formatting, or filtering falls under Mental Processes. Humans routinely organize, sort, or clean data before performing an analysis. (3) The step of “Determining features associated with financial information for horizons using AI models”: Determining features, relationships, or markers in financial data is a Mental Processes exception (evaluating information, mathematical calculation, and analysis). (4) The step of “Generating feature combinations by integrating features for each horizon”: Combining, grouping, or cross-referencing features is a mental process and a method of organizing human data/information. (5) The step of “Generating forecasts for each horizon using a stacked AI model”: Forecasting future cash flows or outcomes based on past data is a quintessential Mental Process and mathematical calculation. Humans (financial analysts) perform forecasting manually or mentally. (6) The step of “Providing the generated forecasts for cash flow via user interfaces”: Displaying, reporting, or outputting information to a user fall under Mental Processes and Methods of Organizing Human Activities.
Therefore, other than reciting the additional elements of (e.g., “one or more hardware processors” & “a memory” & “one or more data sources” & “a plurality of artificial intelligence (AI) models” & “ & “a stacked AI model” & “one or more user interfaces” & “one or more electronic devices” & “wherein the plurality of subsystems” & “a data obtaining subsystem” & “a data pre-processing subsystem” & “data obtaining subsystem” & “a feature determining subsystem” & “a feature combination generating subsystem” & “a forecasting generating subsystem” & “an output subsystem”), nothing in the claim elements precludes the steps from being performed as “Certain Methods of Organizing Human Activities” which pertains to (1) commercial interactions (including sales activities or behaviors; business relations) or (2) managing personal behavior or relationships (including teachings or following rules or instructions) and additionally or alternatively as “Mental Processes” which pertains to (3) concepts performed in the human mind (including observations or evaluations or judgments) or (4) using pen and paper as a physical aid and additionally or alternatively as “Mathematical Concepts” which pertains to (5) mathematical calculations or (6) mathematical relationships.
Moreover, the mere recitation of generic computer components such as (e.g., “one or more hardware processors” & “a memory”) does not take the claims out of “Certain Methods of Organizing Human Activities” or “Mental Processes” or “Mathematical Concepts” Groupings.
Therefore, at step 2a prong 1, Yes, Claims 1-20 recites an abstract idea. We proceed onto analyzing the claims at step 2a prong 2.
Step 2A Prong Two: With respect to Step 2A Prong Two of the eligibility inquiry (as explained in MPEP § 2106.04(d)), the judicial exception is not integrated into a practical application. Independent Claim 1 recites additional elements directed to: (e.g., “one or more hardware processors” & “one or more user interfaces” & “one or more electronic devices”). Independent Claim 14 recites additional elements directed to: (e.g., “one or more hardware processors” & “a memory” & “one or more user interfaces” & “one or more electronic devices”). Independent Claim 19 recites additional elements directed to: (e.g., “one or more hardware processors” & “one or more user interfaces” & “one or more electronic devices”). These additional elements have been considered individually and in combination, but fail to integrate the abstract idea into a practical application because they amount to using computing elements or instructions (software) to perform the abstract idea, similar to adding the words “apply it” (or an equivalent), which merely serves to link the use of the judicial exception to a particular technological environment. See MPEP § 2106.05(f) and MPEP § 2106.05(h).
Independent Claims 1, 14 and 19: With respect to reliance on additional elements such as (e.g., “a stacked AI model” & “a plurality of input artificial intelligence (AI) models” & “wherein the plurality of subsystems” & “a data obtaining subsystem” & “a data pre-processing subsystem” & “data obtaining subsystem” & “a feature determining subsystem” & “a feature combination generating subsystem” & “a forecasting generating subsystem” & “an output subsystem”) shown in Independent Claims 1, 14 and 19 when considered individually and as an ordered combination (as a whole) in view of these claim limitations, these additional elements do not provide limitations that are indicative of integration into a practical application under step 2a prong 2 due to: Merely invoking artificial intelligence or machine learning models (like a "stacked AI model" or "input AI models") to perform mathematical forecasting does not improve the functioning of the computer itself. The steps of obtaining data, pre-processing, and outputting forecasts on a generic interface amount to "apply it on a computer" without providing a specific, unconventional technological solution to a technological problem. Thus, the abstract idea is not integrated into a practical application. Considered individually and as an ordered combination, the limitations do no more than recite computer and data-processing activities (fetching data from sources, running standard statistical/AI regressions, and displaying the results). The selection of "cash flow" parameters and "feature combinations" reflects domain-specific financial management rules, not an inventive technical contribution to AI architecture. The hardware components merely act as a tool to run the math, failing to transform the abstract financial forecast into a patent-eligible application.
Furthermore, certain/particular limitations in Independent Claims 1, 14 and 19 recite (1) “mere data gathering” (e.g., “obtaining, by one or more hardware processors, data associated with one or more business units from one or more data sources” (see Independent Claim 1); “a data obtaining subsystem configured to obtain data associated with one or more business units from one or more data sources” (see Independent Claim 14); “obtaining data associated with one or more business units from one or more data sources” (see Independent Claim 19)) & (2) “mere data transmitting/outputting” (e.g., “providing, by the one or more hardware processors, the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users” (see Independent Claim 1); “an output subsystem configured to provide the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users” (see Independent Claim 14);“providing the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users” (see Independent Claim 19)), which when evaluated as additional elements, these activities at most amount to insignificant extra-solution activities (see MPEP § 2106.05 (g)).
In addition, these limitations fail to provide an improvement to the functioning of a computer or to any other technology or technical field, fail to apply the exception with a particular machine, fail to apply the judicial exception to effect a particular treatment or prophylaxis for a disease or medical condition, fail to effect a transformation of a particular article to a different state or thing, and fail to apply/use the abstract idea in a meaningful way beyond generally linking the use of the judicial exception to a particular technological environment.
Accordingly, because the Step 2A Prong One and Prong Two analysis resulted in the conclusion that the claims are directed to an abstract idea, additional analysis under Step 2B of the eligibility inquiry must be conducted in order to determine whether any claim element or combination of elements amount to significantly more than the judicial exception. Therefore, at step 2a prong 2, Claims 1-20 are directed to the abstract idea and do not recite additional elements that integrate into a practical application.
Step 2B: (As explained in MPEP § 2106.05), it has been determined that the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. Independent Claim 1 recites additional elements directed to: (e.g., “one or more hardware processors” & “one or more user interfaces” & “one or more electronic devices”). Independent Claim 14 recites additional elements directed to: (e.g., “one or more hardware processors” & “a memory” & “one or more user interfaces” & “one or more electronic devices”). Independent Claim 19 recites additional elements directed to: (e.g., “one or more hardware processors” & “one or more user interfaces” & “one or more electronic devices”). These elements have been considered individually and in combination, but fail to add significantly more to the claims because they amount to using computing elements or instructions (software) to perform the abstract idea, similar to adding the words “apply it” (or an equivalent), which merely serves to link the use of the judicial exception to a particular technological environment (computing environment) and does not amount to significantly more than the abstract idea itself. See MPEP § 2106.05 (h) and See MPEP § 2106.05 (f). Notably, Applicant’s Specification suggests that the claimed invention relies on nothing more than a general-purpose computer executing the instructions to implement the invention (see at least Applicant’s Specification ¶ [[0061]: “A "module" or "subsystem" may also comprise programmable logic or circuitry (as encompassed within a general-purpose processor or other programmable processor) that is temporarily configured by software to perform certain operations.” See also Applicant’s Specification ¶ [0074]: “The one or more hardware processors 204 may also include embedded controllers, including at least one of: generic or programmable logic devices or arrays, application specific integrated circuits, single-chip computers, and the like).”).
Furthermore, certain/particular limitations in Independent Claims 1, 14 and 19 recite (1) “mere data gathering” (e.g., “obtaining, by one or more hardware processors, data associated with one or more business units from one or more data sources” (see Independent Claim 1); “a data obtaining subsystem configured to obtain data associated with one or more business units from one or more data sources” (see Independent Claim 14); “obtaining data associated with one or more business units from one or more data sources” (see Independent Claim 19)) & (2) “mere data transmitting/outputting” (e.g., “providing, by the one or more hardware processors, the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users” (see Independent Claim 1); “an output subsystem configured to provide the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users” (see Independent Claim 14);“providing the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users” (see Independent Claim 19)), which when evaluated as additional elements, these activities at most amount to insignificant extra-solution activities (see MPEP § 2106.05 (g)), and have been recognized as Well-Understood, Routine and Conventional (WURC), and thus insufficient to add significantly more to the abstract idea. See MPEP § 2106.05(d) ii - Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607, 610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016) (using a telephone for image transmission); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network); buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network). See also MPEP § 2106.05(d) ii - Storing and Retrieving information in memory, Versata Dev. Group, Inc. v. SAP Am., Inc.,793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); OIP Techs., 788 F.3d at 1363, 115USPQ2d at 1092-93.
The additional element of “AI models” or “artificial intelligence” in general in the claims does not amount to significantly more than the judicial exceptions under step 2B due to being expressly recognized as Well-Understood, Routine and Conventional (WURC) in the art. For example, see also US PG Pub (US 2024/0037370 A1) – “Automated Data Forecasting Using Machine Learning” hereinafter Chan, et. al. See also Chan at ¶ [0025]: “Generalizability (e.g., applicability of prediction model to accounts and/or time series data on which the prediction model has not been trained). See also Chan at ¶ [0047]: “The historical data may also include an out-of-sample data set that may be utilized to evaluate generalizability of the machine learning model 216. For example, the out-of-sample data set may include a portion of the historical data that was not included in the training data set, the validation data set or the test data set. This type of evaluation may indicate the generalizability of the machine learning model 216 (e.g., ability of model to generate forecasts for accounts without having been trained, validated or tested with historical data from such accounts).” See also Chan at ¶ [0058]: “The model training data generator may further be configured to introduce randomness into the order of the input sequences, so as the machine learning model 216 may learn in a generalizable and robust manner.”
In addition, when taken as an ordered combination, the ordered combination adds nothing that is not already present as when the elements are taken individually. There is no indication that the combination of elements integrates the abstract idea into a practical application. Therefore, when viewed as a whole, these additional claim elements do not provide meaningful limitations to transform the abstract idea into a practical application of the abstract idea or that, as an ordered combination, amount to significantly more than the abstract idea itself.
Dependent Claims 2-13, 15-18 and 20 recite additional elements directed to: (e.g., “an intelligence AI-inference process” (see Dependent Claim 2) & “the plurality of input AI models” (see Dependent Claims 3-4) & “first input AI model” (see Dependent Claims 5, 15 and 20) & “second input AI model” (see Dependent Claims 6, 15 and 20) & “third input AI model” (see Dependent Claims 7, 15 and 20) & “fourth input AI model” (see Dependent Claims 8, 15 and 20) & “fifth input AI model” (see Dependent Claims 9, 15 and 20), etc…), and when considered individually and as an ordered combination (as a whole) with the limitations recite the same abstract idea(s) as shown in Independent Claims 1, 14 and 19 along with further steps/details that could be performed as “Certain Methods of Organizing Human Activities” which pertains to (1) commercial interactions (including sales activities or behaviors; business relations) or (2) managing personal behavior or relationships (including teachings or following rules or instructions) and additionally or alternatively as “Mental Processes” which pertains to (3) concepts performed in the human mind (including observations or evaluations or judgments) or (4) using pen and paper as a physical aid and additionally or alternatively as “Mathematical Concepts” which pertains to (5) mathematical calculations or (6) mathematical relationships.
The following Dependent Claims 2-13, 15-18 and 20 add further different AI input models and subsystem elements to the abstract framework of Independent Claims 1, 14 and 19. They are analyzed collectively below because they all share a common legal failure under § 101.
Claim 2: Claim 2 recites pre-processing data "to transform the data into an organized format being adaptable for performing an intelligent AI-inference process". Under Prong 2, merely stating that data is prepared "for an AI process" or "for an inference model” amount to generic data manipulation or extra-solution activity. The claim does not improve the internal functioning of the computer or the artificial intelligence algorithm itself (e.g., it does not claim a novel neural network architecture, a unique matrix multiplication technique, or a specific memory-management breakthrough). Instead, it uses a generic hardware processor to apply routine data formatting to a specific field of use (financial datasets), which fails to integrate the abstract idea into a patent-eligible practical application.
Looking at the additional elements individually and as an ordered combination. The operations recite generic components ("one or more hardware processors") performing computer tasks (storing, parsing, formatting strings, converting numeric values). Standardizing column names, parsing dates, and checking for null values are activities in the data-processing art. Ordered together, they do not transform the abstract data-cleaning concept into something significantly more than the abstract idea itself. Therefore, the limitation lacks an inventive concept and remains ineligible.
Claim 3: The data pre-processing and AI-inference claim limitations are patent ineligible because they recite mental processes and methods of organizing human financial activity (abstract ideas) without providing an inventive concept or technological improvement. Claim 3 does not integrate the abstract operations into a practical application that saves or improves computer functionality. The steps merely state results—using "one or more hardware processors" and "ingestion by a plurality of input AI models"—to organize financial records. The claim does not improve how a computer, processor, or machine learning model functions internally. It uses generic AI as a tool to automate accounting and data-wrangling routines in a new data environment, which fails Prong 2.
The additional elements do not amount to "significantly more" than the abstract idea itself. When considered individually and as an ordered combination, the hardware components ("one or more hardware processors") are recited at a high level as pure tools. Performing data grouping, synthetic invoice generation, and updating posting dates using machine learning input routines represents activity in the data-processing arts.
Claim 4: The data pre-processing and noise-removal limitation in claim 4 is ineligible under 35 U.S.C. § 101 because it recites a mental process and mathematical/data-filtering concept grouped as an abstract idea, fails to integrate it into a practical application, and lacks an inventive concept. The abstract idea is not integrated into a practical application. The claim merely uses generic computer components ("one or more hardware processors") to execute routine data filtration. It does not improve how a computer or processor functions internally. The data manipulation serves only as a preparatory step for an AI model rather than defining a specific technological solution to a technical problem in computer science. It stands as an instruction to filter business records before analysis.
The claim lacks an inventive concept ("significantly more"). Looking at the additional elements individually and as an ordered combination, the operations recite data-processing steps. Filtering invoice dates, credit notes, and purchase orders represent activities in the financial data analysis field.
Claim 5: These claim limitations are patent ineligible under 35 U.S.C. 101. It falls under the abstract idea of data processing / financial forecasting, and it lacks an inventive concept to transform it into a patent-eligible application. The claim recites hardware components ("one or more hardware processors") and a "first input AI model." Using processors to run a statistical or machine learning model does not impose meaningful limits on the abstract idea. The AI model here serves merely as a tool to automate mathematical and predictive calculations. It does not improve the functioning of the computer itself, nor does it alter any computer technology. The abstract idea is not integrated into a practical application. It is merely instructions to apply an abstract concept on generic computer hardware using generic machine learning terms.
The claim requires "determining" payment dates and "computing" expected amounts based on historical data. Collecting historical payment patterns and applying a mathematical model to forecast a future date and amount is practice in data analysis. The use of "an AI model" is stated at a high level without reciting any specific, unconventional algorithmic improvement or specialized training architecture that solves a technical problem. Granting a patent on this limitation would preempt a massive field of financial cash-flow forecasting using machine learning. The claim does not provide an inventive concept (an element or combination of elements sufficient to ensure that the patent in practice amounts to significantly more than a patent upon the natural law or abstract idea itself).
Claim 6: The selected claim limitation recites an abstract idea (a method of organizing human activity and mental/mathematical processes) regarding financial forecasting, fails to integrate it into a technical application at Step 2A Prong Two, and lacks an inventive concept at Step 2B. Under step 2a prong 2, the claim must be evaluated to see if the abstract financial prediction is integrated into a practical application that improves computer functionality or a technical field. The limitation uses machine learning models and processors merely as a tool or a transparent proxy to automate commercial calculations. The claim does not improve how the computer, processor, or machine learning model functions technically (e.g., it does not claim an unconventional neural network architecture, optimized matrix memory handling, or accelerated training pipeline). Instead, it applies predictive modeling to a specific, non-technical business field (financial accounting and invoice management), which leaves the claim "directed to" the abstract idea. Because the claim is directed to a judicial exception and fails Step 2A Prong Two, we look at the additional elements individually and as an ordered combination to see if they amount to an inventive concept. The recitation of "one or more hardware processors" is computer implementation language. Using an "input AI model" to process features and output business predictions in the art. The combination of steps lacks any unconventional or non-routine technical ordering. It merely instructs the user to "apply it" using tools, which is legally insufficient to confer patent eligibility under 35 U.S.C. § 101.
Claim 7: The claim does not improve the functioning of a computer, nor does it improve the machine learning models or algorithms themselves (such as a novel neural network architecture or training methodology). Applying standard AI models to a specific data environment—namely, accounts receivable and bank statements for financial forecasting—amounts to an ineligible field-of-use restriction. The additional elements (using hardware processors and input AI models) are recited at a high, functional level of generality, serving merely as an instruction to "apply it" using standard automation rather than integrating the exception into a practical technological application. Looking at the claim elements individually and as an ordered combination, the system does not provide "significantly more" than the abstract financial calculation itself. The hardware processors, bank data sources, and AR invoices are conventional, well-understood components in the financial sector. Combining routine data inputs (bank statements and invoices) with generic predictive math/AI models does not transform the abstract financial deduction calculation into patent-eligible subject matter.
Claim 8: The claim does not improve the operation of the computer or the machine learning model itself; it uses a generic fourth input AI model as a tool to perform routine business accounting. The operations described (calculating averages, distributions, and future payment windows) are merely instructions to apply conventional data analysis to financial information, meaning the exception is not integrated into a practical technical application. Executing these steps via "one or more hardware processors" and a "fourth input AI model" amounts to generic computer implementation, which the Supreme Court ruled is insufficient to transform an abstract idea into a patent-eligible invention. The combination of tracking clearance times and forecasting future invoice payments does not provide an unconventional or significantly transformative twist; it is routine automation of conventional commercial finance work.
Claim 9: The recited claim limitation directed to determining an expected credit application against open invoices using a fifth input AI model based on historical patterns and claim rates is patent ineligible because it falls under a mental process/abstract idea and fails to provide an inventive concept. The limitation uses generic hardware ("one or more hardware processors") and a generic functional tool ("a fifth input AI model") to execute the prediction. Under recent Federal Circuit precedent (e.g., [Recentive Analytics v. Fox Corp.]), applying a standard, established machine learning or AI technique to a new data environment or a specific field of use (finance/invoicing) without improving the underlying AI architecture or computer functionality itself does not integrate the abstract idea into a patent-eligible practical application. It merely uses generic tools to automate a manual business task. The additional elements—such as a "fifth input AI model," "hardware processors," and gathering historical data features—represent generic, well-understood, routine, and conventional computer activities. There is no recitation of a specific, unconventional technological improvement in how the AI model is trained, structured, or compiled (such as an improvement to computer memory efficiency or specific algorithmic enhancements). As an ordered combination, they simply instruct the system to apply generic predictive modeling to financial records, leaving the claim directed solely to the ineligible abstract concept itself.
Claim 10: The recited limitation—determining cash flow over a predetermined horizon using historical trends, seasonality, and an AI model—is ineligible under 35 U.S.C. § 101 because it falls under the abstract idea exception (mathematical concepts and certain methods of organizing human activity) and fails to integrate that exception into a practical application. The claim does not improve the functioning of a computer or the underlying machine learning architecture itself. It does not claim a new training method, a novel neural network topology, or an improved data storage/retrieval technique. Instead, the AI model is invoked merely as a generic tool ("a sixth input AI model") to process financial data. The focus of the limitation is entirely on the result (cash flow prediction) in the field of finance, which amounts to a generic field-of-use restriction. It does not integrate the mathematical calculation into a specific technological solution. "One or more hardware processors": This invokes generic computer hardware to perform generic calculations. Under well-established case law, reciting generic hardware ("using a hardware processor") to run an abstract mathematical process does not transform the claim into significantly more. "Time series data / historical trends / seasonality": This merely represents data gathering and generalized data input, which are routine activities ancillary to the underlying calculation. Ordered Combination: Combining multiple AI models or using a "sixth input AI model" to look at seasonal financial data does not establish an inventive concept. The combination relies purely on the conventional execution of machine learning for data analysis without providing an inventive technical contribution to the computing arts.
Claim 11: The incremental training limitation for a stacked AI forecasting model is patent-ineligible because it recites a generalized mathematical concept and a mental process without improving computer functionality or offering a specific, non-conventional technical fix. The training is executed generically "by the one or more hardware processors" using standard "incremental learning techniques." It does not restrict the hardware architecture or internal computer memory management. The claim fails to improve how a computer or processor functions. It uses standard machine learning as a tool to execute a routine mathematical updating step rather than solving a technical problem native to computers. The abstract idea is not integrated into a practical application because the additional elements (hardware processors) amount to generic computer implementation.
Incremental learning, feature combination evaluation, and multi-horizon forecasting are activities in the field of machine learning. Combining a stacked AI model with sequential horizon training, viewed as an ordered combination, relies entirely on the execution of machine learning principles. The limitation does not provide an inventive concept "significantly more" than the abstract mathematical algorithm itself.
Claim 12: The selected claim limitations directed to assessing accuracy, ranking forecasts via mathematical parameters (R2 score), weighted accuracy, error percentiles), and selecting a model is ineligible. The claim is not integrated into a practical application because it stops at evaluating, ranking, and selecting a model based on abstract statistical metrics. Improving the mathematical or computational accuracy of a prediction via algorithmic ranking does not qualify as an improvement to computer capabilities or a technological process; rather, it is an enhancement of the abstract mathematical calculation itself. The claim uses generic processors merely as a tool to run math equations. The steps call upon standard, generic hardware components ("one or more hardware processors") and conventional statistical concepts (R2), error percentiles) used in ordinary machine learning workflows. Performing routine data sorting, statistical evaluation, and model selection does not transform the abstract mathematical/mental process into "significantly more." It amounts to applying generic machine learning evaluation routines as a generic black box, which fails to provide an inventive concept under Alice Step 2B.
Claim 13: The specified claim limitations—directed to monitoring performance, determining threshold values, fine-tuning machine learning parameters/features, and evaluating weighted average forecast accuracy—is patent-ineligible under 35 U.S.C. 101. The steps are executed by "one or more hardware processors" in a generic manner. Simply appending generic computer hardware ("hardware processors") to perform math, data collection, and routine model evaluation does not transform the abstract idea. The focus of the claim is statistical optimization (accuracy across horizons) rather than improving how a computer network, memory hierarchy, or processor operates internally. The limitations represent standard business or analytical utility rather than a technological solution to a technological problem. Viewed individually and as an ordered combination, the elements of monitoring performance over time, checking against a threshold, and fine-tuning parameters represent conventional, routine, and well-understood activities in the field of machine learning. The combination of evaluating forecast accuracy and tweaking parameters does not amount to an inventive concept that significantly departs from routine data manipulation. It applies established machine learning training principles to a generalized predictive workflow, which federal precedent treats as ineligible.
For Dependent Claims 15-18 and 20, see similar 35 U.S.C. § 101 reasons and rationale from Dependent Claims 2-13 applied here as well and are therefore still ineligible under Steps 2A Prong 2 and 2B.
The ordered combination of elements in the Claims (including the limitations inherited from the parent claim(s)) add nothing that is not already present as when the elements are taken individually. There is no indication that the combination of elements improves the functioning of a computer or improves any other technology. Accordingly, the subject matter encompassed by the claims fails to amount to a practical application or significantly more than the abstract idea itself. Therefore, under Step 2B, Claims 1-20 do not include additional elements that are sufficient to amount to significantly more than the recited judicial exceptions. Thus, Claims 1-20 are ineligible with respect to the 35 U.S.C. § 101 analysis.
Claim Rejections - 35 USC § 103
5. This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
6. In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
7. The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
8. Claims 1-3, 10-11, 14, 16 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over US PG Pub (US 2025/0103605 A1) hereinafter Gianelle, et. al., in view of US PG Pub (US 2024/0037370 A1) hereinafter Chan, et. al., in further view of NPL Document: Predictive analytics and AI-driven strategies for enhanced cash flow forecasting. In Intelligent Systems Conference (pp. 296-315). Cham: Springer Nature Switzerland, hereinafter Mehra, et. al., and in further view of US PG Pub (US 2021/0303970 A1) hereinafter Ma, et. al.
Regarding Independent Claim 1, Gianelle machine-learning based (ML-based) method for forecasting cash flow teaches the following:
- obtaining, by one or more hardware processors (see at least Gianelle: ¶ [0117-0119].), data associated with one or more business units (see at least Gianella: ¶ [0195] & ¶ [0199-0200]. Gianelle teaches that each user profile may have specific investment requirements. These requirements may be based on the investment itself (e.g., corresponding to a specific company, type of company, etc.) or a state of the portfolio (e.g., a required rate of return, level of risk, etc.). See also Gianelle at ¶ [0199-0200]: The system may analyze multiple sources to discern whether a user has any preferences regarding the types of companies associated with the bonds or equities that make up a given structured note. The system may employ web crawlers to determine whether the user has publicly expressed specific interests either for or against specific companies, technologies, ecological interests, social interests, governmental entities, high-profile persons, etc. ) from one or more data sources (see at least Gianelle: ¶ [0057-0058] & ¶ [0211] & ¶ [0218] & ¶ [0224]. Gianelle teaches that the system may determine potential data streams for the time-series data stream clusters based on the user's interests, respective cohorts, and/or other information about the user using data sourced from multiple sources such as account transaction data, portfolio management experiential data, external public data sources, market data, customer social media interests, user-specified directions, etc. to match a user with a structured note (e.g., represented by a plurality of time-series data streams and/or clusters thereof).) See also Gianelle at [0057-0058]: This procedure often requires data cleaning, normalization, and transformation to ensure consistency and accuracy across different data sources before aggregation. See also Gianelle at [0218]: For example, the system may determine that different data types reflect different aspects of data (e.g., the same time period, sampling frequency, data sources, data formats, etc.). See also Gianelle at [0224]: The system may receive a first dataset by first interfacing with a data source or repository that provides access to non-synthetic, real-world data.));
- pre-processing, by the one or more hardware processors (see at least Gianelle: ¶ [0117-0119].), the data to generate pre-processed data (see at least Gianelle: ¶ [0058]. Gianelle teaches that after ingestion, the data often requires preprocessing, including cleaning, deduplication, and normalization, to ensure consistency across the different data streams. This step may involve handling missing data, resolving conflicts in data types, and applying transformations like date formatting or unit conversion. The cleaned data is then typically stored in a staging area, such as a data warehouse or a data lake.).
Gianelle machine-learning based (ML-based) method for forecasting cash flow does not explicitly disclose, but Chan in the analogous art for machine-learning based (ML-based) method for forecasting cash flow teaches the following limitations:
- determining, by the one or more hardware processors (see at least Chan: Fig. 2 & ¶ [0036]. Chan teaches one or more computer processors 204 in Fig. 2.), one or more features (see at least Chan: ¶ [0042] & ¶ [0101].) associated with one or more financial information for one or more horizons (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) based on the pre-processed data (see at least Chan: ¶ [0085-0089] & ¶ [0101-0102] & ¶ [0109]. Chan notes generating the training data set may include pre-processing module 208 deriving additional data from the user-level information. For example, pre-processing module 208 derive a transaction type for various transactions included in the user-level data. Exemplary transaction type inputs may include, but are not limited to, rent payments, loan payments, one-time loan funds, and the like. These transactions may be further classified as usual and/or unusual payments. See also Chan at ¶ [0101-0102]: The system may then be configured to pre-process the historical account activity data. Pre-processing may include a combination of data imputation, data filtering, feature engineering and/or data scaling. Data imputation may comprise identifying data among the historical account activity data having missing values, and replacing the missing values with system-generated imputed values. See also Chan at ¶ [0109]: The system may be configured to perform further pre-processing operations on the current inflow activity data, the current outflow activity data, and the current balance information associated with the target account(s). The further pre-processing operations may include, for example, data imputation, feature engineering and data scaling.) using a plurality of input artificial intelligence (AI) models (see at least Chan: Fig. 2 noting “ML models 216” & ¶ [0019] & ¶ [0056]. Chan notes forecasting cash flow for one or more customers using one or more artificial intelligence processes. The forecasting may be implemented on an account level (e.g., for selected account(s) from among any number of accounts associated with a particular customer), on the customer level (e.g., across all accounts associated with the particular customer), and/or across a user base clusters (e.g., across multiple customers' accounts). See also Chan at ¶ [0056]: Machine learning model 216 may be representative of a long short-term memory (LSTM). A LSTM model is a specific artificial recurrent neural network architecture capable of learning long term dependencies. LSTM model may learn to capture and/or determine multiple periodic patterns in a user's or client's data, such as weekly, semi-weekly, monthly, and the like. Using an LSTM model may allow machine learning platform 116 to model inflow, outflow, balance and/or any number of other variables together in a combined model, for any number of time steps, and for any number of users, all at once.);
- generating, by the one or more hardware processors (see at least Chan: Fig. 2 & ¶ [0036]. Chan teaches one or more computer processors 204 in Fig. 2.), one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: determining, by the one or more hardware processors, one or more features associated with one or more financial information for one or more horizons based on the pre-processed data using a plurality of input AI models & generating, by the one or more hardware processors, one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons, and in view of Chan, whereby the method of Chan forecasts cash flow for one or more customers using one or more artificial intelligence processes. The forecasting may be implemented on an account level (e.g., for selected account(s) from among any number of accounts associated with a particular customer), on the customer level (e.g., across all accounts associated with the particular customer), and/or across a user base clusters (e.g., across multiple customers' accounts). The system may include a machine learning platform trained to forecast future customer activity based on historical customer activity, including for accounts for which the system has not been trained. Such functionality aims to assist clients to forecast their cash balances by considering net inflow and outflow activity (see at least Chan: ¶ [0019]). Moreover, the weight parameter may provide for a tradeoff between inflow/outflow prediction errors and the balance prediction errors. In other words, by adjusting the weight parameter, the objective function may be optimized for improving the balance forecast accuracy at the expense of a slightly less accurate inflow/outflow forecast (see at least Chan: ¶ [0050]).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Chan, the results of the combination were predictable.
Gianelle / Chan machine-learning based (ML-based) method for forecasting cash flow does not explicitly disclose, but Mehra, et. al. in the analogous art for machine-learning based (ML-based) method for forecasting cash flow teaches the following limitations:
- generating, by the one or more hardware processors, one or more forecasts for each horizon for a pre-determined time interval, using a stacked AI model comprising one or more forecasting models (see at least Mehra: (Under Proposed Methodology pages 299-300) & (Pages 309-310). Mehra teaches that Fig. 1, is to combine the predictive powers of a Random Forest model with a Custom Neural Net model to create an advanced forecasting system. Our goal is to improve the precision and dependability of upcoming three-month cash flow forecasts by utilizing the advantages of both models. By calculating the average of these models’ outputs, it will be possible to integrate them and provide a thorough and reliable method of financial forecasting. This study uses a unique strategy that combines the strengths of a Random Forest (RF) model with a Custom Neural Net model to improve the accuracy of future three-month forecasting. See also Mehra at Pages 302-303: “Using daily user datasets, the models are retrained independently by adjusting hyperparameters, adding fresh data, and removing the oldest data from the training set. The goal of reinforcement learning is to predict the inflow and outflow over the upcoming three months. A weighted average, with the weights derived from each model’s past performance, is used to combine the forecasts of the two models. This approach of dynamic reinforcement learning makes sure that the forecasting models are always able to adjust to the evolving patterns of user behavior.” See also Mehra at Page 309: The model unfolds in a systematic three-step process, each geared towards optimizing financial forecasting through the use of pretrained custom models. In the first step, the pretrained LSTM model is utilized to forecast the inflow and outflow for the next three months. Following the initial LSTM forecasting, the methodology seamlessly transitions into the second step, employing a pretrained Random Forest model. The Random Forest model is applied to forecast inflow and outflow for the same three-month horizon. The next critical phase involves parameter selection for both the LSTM and Random Forest models. In the third and final step, the predictions from the LSTM and Random Forest models are combined to form a unified forecast. Taking a weighted average of the individual model predictions allows for a balanced and comprehensive outlook on future inflow and outflow trends. The weighted averaging approach ensures that the unique strengths of each model contribute meaningfully to the final forecast. The outcomes of this combined prediction are then plotted, providing a visual representation of the forecasted financial trends over the next three months as per Fig. 5.), wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon (see at least Mehra: Fig. 1 & Fig. 5 & (Pages 302-303) & (Page 309-310). Mehra teaches that LSTM and random forest models that have already been trained are used in this stage. Using daily user datasets, the models are retrained independently by adjusting hyperparameters, adding fresh data, and removing the oldest data from the training set. The goal of reinforcement learning is to predict the inflow and outflow over the upcoming three months. A weighted average, with the weights derived from each model’s past performance, is used to combine the forecasts of the two models. This approach of dynamic reinforcement learning makes sure that the forecasting models are always able to adjust to the evolving patterns of user behavior. The models are kept current and adaptable to changing trends by incorporating daily user input into the training process. See also Page 309 of Mehra. See also Page 310 of Mehra: “The combination harnesses the power of LSTM’s sequential memory and Random Forest’s ensemble learning, resulting in a more nuanced understanding of the underlying patterns in cashflow data.”).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: generating, by the one or more hardware processors, one or more forecasts for each horizon for a pre-determined time interval, using a stacked AI model comprising one or more forecasting models, wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon, and in further view of Mehra, whereby the hybrid model, which incorporates Random Forest, can handle complex interactions and enhance predictive accuracy by aggregating predictions from multiple decision trees. This approach aligns with the principle of model diversification to improve forecasting accuracy. The synergistic fusion of LSTM and Random Forest offers a comprehensive solution that leverages both models’ strengths, resulting in more accurate and reliable cash flow forecasting tools. This innovative approach aims to provide businesses with a more nuanced and resilient forecasting capability, enhancing their financial decision-making processes. The synergistic fusion of LSTM and both models, resulting in a more accurate and reliable cash flow forecasting tool. The combined model not only improves prediction accuracy but also mitigates the risk of overfitting, providing a sophisticated and well-rounded solution for financial analysts and decision-makers. This integration represents a significant leap forward in the quest for precise and dependable cash flow predictions in complex financial landscapes (see at least Mehra: Page 298).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Mehra, the results of the combination were predictable.
Gianelle / Chan / Mehra machine-learning based (ML-based) method for forecasting cash flow does not explicitly disclose, but Ma in the analogous art for machine-learning based (ML-based) method for forecasting cash flow teaches the following limitations:
- providing, by the one or more hardware processors (see at least Ma: ¶ [0112]. Ma teaches that a hardware-implemented module is a tangible unit capable of performing certain operations and may be configured or arranged in a certain manner. One or more computer systems (e.g., a standalone, client, or server computer system) or one or more hardware processors may be configured by software (e.g., an application or application portion) as a hardware-implemented module that operates to perform certain operations as described herein.), the generated one or more forecasts for the cash flow of the one or more business units (see at least Ma: ¶ [0015] & ¶ [0071] & Figs. 3-4. Ma teaches that liquidity forecasting is performed directly on corporate cash flow data, the results have high noise and low accuracy. By considering additional data instead of only the aggregate corporate cash flow, greater accuracy results. Separate models are used for each division or subsidiary of the corporation, and the final model 740 uses the results from those models to generate the final result 750 for the cash flow of the corporation. A second hierarchy, within each division or subsidiary, may be by data source (e.g., Associated Press, Bureau of Labor Statistics, World Bank, International Monetary Fund, Yahoo! ® Finance, Google® Finance, or any suitable combination thereof), by currency (e.g., dollar, euro, yen, yuan, or any suitable combination thereof), by political region (e.g., the European Union, the United States, China, or any suitable combination thereof). See also Ma at ¶ [0015]: “Predict liquidity using neural networks would be to use the historical cash position of a business over a period of time to train a neural network. The neural network is used to predict future liquidity based on the cash position to date.” See also Ma at Figs. 3-4.), as an output (see at least Ma: Fig. 5 & ¶ [0033-0035] & ¶ [0039] & ¶ [0043]. Ma notes that machine-learning algorithms operate by building an ML model 516 from example training data 512 in order to make data-driven predictions or decisions expressed as outputs or assessments 520. See also Ma at ¶ [0039]: The machine-learning algorithms utilize the training data 512 to find correlations among identified features 502 that affect the outcome. A feature 502 is an individual measurable property of a phenomenon being observed. The concept of a feature is related to that of an explanatory variable used in statistical techniques such as linear regression. Further, deep features represent the output of nodes in hidden layers of the deep neural network. See also Ma at ¶ [0043]: The training data 512 is time-series data comprising a sequence of values and the output of the machine learning model 516 is a predicted next value of the sequence. For example, 256 previous values may be used as an input feature and the 257th value used as the labeled output. ), to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users (see at least Ma: ¶ [0021] & Fig. 1 & Figs. 9-10. Ma notes client devices 160A and 160B in Fig. 1 and causing the predicted liquidity to be presented on a user interface of a client device at step 930 of Fig. 9. See also Ma at ¶ [0021]: “The application server 120 causes the trained neural network 150 to process the business data 140 to generate a liquidity forecast. The liquidity forecast is provided by the application server 120 to a client device 160 via the network 190 for display to a user.”).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: providing, by the one or more hardware processors, the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users, and in further view of Ma, whereby multiple neural networks are each trained on time-series data from a different domain. Each of the trained neural networks is used to make a domain-specific prediction for each point in time. Thus, time-series prediction data is generated by each of the trained neural networks. The domain-specific time-series prediction data are combined into a vector and used to train a final model that predicts a value. By breaking down the problem of forecasting into domain-specific forecasting models and a forecasting model, accuracy is improved over traditional document-based forecasting and computational resources are saved over traditional neural network designs (see at least Ma: ¶ [abstract].). Furthermore, one way to improve the performance of DNNs is to identify newer structures for the feature-extraction layers, and another way is by improving the way the parameters are identified at the different layers for accomplishing a desired task. For a given neural network, there may be millions of parameters to be optimized. Trying to optimize all these parameters from scratch may take hours, days, or even weeks, depending on the amount of computing resources available and the amount of data in the training set (see at least Ma: ¶ [0064].).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Ma, the results of the combination were predictable.
Regarding Independent Claim 14, Gianelle machine-learning based (ML-based) system for forecasting cash flow teaches the following:
- one or more hardware processors (see at least Gianelle: ¶ [0117-0119].);
- a memory coupled to the one or more hardware processors, wherein the memory comprises a plurality of subsystems in form of programmable instructions executable by the one or more hardware processors, and wherein the plurality of subsystems (see at least Gianelle: Fig. 3B & ¶ [0117-0119].) comprises:
- data obtaining subsystem (see at least Gianelle: Fig. 3B & ¶ [0117-0119].) configured to obtain data associated with one or more business units (see at least Gianella: ¶ [0195] & ¶ [0199-0200]. Gianelle teaches that each user profile may have specific investment requirements. These requirements may be based on the investment itself (e.g., corresponding to a specific company, type of company, etc.) or a state of the portfolio (e.g., a required rate of return, level of risk, etc.). See also Gianelle at ¶ [0199-0200]: The system may analyze multiple sources to discern whether a user has any preferences regarding the types of companies associated with the bonds or equities that make up a given structured note. The system may employ web crawlers to determine whether the user has publicly expressed specific interests either for or against specific companies, technologies, ecological interests, social interests, governmental entities, high-profile persons, etc. ) from one or more data sources (see at least Gianelle: ¶ [0057-0058] & ¶ [0211] & ¶ [0218] & ¶ [0224]. Gianelle teaches that the system may determine potential data streams for the time-series data stream clusters based on the user's interests, respective cohorts, and/or other information about the user using data sourced from multiple sources such as account transaction data, portfolio management experiential data, external public data sources, market data, customer social media interests, user-specified directions, etc. to match a user with a structured note (e.g., represented by a plurality of time-series data streams and/or clusters thereof).) See also Gianelle at [0057-0058]: This procedure often requires data cleaning, normalization, and transformation to ensure consistency and accuracy across different data sources before aggregation. See also Gianelle at [0218]: For example, the system may determine that different data types reflect different aspects of data (e.g., the same time period, sampling frequency, data sources, data formats, etc.). See also Gianelle at [0224]: The system may receive a first dataset by first interfacing with a data source or repository that provides access to non-synthetic, real-world data.));
- a data pre-processing subsystem (see at least Gianelle: Fig. 3B & ¶ [0117-0119].) configured to pre-process the data to generate pre-processed data (see at least Gianelle: ¶ [0058]. Gianelle teaches that after ingestion, the data often requires preprocessing, including cleaning, deduplication, and normalization, to ensure consistency across the different data streams. This step may involve handling missing data, resolving conflicts in data types, and applying transformations like date formatting or unit conversion. The cleaned data is then typically stored in a staging area, such as a data warehouse or a data lake.).
Gianelle machine-learning based (ML-based) system for forecasting cash flow does not explicitly disclose, but Chan in the analogous art for machine-learning based (ML-based) system for forecasting cash flow teaches the following limitations:
- a feature determining subsystem configured (see at least Chan: Figs. 1-2 & Figs. 6A-6B.) to determine one or more features (see at least Chan: ¶ [0042] & ¶ [0101].) associated with one or more financial information for one or more horizons (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) based on the pre-processed data (see at least Chan: ¶ [0085-0089] & ¶ [0101-0102] & ¶ [0109]. Chan notes generating the training data set may include pre-processing module 208 deriving additional data from the user-level information. For example, pre-processing module 208 derive a transaction type for various transactions included in the user-level data. Exemplary transaction type inputs may include, but are not limited to, rent payments, loan payments, one-time loan funds, and the like. These transactions may be further classified as usual and/or unusual payments. See also Chan at ¶ [0101-0102]: The system may then be configured to pre-process the historical account activity data. Pre-processing may include a combination of data imputation, data filtering, feature engineering and/or data scaling. Data imputation may comprise identifying data among the historical account activity data having missing values, and replacing the missing values with system-generated imputed values. See also Chan at ¶ [0109]: The system may be configured to perform further pre-processing operations on the current inflow activity data, the current outflow activity data, and the current balance information associated with the target account(s). The further pre-processing operations may include, for example, data imputation, feature engineering and data scaling.) using a plurality of input artificial intelligence (AI) models (see at least Chan: Fig. 2 noting “ML models 216” & ¶ [0019] & ¶ [0056]. Chan notes forecasting cash flow for one or more customers using one or more artificial intelligence processes. The forecasting may be implemented on an account level (e.g., for selected account(s) from among any number of accounts associated with a particular customer), on the customer level (e.g., across all accounts associated with the particular customer), and/or across a user base clusters (e.g., across multiple customers' accounts). See also Chan at ¶ [0056]: Machine learning model 216 may be representative of a long short-term memory (LSTM). A LSTM model is a specific artificial recurrent neural network architecture capable of learning long term dependencies. LSTM model may learn to capture and/or determine multiple periodic patterns in a user's or client's data, such as weekly, semi-weekly, monthly, and the like. Using an LSTM model may allow machine learning platform 116 to model inflow, outflow, balance and/or any number of other variables together in a combined model, for any number of time steps, and for any number of users, all at once.);
- a feature combination generating subsystem configured to (see at least Chan: Figs. 1-2 & Figs. 6A-6B.) generate one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle machine-learning based (ML-based) system for forecasting cash flow with the aforementioned teachings of: a feature determining subsystem configured to determine one or more features associated with one or more financial information for one or more horizons based on the pre-processed data using a plurality of input AI models & a feature combination generating subsystem configured to generate one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons, and in view of Chan, whereby the method of Chan forecasts cash flow for one or more customers using one or more artificial intelligence processes. The forecasting may be implemented on an account level (e.g., for selected account(s) from among any number of accounts associated with a particular customer), on the customer level (e.g., across all accounts associated with the particular customer), and/or across a user base clusters (e.g., across multiple customers' accounts). The system may include a machine learning platform trained to forecast future customer activity based on historical customer activity, including for accounts for which the system has not been trained. Such functionality aims to assist clients to forecast their cash balances by considering net inflow and outflow activity (see at least Chan: ¶ [0019]). Moreover, the weight parameter may provide for a tradeoff between inflow/outflow prediction errors and the balance prediction errors. In other words, by adjusting the weight parameter, the objective function may be optimized for improving the balance forecast accuracy at the expense of a slightly less accurate inflow/outflow forecast (see at least Chan: ¶ [0050]).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) system for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Chan, the results of the combination were predictable.
Gianelle / Chan machine-learning based (ML-based) system for forecasting cash flow does not explicitly disclose, but Mehra, et. al. in the analogous art for machine-learning based (ML-based) system for forecasting cash flow teaches the following limitations:
- a forecast generating subsystem configured to generate one or more forecasts for each horizon for a pre-determined time interval, using a stacked AI model comprising one or more forecasting models (see at least Mehra: (Under Proposed Methodology pages 299-300) & (Pages 309-310). Mehra teaches that Fig. 1, is to combine the predictive powers of a Random Forest model with a Custom Neural Net model to create an advanced forecasting system. Our goal is to improve the precision and dependability of upcoming three-month cash flow forecasts by utilizing the advantages of both models. By calculating the average of these models’ outputs, it will be possible to integrate them and provide a thorough and reliable method of financial forecasting. This study uses a unique strategy that combines the strengths of a Random Forest (RF) model with a Custom Neural Net model to improve the accuracy of future three-month forecasting. See also Mehra at Pages 302-303: “Using daily user datasets, the models are retrained independently by adjusting hyperparameters, adding fresh data, and removing the oldest data from the training set. The goal of reinforcement learning is to predict the inflow and outflow over the upcoming three months. A weighted average, with the weights derived from each model’s past performance, is used to combine the forecasts of the two models. This approach of dynamic reinforcement learning makes sure that the forecasting models are always able to adjust to the evolving patterns of user behavior.” See also Mehra at Page 309: The model unfolds in a systematic three-step process, each geared towards optimizing financial forecasting through the use of pretrained custom models. In the first step, the pretrained LSTM model is utilized to forecast the inflow and outflow for the next three months. Following the initial LSTM forecasting, the methodology seamlessly transitions into the second step, employing a pretrained Random Forest model. The Random Forest model is applied to forecast inflow and outflow for the same three-month horizon. The next critical phase involves parameter selection for both the LSTM and Random Forest models. In the third and final step, the predictions from the LSTM and Random Forest models are combined to form a unified forecast. Taking a weighted average of the individual model predictions allows for a balanced and comprehensive outlook on future inflow and outflow trends. The weighted averaging approach ensures that the unique strengths of each model contribute meaningfully to the final forecast. The outcomes of this combined prediction are then plotted, providing a visual representation of the forecasted financial trends over the next three months as per Fig. 5.), wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon (see at least Mehra: Fig. 1 & Fig. 5 & (Pages 302-303) & (Page 309-310). Mehra teaches that LSTM and random forest models that have already been trained are used in this stage. Using daily user datasets, the models are retrained independently by adjusting hyperparameters, adding fresh data, and removing the oldest data from the training set. The goal of reinforcement learning is to predict the inflow and outflow over the upcoming three months. A weighted average, with the weights derived from each model’s past performance, is used to combine the forecasts of the two models. This approach of dynamic reinforcement learning makes sure that the forecasting models are always able to adjust to the evolving patterns of user behavior. The models are kept current and adaptable to changing trends by incorporating daily user input into the training process. See also Page 309 of Mehra. See also Page 310 of Mehra: “The combination harnesses the power of LSTM’s sequential memory and Random Forest’s ensemble learning, resulting in a more nuanced understanding of the underlying patterns in cashflow data.”).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan machine-learning based (ML-based) system for forecasting cash flow with the aforementioned teachings of: a forecast generating subsystem configured to generate one or more forecasts for each horizon for a pre-determined time interval, using a stacked AI model comprising one or more forecasting models, wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon, and in further view of Mehra, whereby the hybrid model, which incorporates Random Forest, can handle complex interactions and enhance predictive accuracy by aggregating predictions from multiple decision trees. This approach aligns with the principle of model diversification to improve forecasting accuracy. The synergistic fusion of LSTM and Random Forest offers a comprehensive solution that leverages both models’ strengths, resulting in more accurate and reliable cash flow forecasting tools. This innovative approach aims to provide businesses with a more nuanced and resilient forecasting capability, enhancing their financial decision-making processes. The synergistic fusion of LSTM and both models, resulting in a more accurate and reliable cash flow forecasting tool. The combined model not only improves prediction accuracy but also mitigates the risk of overfitting, providing a sophisticated and well-rounded solution for financial analysts and decision-makers. This integration represents a significant leap forward in the quest for precise and dependable cash flow predictions in complex financial landscapes (see at least Mehra: Page 298).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) system for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Mehra, the results of the combination were predictable.
Gianelle / Chan / Mehra machine-learning based (ML-based) system for forecasting cash flow does not explicitly disclose, but Ma in the analogous art for machine-learning based (ML-based) system for forecasting cash flow teaches the following limitations:
- an output subsystem (see at least Ma: Fig. 1 & Figs. 9-10) configured to provide the generated one or more forecasts for the cash flow of the one or more business units (see at least Ma: ¶ [0015] & ¶ [0071] & Figs. 3-4. Ma teaches that liquidity forecasting is performed directly on corporate cash flow data, the results have high noise and low accuracy. By considering additional data instead of only the aggregate corporate cash flow, greater accuracy results. Separate models are used for each division or subsidiary of the corporation, and the final model 740 uses the results from those models to generate the final result 750 for the cash flow of the corporation. A second hierarchy, within each division or subsidiary, may be by data source (e.g., Associated Press, Bureau of Labor Statistics, World Bank, International Monetary Fund, Yahoo! ® Finance, Google® Finance, or any suitable combination thereof), by currency (e.g., dollar, euro, yen, yuan, or any suitable combination thereof), by political region (e.g., the European Union, the United States, China, or any suitable combination thereof). See also Ma at ¶ [0015]: “Predict liquidity using neural networks would be to use the historical cash position of a business over a period of time to train a neural network. The neural network is used to predict future liquidity based on the cash position to date.” See also Ma at Figs. 3-4.), as an output (see at least Ma: Fig. 5 & ¶ [0033-0035] & ¶ [0039] & ¶ [0043]. Ma notes that machine-learning algorithms operate by building an ML model 516 from example training data 512 in order to make data-driven predictions or decisions expressed as outputs or assessments 520. See also Ma at ¶ [0039]: The machine-learning algorithms utilize the training data 512 to find correlations among identified features 502 that affect the outcome. A feature 502 is an individual measurable property of a phenomenon being observed. The concept of a feature is related to that of an explanatory variable used in statistical techniques such as linear regression. Further, deep features represent the output of nodes in hidden layers of the deep neural network. See also Ma at ¶ [0043]: The training data 512 is time-series data comprising a sequence of values and the output of the machine learning model 516 is a predicted next value of the sequence. For example, 256 previous values may be used as an input feature and the 257th value used as the labeled output. ), to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users (see at least Ma: ¶ [0021] & Fig. 1 & Figs. 9-10. Ma notes client devices 160A and 160B in Fig. 1 and causing the predicted liquidity to be presented on a user interface of a client device at step 930 of Fig. 9. See also Ma at ¶ [0021]: “The application server 120 causes the trained neural network 150 to process the business data 140 to generate a liquidity forecast. The liquidity forecast is provided by the application server 120 to a client device 160 via the network 190 for display to a user.”).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra machine-learning based (ML-based) system for forecasting cash flow with the aforementioned teachings of: an output subsystem configured to provide the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users, and in further view of Ma, whereby multiple neural networks are each trained on time-series data from a different domain. Each of the trained neural networks is used to make a domain-specific prediction for each point in time. Thus, time-series prediction data is generated by each of the trained neural networks. The domain-specific time-series prediction data are combined into a vector and used to train a final model that predicts a value. By breaking down the problem of forecasting into domain-specific forecasting models and a forecasting model, accuracy is improved over traditional document-based forecasting and computational resources are saved over traditional neural network designs (see at least Ma: ¶ [abstract].). Furthermore, one way to improve the performance of DNNs is to identify newer structures for the feature-extraction layers, and another way is by improving the way the parameters are identified at the different layers for accomplishing a desired task. For a given neural network, there may be millions of parameters to be optimized. Trying to optimize all these parameters from scratch may take hours, days, or even weeks, depending on the amount of computing resources available and the amount of data in the training set (see at least Ma: ¶ [0064].).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) system for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Ma, the results of the combination were predictable.
Regarding Independent Claim 19, Gianelle machine-learning based (ML-based) non-transitory computer readable storage medium for forecasting cash flow teaches the following:
- having instructions stored therein that when executed by one or more hardware processors, cause the one or more hardware processors to execute (see at least Gianelle: Fig. 3B & ¶ [0117-0119].) operations of:
- obtain data associated with one or more business units (see at least Gianella: ¶ [0195] & ¶ [0199-0200]. Gianelle teaches that each user profile may have specific investment requirements. These requirements may be based on the investment itself (e.g., corresponding to a specific company, type of company, etc.) or a state of the portfolio (e.g., a required rate of return, level of risk, etc.). See also Gianelle at ¶ [0199-0200]: The system may analyze multiple sources to discern whether a user has any preferences regarding the types of companies associated with the bonds or equities that make up a given structured note. The system may employ web crawlers to determine whether the user has publicly expressed specific interests either for or against specific companies, technologies, ecological interests, social interests, governmental entities, high-profile persons, etc. ) from one or more data sources (see at least Gianelle: ¶ [0057-0058] & ¶ [0211] & ¶ [0218] & ¶ [0224]. Gianelle teaches that the system may determine potential data streams for the time-series data stream clusters based on the user's interests, respective cohorts, and/or other information about the user using data sourced from multiple sources such as account transaction data, portfolio management experiential data, external public data sources, market data, customer social media interests, user-specified directions, etc. to match a user with a structured note (e.g., represented by a plurality of time-series data streams and/or clusters thereof).) See also Gianelle at [0057-0058]: This procedure often requires data cleaning, normalization, and transformation to ensure consistency and accuracy across different data sources before aggregation. See also Gianelle at [0218]: For example, the system may determine that different data types reflect different aspects of data (e.g., the same time period, sampling frequency, data sources, data formats, etc.). See also Gianelle at [0224]: The system may receive a first dataset by first interfacing with a data source or repository that provides access to non-synthetic, real-world data.));
- pre-processing the data to generate pre-processed data (see at least Gianelle: ¶ [0058]. Gianelle teaches that after ingestion, the data often requires preprocessing, including cleaning, deduplication, and normalization, to ensure consistency across the different data streams. This step may involve handling missing data, resolving conflicts in data types, and applying transformations like date formatting or unit conversion. The cleaned data is then typically stored in a staging area, such as a data warehouse or a data lake.)
Gianelle machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow does not explicitly disclose, but Chan in the analogous art for machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow teaches the following limitations:
- determining one or more features (see at least Chan: ¶ [0042] & ¶ [0101].) associated with one or more financial information for one or more horizons (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) based on the pre-processed data (see at least Chan: ¶ [0085-0089] & ¶ [0101-0102] & ¶ [0109]. Chan notes generating the training data set may include pre-processing module 208 deriving additional data from the user-level information. For example, pre-processing module 208 derive a transaction type for various transactions included in the user-level data. Exemplary transaction type inputs may include, but are not limited to, rent payments, loan payments, one-time loan funds, and the like. These transactions may be further classified as usual and/or unusual payments. See also Chan at ¶ [0101-0102]: The system may then be configured to pre-process the historical account activity data. Pre-processing may include a combination of data imputation, data filtering, feature engineering and/or data scaling. Data imputation may comprise identifying data among the historical account activity data having missing values, and replacing the missing values with system-generated imputed values. See also Chan at ¶ [0109]: The system may be configured to perform further pre-processing operations on the current inflow activity data, the current outflow activity data, and the current balance information associated with the target account(s). The further pre-processing operations may include, for example, data imputation, feature engineering and data scaling.) using a plurality of input artificial intelligence (AI) models (see at least Chan: Fig. 2 noting “ML models 216” & ¶ [0019] & ¶ [0056]. Chan notes forecasting cash flow for one or more customers using one or more artificial intelligence processes. The forecasting may be implemented on an account level (e.g., for selected account(s) from among any number of accounts associated with a particular customer), on the customer level (e.g., across all accounts associated with the particular customer), and/or across a user base clusters (e.g., across multiple customers' accounts). See also Chan at ¶ [0056]: Machine learning model 216 may be representative of a long short-term memory (LSTM). A LSTM model is a specific artificial recurrent neural network architecture capable of learning long term dependencies. LSTM model may learn to capture and/or determine multiple periodic patterns in a user's or client's data, such as weekly, semi-weekly, monthly, and the like. Using an LSTM model may allow machine learning platform 116 to model inflow, outflow, balance and/or any number of other variables together in a combined model, for any number of time steps, and for any number of users, all at once.);
- generating one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow with the aforementioned teachings of: determining one or more features associated with one or more financial information for one or more horizons based on the pre-processed data using a plurality of input AI models & generating, by the one or more hardware processors, one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons, and in view of Chan, whereby the method of Chan forecasts cash flow for one or more customers using one or more artificial intelligence processes. The forecasting may be implemented on an account level (e.g., for selected account(s) from among any number of accounts associated with a particular customer), on the customer level (e.g., across all accounts associated with the particular customer), and/or across a user base clusters (e.g., across multiple customers' accounts). The system may include a machine learning platform trained to forecast future customer activity based on historical customer activity, including for accounts for which the system has not been trained. Such functionality aims to assist clients to forecast their cash balances by considering net inflow and outflow activity (see at least Chan: ¶ [0019]). Moreover, the weight parameter may provide for a tradeoff between inflow/outflow prediction errors and the balance prediction errors. In other words, by adjusting the weight parameter, the objective function may be optimized for improving the balance forecast accuracy at the expense of a slightly less accurate inflow/outflow forecast (see at least Chan: ¶ [0050]).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Chan, the results of the combination were predictable.
Gianelle / Chan machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow does not explicitly disclose, but Mehra, et. al. in the analogous art for machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow teaches the following limitations:
- generating one or more forecasts for each horizon for a pre-determined time interval, using a stacked AI model comprising one or more forecasting models (see at least Mehra: (Under Proposed Methodology pages 299-300) & (Pages 309-310). Mehra teaches that Fig. 1, is to combine the predictive powers of a Random Forest model with a Custom Neural Net model to create an advanced forecasting system. Our goal is to improve the precision and dependability of upcoming three-month cash flow forecasts by utilizing the advantages of both models. By calculating the average of these models’ outputs, it will be possible to integrate them and provide a thorough and reliable method of financial forecasting. This study uses a unique strategy that combines the strengths of a Random Forest (RF) model with a Custom Neural Net model to improve the accuracy of future three-month forecasting. See also Mehra at Pages 302-303: “Using daily user datasets, the models are retrained independently by adjusting hyperparameters, adding fresh data, and removing the oldest data from the training set. The goal of reinforcement learning is to predict the inflow and outflow over the upcoming three months. A weighted average, with the weights derived from each model’s past performance, is used to combine the forecasts of the two models. This approach of dynamic reinforcement learning makes sure that the forecasting models are always able to adjust to the evolving patterns of user behavior.” See also Mehra at Page 309: The model unfolds in a systematic three-step process, each geared towards optimizing financial forecasting through the use of pretrained custom models. In the first step, the pretrained LSTM model is utilized to forecast the inflow and outflow for the next three months. Following the initial LSTM forecasting, the methodology seamlessly transitions into the second step, employing a pretrained Random Forest model. The Random Forest model is applied to forecast inflow and outflow for the same three-month horizon. The next critical phase involves parameter selection for both the LSTM and Random Forest models. In the third and final step, the predictions from the LSTM and Random Forest models are combined to form a unified forecast. Taking a weighted average of the individual model predictions allows for a balanced and comprehensive outlook on future inflow and outflow trends. The weighted averaging approach ensures that the unique strengths of each model contribute meaningfully to the final forecast. The outcomes of this combined prediction are then plotted, providing a visual representation of the forecasted financial trends over the next three months as per Fig. 5.), wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon (see at least Mehra: Fig. 1 & Fig. 5 & (Pages 302-303) & (Page 309-310). Mehra teaches that LSTM and random forest models that have already been trained are used in this stage. Using daily user datasets, the models are retrained independently by adjusting hyperparameters, adding fresh data, and removing the oldest data from the training set. The goal of reinforcement learning is to predict the inflow and outflow over the upcoming three months. A weighted average, with the weights derived from each model’s past performance, is used to combine the forecasts of the two models. This approach of dynamic reinforcement learning makes sure that the forecasting models are always able to adjust to the evolving patterns of user behavior. The models are kept current and adaptable to changing trends by incorporating daily user input into the training process. See also Page 309 of Mehra. See also Page 310 of Mehra: “The combination harnesses the power of LSTM’s sequential memory and Random Forest’s ensemble learning, resulting in a more nuanced understanding of the underlying patterns in cashflow data.”)
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow with the aforementioned teachings of: generating one or more forecasts for each horizon for a pre-determined time interval, using a stacked AI model comprising one or more forecasting models, wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon, and in further view of Mehra, whereby the hybrid model, which incorporates Random Forest, can handle complex interactions and enhance predictive accuracy by aggregating predictions from multiple decision trees. This approach aligns with the principle of model diversification to improve forecasting accuracy. The synergistic fusion of LSTM and Random Forest offers a comprehensive solution that leverages both models’ strengths, resulting in more accurate and reliable cash flow forecasting tools. This innovative approach aims to provide businesses with a more nuanced and resilient forecasting capability, enhancing their financial decision-making processes. The synergistic fusion of LSTM and both models, resulting in a more accurate and reliable cash flow forecasting tool. The combined model not only improves prediction accuracy but also mitigates the risk of overfitting, providing a sophisticated and well-rounded solution for financial analysts and decision-makers. This integration represents a significant leap forward in the quest for precise and dependable cash flow predictions in complex financial landscapes (see at least Mehra: Page 298).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Mehra, the results of the combination were predictable.
Gianelle / Chan / Mehra machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow does not explicitly disclose, but Ma in the analogous art for machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow teaches the following limitations:
- providing the generated one or more forecasts for the cash flow of the one or more business units (see at least Ma: ¶ [0015] & ¶ [0071] & Figs. 3-4. Ma teaches that liquidity forecasting is performed directly on corporate cash flow data, the results have high noise and low accuracy. By considering additional data instead of only the aggregate corporate cash flow, greater accuracy results. Separate models are used for each division or subsidiary of the corporation, and the final model 740 uses the results from those models to generate the final result 750 for the cash flow of the corporation. A second hierarchy, within each division or subsidiary, may be by data source (e.g., Associated Press, Bureau of Labor Statistics, World Bank, International Monetary Fund, Yahoo! ® Finance, Google® Finance, or any suitable combination thereof), by currency (e.g., dollar, euro, yen, yuan, or any suitable combination thereof), by political region (e.g., the European Union, the United States, China, or any suitable combination thereof). See also Ma at ¶ [0015]: “Predict liquidity using neural networks would be to use the historical cash position of a business over a period of time to train a neural network. The neural network is used to predict future liquidity based on the cash position to date.” See also Ma at Figs. 3-4.), as an output (see at least Ma: Fig. 5 & ¶ [0033-0035] & ¶ [0039] & ¶ [0043]. Ma notes that machine-learning algorithms operate by building an ML model 516 from example training data 512 in order to make data-driven predictions or decisions expressed as outputs or assessments 520. See also Ma at ¶ [0039]: The machine-learning algorithms utilize the training data 512 to find correlations among identified features 502 that affect the outcome. A feature 502 is an individual measurable property of a phenomenon being observed. The concept of a feature is related to that of an explanatory variable used in statistical techniques such as linear regression. Further, deep features represent the output of nodes in hidden layers of the deep neural network. See also Ma at ¶ [0043]: The training data 512 is time-series data comprising a sequence of values and the output of the machine learning model 516 is a predicted next value of the sequence. For example, 256 previous values may be used as an input feature and the 257th value used as the labeled output. ), to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users (see at least Ma: ¶ [0021] & Fig. 1 & Figs. 9-10. Ma notes client devices 160A and 160B in Fig. 1 and causing the predicted liquidity to be presented on a user interface of a client device at step 930 of Fig. 9. See also Ma at ¶ [0021]: “The application server 120 causes the trained neural network 150 to process the business data 140 to generate a liquidity forecast. The liquidity forecast is provided by the application server 120 to a client device 160 via the network 190 for display to a user.”)
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow with the aforementioned teachings of: providing the generated one or more forecasts for the cash flow of the one or more business units, as an output, to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users, and in further view of Ma, whereby multiple neural networks are each trained on time-series data from a different domain. Each of the trained neural networks is used to make a domain-specific prediction for each point in time. Thus, time-series prediction data is generated by each of the trained neural networks. The domain-specific time-series prediction data are combined into a vector and used to train a final model that predicts a value. By breaking down the problem of forecasting into domain-specific forecasting models and a forecasting model, accuracy is improved over traditional document-based forecasting and computational resources are saved over traditional neural network designs (see at least Ma: ¶ [abstract].). Furthermore, one way to improve the performance of DNNs is to identify newer structures for the feature-extraction layers, and another way is by improving the way the parameters are identified at the different layers for accomplishing a desired task. For a given neural network, there may be millions of parameters to be optimized. Trying to optimize all these parameters from scratch may take hours, days, or even weeks, depending on the amount of computing resources available and the amount of data in the training set (see at least Ma: ¶ [0064].).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Ma, the results of the combination were predictable.
Regarding Dependent Claim 2, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow teaches the limitations of Independent Claim 1 above, and Gianelle further teaches the machine-learning based (ML-based) method for forecasting cash flow comprising:
- wherein pre-processing the data (see at least Gianelle: ¶ [0058]. Gianelle teaches that after ingestion, the data often requires preprocessing, including cleaning, deduplication, and normalization, to ensure consistency across the different data streams. This step may involve handling missing data, resolving conflicts in data types, and applying transformations like date formatting or unit conversion. The cleaned data is then typically stored in a staging area, such as a data warehouse or a data lake.) comprises:
- standardizing (see at least Gianelle: ¶ [0061] & ¶ [0122]. Gianelle teaches that the system may normalize data by transforming the data into a standard format so that it can be easily combined and compared across different sources. This may involve standardizing naming conventions, data units, and encoding formats, ensuring all the data conforms to the same rules. For example, if different streams use varying currency formats, the system will convert all values to a common currency. Similarly, it may adjust data granularities (e.g., daily vs. monthly data) to make them compatible. Once the data is cleaned, deduplicated, and normalized, it becomes consistent and ready for aggregation, ensuring the analysis is accurate and meaningful across all datasets.), by the one or more hardware processors (see at least Gianelle: ¶ [0117-0119]. Gianelle notes that each of these devices may also include processors and/or control circuitry to send and receive commands, requests, and other suitable data using the I/O paths.), the data by transforming the data (see at least Gianelle: ¶ [0057-0061] & ¶ [0197]. Gianelle teaches that data may be grouped by time periods, cohorts, regions, and/or product categories, and then aggregated based on metrics like time period, start dates, and/or similar characteristics. This procedure often requires data cleaning, normalization, and transformation to ensure consistency and accuracy across different data sources before aggregation. The results are usually used for analysis, reporting, or decision-making purposes. This step may involve handling missing data, resolving conflicts in data types, and applying transformations like date formatting or unit conversion. The cleaned data is then typically stored in a staging area, such as a data warehouse or a data lake. See also Gianelle at ¶ [0197]: Feature engineering, the process of selecting, transforming, or creating features, is often performed to improve the model's performance and extract meaningful patterns from the data.) into an organized format being adaptable for performing an intelligent Al-inference process (see at least Gianelle: Fig. 6 & ¶ [0057-0061] & ¶ [0190]. Gianelle notes that data may be grouped by time periods, cohorts, regions, and/or product categories, and then aggregated based on metrics like time period, start dates, and/or similar characteristics. This procedure often requires data cleaning, normalization, and transformation to ensure consistency and accuracy across different data sources before aggregation. The results are usually used for analysis, reporting, or decision-making purposes. See also Gianelle at ¶ [0061]: The system may normalize data by transforming the data into a standard format so that it can be easily combined and compared across different sources. This may involve standardizing naming conventions, data units, and encoding formats, ensuring all the data conforms to the same rules. For example, if different streams use varying currency formats, the system will convert all values to a common currency. Similarly, it may adjust data granularities (e.g., daily vs. monthly data) to make them compatible. Once the data is cleaned, deduplicated, and normalized, it becomes consistent and ready for aggregation, ensuring the analysis is accurate and meaningful across all datasets. See also Gianelle at ¶ [abstract] & Fig. 6: Systems and methods are described herein for novel uses and/or improvements to data aggregation related to artificial intelligence applications, specifically applications related to aggregating time-series data. As one example, systems and methods are described herein for predicting effects when aggregating time-series data and modifying the one or more data streams used to populate a model profile and/or feed an artificial intelligence application with the time-series data. FIG. 6 shows a flowchart of the steps involved in generating time-series predictions using artificial intelligence models based on cohort clusters. See also Gianelle at ¶ [0190]: By clustering the data streams, the model can identify and filter out the noise or outliers. The clusters representing the noise can be ignored or treated separately, reducing the impact of noisy data on the model's training and inference processes. ), wherein standardizing the data (see at least Gianelle: ¶ [0061] & ¶ [0122]. Gianelle teaches that the system may normalize data by transforming the data into a standard format so that it can be easily combined and compared across different sources. This may involve standardizing naming conventions, data units, and encoding formats, ensuring all the data conforms to the same rules. For example, if different streams use varying currency formats, the system will convert all values to a common currency. Similarly, it may adjust data granularities (e.g., daily vs. monthly data) to make them compatible. Once the data is cleaned, deduplicated, and normalized, it becomes consistent and ready for aggregation, ensuring the analysis is accurate and meaningful across all datasets.) comprises at least one of: standardizing one or more column names, converting date and numeric formats across one or more dataset fields (see at least Gianelle: ¶ [0058-0061] & ¶ [0196-0198] & ¶ [0218]. Gianelle notes filling in missing values with default or estimated figures, correcting invalid data types, or removing entries that are clearly outliers or do not meet predefined quality standards. Cleaning can also involve formatting data properly, such as ensuring consistent date and time formats or applying standard units of measure across all data streams. Also the system may then apply the normalization factor to the second time-series data stream cluster to generate the first state characteristic. For example, when dealing with different data streams, normalization techniques can be applied to ensure that the data from each stream is brought to a common scale or range. The system may first determine the characteristics and properties of each data stream such as the type of data (numerical, categorical, etc. See also Gianelle at ¶ [0196-0198]: For example, the system may determine a type of data in the first model profile and select a feature input (or vector array type) based on the type of data. The system may select a numeric type. These are numeric values that represent measurable quantities. Examples include temperature, age, height, or any other continuous or discrete numerical variables. Numerical features are often used in regression or numerical prediction tasks. The system may select categorical features. These are non-numeric variables that represent different categories or classes.), handling null records, standardizing one or more financial transactions by currency conversion (see at least Gianelle: ¶ [0061]. The system may normalize data by transforming the data into a standard format so that it can be easily combined and compared across different sources. This may involve standardizing naming conventions, data units, and encoding formats, ensuring all the data conforms to the same rules. For example, if different streams use varying currency formats, the system will convert all values to a common currency. Similarly, it may adjust data granularities (e.g., daily vs. monthly data) to make them compatible. Once the data is cleaned, deduplicated, and normalized, it becomes consistent and ready for aggregation, ensuring the analysis is accurate and meaningful across all datasets.), and removing one or more intra-company financial transactions in the one or more financial transactions.).
Regarding Dependent Claim 3, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow teaches the limitations of Claims 1-2 above, and Gianelle further teaches the machine-learning based (ML-based) method for forecasting cash flow comprising:
- wherein pre-processing the data (see at least Gianelle: ¶ [0058]. Gianelle teaches that after ingestion, the data often requires preprocessing, including cleaning, deduplication, and normalization, to ensure consistency across the different data streams. This step may involve handling missing data, resolving conflicts in data types, and applying transformations like date formatting or unit conversion. The cleaned data is then typically stored in a staging area, such as a data warehouse or a data lake.) further comprises performing, by the one or more hardware processors (see at least Gianelle: ¶ [0117-0119]. Gianelle notes that each of these devices may also include processors and/or control circuitry to send and receive commands, requests, and other suitable data using the I/O paths.), the intelligent Al-inference process (see at least Gianelle: Fig. 6 & ¶ [0057-0061] & ¶ [0190]. Gianelle notes that data may be grouped by time periods, cohorts, regions, and/or product categories, and then aggregated based on metrics like time period, start dates, and/or similar characteristics. This procedure often requires data cleaning, normalization, and transformation to ensure consistency and accuracy across different data sources before aggregation. The results are usually used for analysis, reporting, or decision-making purposes. See also Gianelle at ¶ [0061]: The system may normalize data by transforming the data into a standard format so that it can be easily combined and compared across different sources. This may involve standardizing naming conventions, data units, and encoding formats, ensuring all the data conforms to the same rules. For example, if different streams use varying currency formats, the system will convert all values to a common currency. Similarly, it may adjust data granularities (e.g., daily vs. monthly data) to make them compatible. Once the data is cleaned, deduplicated, and normalized, it becomes consistent and ready for aggregation, ensuring the analysis is accurate and meaningful across all datasets. See also Gianelle at ¶ [abstract] & Fig. 6: Systems and methods are described herein for novel uses and/or improvements to data aggregation related to artificial intelligence applications, specifically applications related to aggregating time-series data. As one example, systems and methods are described herein for predicting effects when aggregating time-series data and modifying the one or more data streams used to populate a model profile and/or feed an artificial intelligence application with the time-series data. FIG. 6 shows a flowchart of the steps involved in generating time-series predictions using artificial intelligence models based on cohort clusters. See also Gianelle at ¶ [0190]: By clustering the data streams, the model can identify and filter out the noise or outliers. The clusters representing the noise can be ignored or treated separately, reducing the impact of noisy data on the model's training and inference processes. ) for adapting the standardized data (see at least Gianelle: ¶ [0061] & ¶ [0122]. Gianelle teaches that the system may normalize data by transforming the data into a standard format so that it can be easily combined and compared across different sources. This may involve standardizing naming conventions, data units, and encoding formats, ensuring all the data conforms to the same rules. For example, if different streams use varying currency formats, the system will convert all values to a common currency. Similarly, it may adjust data granularities (e.g., daily vs. monthly data) to make them compatible. Once the data is cleaned, deduplicated, and normalized, it becomes consistent and ready for aggregation, ensuring the analysis is accurate and meaningful across all datasets.) for ingestion (see at least Gianelle: ¶ [0058].) by the plurality of input AI models (see at least Gianelle: Figs. 3A-B & Fig. 6 & ¶ [0021].), wherein performing the intelligent AI-inference process (see at least Gianelle: Fig. 6 & ¶ [0057-0061] & ¶ [0190]. Gianelle notes that data may be grouped by time periods, cohorts, regions, and/or product categories, and then aggregated based on metrics like time period, start dates, and/or similar characteristics. This procedure often requires data cleaning, normalization, and transformation to ensure consistency and accuracy across different data sources before aggregation. The results are usually used for analysis, reporting, or decision-making purposes. See also Gianelle at ¶ [0061]: The system may normalize data by transforming the data into a standard format so that it can be easily combined and compared across different sources. This may involve standardizing naming conventions, data units, and encoding formats, ensuring all the data conforms to the same rules. For example, if different streams use varying currency formats, the system will convert all values to a common currency. Similarly, it may adjust data granularities (e.g., daily vs. monthly data) to make them compatible. Once the data is cleaned, deduplicated, and normalized, it becomes consistent and ready for aggregation, ensuring the analysis is accurate and meaningful across all datasets. See also Gianelle at ¶ [abstract] & Fig. 6: Systems and methods are described herein for novel uses and/or improvements to data aggregation related to artificial intelligence applications, specifically applications related to aggregating time-series data. As one example, systems and methods are described herein for predicting effects when aggregating time-series data and modifying the one or more data streams used to populate a model profile and/or feed an artificial intelligence application with the time-series data. FIG. 6 shows a flowchart of the steps involved in generating time-series predictions using artificial intelligence models based on cohort clusters. See also Gianelle at ¶ [0190]: By clustering the data streams, the model can identify and filter out the noise or outliers. The clusters representing the noise can be ignored or treated separately, reducing the impact of noisy data on the model's training and inference processes.) comprises at least one of:
- grouping, by the one or more hardware processors (see at least Gianelle: ¶ [0117-0119]. Gianelle notes that each of these devices may also include processors and/or control circuitry to send and receive commands, requests, and other suitable data using the I/O paths.), one or more information associated with equivalent entities of the one or more business units (see at least Gianelle: [0025-0026] & ¶ [0057] & ¶ [0084] & ¶ [0168]. Gianelle notes that at step 604, process 600 determines a cohort of a user using artificial intelligence models based on cohort clusters. For example, the methods and systems may include a first artificial intelligence model, wherein the first artificial intelligence model is trained to cluster a plurality of separate time-series data streams into a plurality of cohort clusters (e.g., through unsupervised hierarchical clustering). For example, as opposed to manually grouping potential cohorts, the system may train an artificial intelligence model to identify common user characteristics that correspond to a group of cohorts. See also Gianelle at ¶ [0025-0026]: The UOM of the participating entities may be the same, but this is not a requirement. For example, the data used to determine the correlation between the entities may be actual or synthetic time-series data. See also Gianelle at ¶ [0057]: Depending on the complexity, the system may filter, group, and/or merge data to organize it in meaningful categories. For example, data may be grouped by time periods, cohorts, regions, and/or product categories, and then aggregated based on metrics like time period, start dates, and/or similar characteristics. See also Gianelle at ¶ [0084]: A cohort may comprise a set that consists of the entities that experienced the data being captured and analyzed. Cohorts may share common attributes that cause them to be selected as members of the set. The system may analyze the cohort data to discern patterns that lead to predicted outcomes. The cohort data may comprise numerous data streams with many units of measures. The system may then segregate the data streams by cohort);
- generating, by the one or more hardware processors (see at least Gianelle: ¶ [0117-0119]. Gianelle notes that each of these devices may also include processors and/or control circuitry to send and receive commands, requests, and other suitable data using the I/O paths.), one or more synthetic invoices with unpaid open amount upon identifying one or more partially paid invoices (see at least Gianelle: ¶ [0031-0033] & ¶ [0109] & ¶ [0194]. Gianelle teaches or renders obvious the limitation of generating synthetic invoices with unpaid open amounts upon identifying partially paid invoices because it discloses creating a predictive synthetic profile from filtered historic data based on current state trajectories. The reference teaches using a first data set to determine the "trajectory of a current state". In the context of invoicing, identifying a partially paid invoice represents determining this current state and trajectory. The reference discloses creating a "synthetic profile" using actual historic time-series data that reflects rates of change [Reference]. Generating a synthetic invoice with an unpaid open amount is structurally and functionally equivalent to this synthetic profile used to project future financial states. The reference teaches filtering a historic data set based on "similarities between the current state characteristics" (e.g., similar states at the beginning or ending of a period, trajectories, or user profiles). Using a partially paid invoice characteristic to select matching historic payment data to generate the synthetic unpaid invoice is a direct application of this filtering teaching. The reference analyzes the synthetic data set for "potentially significant events" beyond a threshold [Reference], which aligns with predicting remaining unpaid open amounts on partial payments. It would have been obvious to a person of ordinary skill in the art at the time of the invention to apply the reference’s method of generating a synthetic profile using similarity-filtered historic data to the field of financial or invoice management (see at least Gianelle: ¶ [0094-0096] & ¶ [0199-0200]). The reference explicitly provides a general-purpose data prediction engine that operates on non-homogenous, time-series data to forecast outlier events or trajectories. Applying this known predictive technique to financial records—specifically treating a partially paid invoice as a "current state" and generating a synthetic profile representing the unpaid balance—would yield predictable results using known programming and data analysis methods.);
- aggregating, by the one or more hardware processors, one or more invoices with
information associated with in-period invoice creation and clearance, on week-level;
- updating, by the one or more hardware processors, a clearing date of the one or more
invoices with posting of payments; and
- automatically detecting, by the one or more hardware processors, the data associated
with at least one of: the one or more invoices, one or more payments, one or more deductions,
and one or more credit memos, for ingestion by the plurality of input Al models.
Regarding Dependent Claim 10, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow teaches the limitations of Independent Claim 1 above, and Chan further teaches the machine-learning based (ML-based) method for forecasting cash flow comprising:
- wherein determining the one or more features associated with the one or more financial information using the plurality of input Al models (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) further comprises:
- determining, by the one or more hardware processors (see at least Chan: ¶ [0036] & Fig. 2.), cash flow for a predetermined horizon based (see at least Chan: ¶ [0019] & ¶ [0074].) on at least one of: one or more historical trends and seasonality with time series data (see at least Chan: ¶ [0023-0025].), using a sixth input AI model (see at least Chan: Fig. 2 & ¶ [0019] & ¶ [0056].)
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow with the aforementioned teachings of: determining, by the one or more hardware processors, cash flow for a predetermined horizon based on at least one of: one or more historical trends and seasonality with time series data, using a sixth input AI model, and in view of Chan, whereby the method of Chan forecasts cash flow for one or more customers using one or more artificial intelligence processes. The forecasting may be implemented on an account level (e.g., for selected account(s) from among any number of accounts associated with a particular customer), on the customer level (e.g., across all accounts associated with the particular customer), and/or across a user base clusters (e.g., across multiple customers' accounts). The system may include a machine learning platform trained to forecast future customer activity based on historical customer activity, including for accounts for which the system has not been trained. Such functionality aims to assist clients to forecast their cash balances by considering net inflow and outflow activity (see at least Chan: ¶ [0019]). Moreover, the weight parameter may provide for a tradeoff between inflow/outflow prediction errors and the balance prediction errors. In other words, by adjusting the weight parameter, the objective function may be optimized for improving the balance forecast accuracy at the expense of a slightly less accurate inflow/outflow forecast (see at least Chan: ¶ [0050]).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Chan, the results of the combination were predictable.
Regarding Dependent Claims 11 and 16, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method / system for forecasting cash flow teaches the limitations of Independent Claims 1 and 14 above, and Gianelle further teaches the machine-learning based (ML-based) method / system for forecasting cash flow comprising:
- training (see at least Gianelle: Figs. 3A-3B & Fig. 4.), by the one or more hardware processors (see at least Gianelle: ¶ [0117-0119].), the one or more forecasting models of the stacked AI model incrementally on each feature combination of the one or more feature combinations for each horizon of the one or more horizons, using an incremental learning technique, wherein the incremental learning technique is configured to adapt the one or more forecasting models to learn knowledge of the stacked AI model in addition to previously acquired information for generating the one or more feature combinations (see at least Gianelle: ¶ [0081] & ¶ [0109] & ¶ [0190] & ¶ [0199]. Gianelle teaches that system 300 includes model 302 a, which may be a machine learning model, artificial intelligence model, etc. (which may be referred to collectively as “models” herein). Model 302 a may take inputs 304 a and provide outputs 306 a. The inputs may include multiple data sets, such as a training data set and a test data set. Each of the plurality of data sets (e.g., inputs 304 a) may include data subsets related to user data, predicted forecasts and/or errors, and/or actual forecasts and/or errors. Outputs 306 a may be fed back to model 302 a as input to train model 302 (e.g., alone or in conjunction with user indications of the accuracy of outputs 306 a, labels associated with the inputs, or with other reference feedback information). For example, the system may receive a first labeled feature input, wherein the first labeled feature input is labeled with a known prediction for the first labeled feature input. The system may then train the first machine learning model to classify the first labeled feature input with the known prediction (e.g., select a second data set from a plurality of available data sets based on similarities between state characteristics for the second data set and the current state characteristic and the required future state characteristic). See also Gianelle at ¶ [0190]: Clustering data streams can be particularly useful in scenarios where new data points continuously arrive in a streaming fashion. Clustering algorithms that support incremental learning can adapt to the evolving data stream by updating the existing clusters or creating new ones. This allows the machine learning model to efficiently process and learn from new incoming data without retraining the entire model. Incremental learning enables real-time or near-real-time analysis, improving efficiency by handling streaming data in a more agile and scalable manner.)
8. Claims 5-9 are rejected under 35 U.S.C. 103 as being unpatentable over US PG Pub (US 2025/0103605 A1) hereinafter Gianelle, et. al., in view of US PG Pub (US 2024/0037370 A1) hereinafter Chan, et. al., in further view of NPL Document: Predictive analytics and AI-driven strategies for enhanced cash flow forecasting. In Intelligent Systems Conference (pp. 296-315). Cham: Springer Nature Switzerland, hereinafter Mehra, et. al., and in further view of US PG Pub (US 2021/0303970 A1) hereinafter Ma, et. al, and in further view of US PG Pub (US 2026/0030662 A1) hereinafter Vlodinger, et. al.
Regarding Dependent Claim 5, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow teaches the limitations of Independent Claim 1 above, and Chan further teaches the machine-learning based (ML-based) method for forecasting cash flow comprising:
- wherein determining the one or more features associated with the one or more financial information using the plurality of input Al models (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) further comprises:
However, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow does not explicitly disclose, but Vlodinger in the analogous art for machine-learning based (ML-based) method for forecasting cash flow teaches the following limitations:
- determining, by the one or more hardware processors (see at least Vlodinger: Fig. 3 & Fig. 10 & ¶ [0319].), a payment date of one or more open invoices at a time of forecast based on one or more historical payment patterns, using a first input AI model of the plurality of input AI models (see at least Vlodinger:
- computing, by the one or more hardware processors (see at least Vlodinger: Fig. 3 & Fig. 10 & ¶ [0319].), an expected amount from the one or more open invoices across n number of weeks (see at least Vlodinger: ¶ [0555] & ¶ [0644] & ¶ [0773]. Vlodinger notes that the actual payment(s) are made per invoice 195 (block 1474). This can of course happen days, weeks or even months after the prediction was provided in step 1465. This can lead to a reconciliation with the invoice, which is associating with the invoice 195 an indication that a payment of a certain amount was made on a certain date, for that invoice. ) from a week of forecast run, based on the determined payment date of the one or more invoices (see at least Vlodinger: Figs. 13-14 & ¶ [0724]. Vlodinger teaches that purely for reasons of simplicity of exposition system 110 is further configured to re-train the machine learning model(s) 1120, 1170, based on an error in the prediction. This retraining can often occur asynchronously with the prediction. For example, when, at a later date, an actual payment is made against invoice #1234, and this information is reconciled or otherwise added to the invoice, the relevant models can be re-trained using this updated information. That is, the level of success of each forecast, indicated by one or more success parameters, measures or metrics, can be fed back into the learning. In one example, the data item indicative of the invoice 1234 is be updated, such that it has populated fields such as “Predicted Payment Date=June 15” and “Actual Payment Date=June 20”, and these can be fed into the re-training. In another example implementation, a “normalized average amount” error method is used instead, or additionally. Examples methods for these are disclosed with reference to FIG. 14.), using the first input AI model of the plurality of input AI models (see at least Vlodinger: ¶ [0407] & Fig. 4. Vlodinger teaches using enricher models 410, 460 disclosed with reference to FIG. 4 . The normalization module 320 works in combination with the enricher models.).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: determining, by the one or more hardware processors, a payment date of one or more open invoices at a time of forecast based on one or more historical payment patterns, using a first input AI model of the plurality of input AI models; and computing, by the one or more hardware processors, an expected amount from the one or more open invoices across n number of weeks from a week of forecast run, based on the determined payment date of the one or more invoices, using the first input AI model of the plurality of input AI models, and in further view of Vlodinger, in order for the enrichers 610 perform at least identification of one of the following: (i) the format of the one or more portions of the first record 190, 510;(ii) financial services associated with the first record;(iii) financial/business entities associated with the first record (including one or more counterparties);(iv) whether the actual financial transaction is an intercompany transaction or (v) transaction type. Thus, having generated enriched first records 530, the system 110 is further configured to perform re-training of the machine learning model(s) 410 using the enriched first record, thereby obtaining updated machine learning model(s) 410. That is, the currently enriched records can be used to improve the models (see at least Vlodinger: ¶ [0471].).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Vlodinger, the results of the combination were predictable.
Regarding Dependent Claim 6, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow teaches the limitations of Independent Claim 1 above, and Chan further teaches the machine-learning based (ML-based) method for forecasting cash flow comprising:
- wherein determining the one or more features associated with the one or more financial information using the plurality of input Al models (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) further comprises:
However, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow does not explicitly disclose, but Vlodinger in the analogous art for machine-learning based (ML-based) method for forecasting cash flow teaches the following limitations:
- determining, by the one or more hardware processors (see at least Vlodinger: Fig. 3 & Fig. 10 & ¶ [0319].), at least one of: an expected creation and clearance of one or more invoices, one or more credit memos (see at least Vlodinger: ¶ [0390]. Vlodinger notes that some of the second records 195 comprise at least one of the following information, e.g. located in fields of the record: entity identification information (e.g. name of the company for whom the records are being processed, or of a subsidiary of it), counterparty identification information, credit memo information, invoice information, and purchase order information.), and pre-deductions across the one or more horizons, using a second input AI model (see at least Vlodinger: ¶ [0407] & Fig. 4.).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: determining, by the one or more hardware processors, at least one of: an expected creation and clearance of one or more invoices, one or more credit memos, and pre-deductions across the one or more horizons, using a second input AI model, and in further view of Vlodinger, in order for the enrichers 610 perform at least identification of one of the following: (i) the format of the one or more portions of the first record 190, 510;(ii) financial services associated with the first record;(iii) financial/business entities associated with the first record (including one or more counterparties);(iv) whether the actual financial transaction is an intercompany transaction or (v) transaction type. Thus, having generated enriched first records 530, the system 110 is further configured to perform re-training of the machine learning model(s) 410 using the enriched first record, thereby obtaining updated machine learning model(s) 410. That is, the currently enriched records can be used to improve the models (see at least Vlodinger: ¶ [0471].).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Vlodinger, the results of the combination were predictable.
Regarding Dependent Claim 7, Gianelle / Chan / Mehra / Ma / Vlodinger machine-learning based (ML-based) method for forecasting cash flow teaches the limitations of Claims 1 and 5 above, and Chan further teaches the machine-learning based (ML-based) method for forecasting cash flow comprising:
- wherein determining the one or more features associated with the one or more financial information using the plurality of input Al models (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) further comprises:
Gianelle / Chan / Mehra / Ma / Vlodinger machine-learning based (ML-based) method for forecasting cash flow teaches the limitations of Claims 1 and 5 above, and Vlodinger further teaches the machine-learning based (ML-based) method for forecasting cash flow comprising:
- determining, by the one or more hardware processors (see at least Vlodinger: Fig. 3 & Fig. 10 & ¶ [0319].), a deduction to be claimed for the n number of weeks from the week of the forecast run, using a third input AI model, wherein a value associated with the n number of weeks depends on a week cutoff from the first input AI model (see at least Vlodinger: ¶ [0724] & Figs. 13-14.)
- computing, by the one or more hardware processors (see at least Vlodinger: Fig. 3 & Fig. 10 & ¶ [0319].), a deduction claim rate using at least one of: one or more closed accounts receivable (AR) invoices, and one or more actuals from bank statements, based on an optimum accuracy in historical time period, using the third input AI model (see at least Vlodinger: ¶ [0359] & ¶ [0391-0392] & ¶ [0815].).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra / Ma / Vlodinger machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: determining, by the one or more hardware processors, a deduction to be claimed for the n number of weeks from the week of the forecast run, using a third input AI model, wherein a value associated with the n number of weeks depends on a week cutoff from the first input AI model; and computing, by the one or more hardware processors, a deduction claim rate using at least one of: one or more closed accounts receivable (AR) invoices, and one or more actuals from bank statements, based on an optimum accuracy in historical time period, using the third input AI model, and in further view of Vlodinger, in order for the enrichers 610 perform at least identification of one of the following: (i) the format of the one or more portions of the first record 190, 510;(ii) financial services associated with the first record;(iii) financial/business entities associated with the first record (including one or more counterparties);(iv) whether the actual financial transaction is an intercompany transaction or (v) transaction type. Thus, having generated enriched first records 530, the system 110 is further configured to perform re-training of the machine learning model(s) 410 using the enriched first record, thereby obtaining updated machine learning model(s) 410. That is, the currently enriched records can be used to improve the models (see at least Vlodinger: ¶ [0471].).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Vlodinger, the results of the combination were predictable.
Regarding Dependent Claim 8, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow teaches the limitations of Independent Claim 1 above, and Chan further teaches the machine-learning based (ML-based) method for forecasting cash flow comprising:
- wherein determining the one or more features associated with the one or more financial information using the plurality of input Al models (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) further comprises:
However, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow does not explicitly disclose, but Vlodinger in the analogous art for machine-learning based (ML-based) method for forecasting cash flow teaches the following limitations:
- determining, by the one or more hardware processors (see at least Vlodinger: Fig. 3 & Fig. 10 & ¶ [0319].), an expected payment against at least one of (see at least Vlodinger: ¶ [0633-0634] & ¶ [0699] & ¶ [0770].): one or more sales orders and one or more purchase orders (see at least Vlodinger: ¶ [0390-0394] & ¶ [0571].), based on an average time taken from posting of (see at least Vlodinger: ¶ [0649] & ¶ [0663-0664] & ¶ [0779-0784].) at least one of: the one or more sales orders and the one or more purchase orders (see at least Vlodinger: ¶ [0390-0394] & ¶ [0571].), to clearance of the one or more invoices linked to (see at least Vlodinger: ¶ [0147-0151].) the at least one of: the one or more sales orders and the one or more purchase orders, on a user level (see at least Vlodinger: ¶ [0543-0544] & ¶ [0589] & & ¶ [0671].), using a fourth input AI model (see at least Vlodinger: ¶ [0407] & Fig. 4.);
- computing, by the one or more hardware processors (see at least Vlodinger: Fig. 3 & Fig. 10 & ¶ [0319].), an expected amount from (see at least Vlodinger: ¶ [0633-0634] & ¶ [0649] & ¶ [0779-0785] at least one of: the one or more sales orders and the one or more purchase orders (see at least Vlodinger: ¶ [0390-0394] & ¶ [0571].), across n number of weeks from a week of forecast run, based on the determined payment date of the one or more invoices (see at least Vlodinger: ¶ [0630] & ¶ [0644-0647] & ¶ [0724].), using the fourth input AI model (see at least Vlodinger: ¶ [0407] & Fig. 4.), wherein a value of the n number of weeks depends on a distribution of an actual clearance of at least one of: the one or more sales orders and the one or more purchase orders, across one or more weeks (see at least Vlodinger: ¶ [0390-0394] & ¶ [0571].)
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: determining, by the one or more hardware processors, an expected payment against at least one of: one or more sales orders and one or more purchase orders, based on an average time taken from posting of at least one of: the one or more sales orders and the one or more purchase orders, to clearance of the one or more invoices linked to the at least one of: the one or more sales orders and the one or more purchase orders, on a user level, using a fourth input AI model; and computing, by the one or more hardware processors, an expected amount from at least one of: the one or more sales orders and the one or more purchase orders, across n number of weeks from a week of forecast run, based on the determined payment date of the one or more invoices, using the fourth input AI model, wherein a value of the n number of weeks depends on a distribution of an actual clearance of at least one of: the one or more sales orders and the one or more purchase orders, across one or more weeks, and in further view of Vlodinger, in order for the enrichers 610 perform at least identification of one of the following: (i) the format of the one or more portions of the first record 190, 510;(ii) financial services associated with the first record;(iii) financial/business entities associated with the first record (including one or more counterparties);(iv) whether the actual financial transaction is an intercompany transaction or (v) transaction type. Thus, having generated enriched first records 530, the system 110 is further configured to perform re-training of the machine learning model(s) 410 using the enriched first record, thereby obtaining updated machine learning model(s) 410. That is, the currently enriched records can be used to improve the models (see at least Vlodinger: ¶ [0471].).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Vlodinger, the results of the combination were predictable.
Regarding Dependent Claim 9, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow teaches the limitations of Independent Claim 1 above, and Chan further teaches the machine-learning based (ML-based) method for forecasting cash flow comprising:
- wherein determining the one or more features associated with the one or more financial information using the plurality of input Al models (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) further comprises:
However, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow does not explicitly disclose, but Vlodinger in the analogous art for machine-learning based (ML-based) method for forecasting cash flow teaches the following limitations:
- determining, by the one or more hardware processors (see at least Vlodinger: Fig. 3 & Fig. 10 & ¶ [0319].) an expected credit application against one or more open invoices (see at least Vlodinger: ¶ [0377] & ¶ [0618] & ¶ [0633] & Table 1.) at a time of the forecast based on at (see at least Vlodinger: ¶ [0724] & Figs. 13-14.) least one of: one or more historical credit application patterns and a claim rate of the one or more users (see at least Vlodinger: ¶ [0390] & [0654]. Vlodinger teaches that If machine learning model(s) are created for a specific business entity, company J, they can be configured to provide company-specific payment profiles, by learning patterns of payment-related behavior of the business entity (how much they pay, when, how late, etc.). Each such entity may have its own dedicated model(s), or it may be represented as part of a larger model(s). In some examples, models for predicted date of month, delay, amount, and amount impact on delay are constructed for each relevant business entity with sufficient history.), using a fifth input AI model (see at least Vlodinger: ¶ [0407] & Fig. 4.)
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: determining, by the one or more hardware processors, an expected credit application against one or more open invoices at a time of the forecast based on at least one of: one or more historical credit application patterns and a claim rate of the one or more users, using a fifth input AI model, and in further view of Vlodinger, in order for the enrichers 610 perform at least identification of one of the following: (i) the format of the one or more portions of the first record 190, 510;(ii) financial services associated with the first record;(iii) financial/business entities associated with the first record (including one or more counterparties);(iv) whether the actual financial transaction is an intercompany transaction or (v) transaction type. Thus, having generated enriched first records 530, the system 110 is further configured to perform re-training of the machine learning model(s) 410 using the enriched first record, thereby obtaining updated machine learning model(s) 410. That is, the currently enriched records can be used to improve the models (see at least Vlodinger: ¶ [0471].).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Vlodinger, the results of the combination were predictable.
9. Claims 12 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over US PG Pub (US 2025/0103605 A1) hereinafter Gianelle, et. al., in view of US PG Pub (US 2024/0037370 A1) hereinafter Chan, et. al., in further view of NPL Document: Predictive analytics and AI-driven strategies for enhanced cash flow forecasting. In Intelligent Systems Conference (pp. 296-315). Cham: Springer Nature Switzerland, hereinafter Mehra, et. al., and in further view of US PG Pub (US 2021/0303970 A1) hereinafter Ma, et. al, and in further view of US PG Pub (US 2024/0420026 A1) hereinafter Breed-Allison, et. al.
Regarding Dependent Claims 12 and 17, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method / system for forecasting cash flow teaches the limitations of Independent Claims 1 and 14 above, and Mehra further teaches the machine-learning based (ML-based) method / system for forecasting cash flow comprising:
- assessing, by the one or more hardware processors, an accuracy of the one or more forecasting models of the stacked Al model in generating the one or more forecasts for each horizon for a pre-determined time interval, wherein assessing the accuracy of the stacked AI model (see at least Mehra: (Pages 309-311) & Tables 1-3.)
comprises:
- generating, by the one or more hardware processors, the one or more forecasts for each horizon for the pre-determined time interval (see at least Mehra: (Pages 308-311).)
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow with the aforementioned teachings of: assessing, by the one or more hardware processors, an accuracy of the one or more forecasting models of the stacked Al model in generating the one or more forecasts for each horizon for a pre-determined time interval, wherein assessing the accuracy of the stacked AI model & generating, by the one or more hardware processors, the one or more forecasts for each horizon for the pre-determined time interval, and in further view of Mehra, whereby the hybrid model, which incorporates Random Forest, can handle complex interactions and enhance predictive accuracy by aggregating predictions from multiple decision trees. This approach aligns with the principle of model diversification to improve forecasting accuracy. The synergistic fusion of LSTM and Random Forest offers a comprehensive solution that leverages both models’ strengths, resulting in more accurate and reliable cash flow forecasting tools. This innovative approach aims to provide businesses with a more nuanced and resilient forecasting capability, enhancing their financial decision-making processes. The synergistic fusion of LSTM and both models, resulting in a more accurate and reliable cash flow forecasting tool. The combined model not only improves prediction accuracy but also mitigates the risk of overfitting, providing a sophisticated and well-rounded solution for financial analysts and decision-makers. This integration represents a significant leap forward in the quest for precise and dependable cash flow predictions in complex financial landscapes (see at least Mehra: Page 298).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) non-transitory computer-readable medium for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Mehra, the results of the combination were predictable.
However, regarding Dependent Claims 12 and 17, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method / system for forecasting cash flow does not explicitly disclose, but Breed-Allison in the analogous art for machine-learning based (ML-based) method for forecasting cash flow teaches the following limitations:
- ranking (see at least Breed-Allison: ¶ [0089]. Breed-Allison teaches that the track models may stop running when the loss metric has reached a minimum. The register stage of the workflow 265 may sort through all model runs, rank the model with the best out of the sample performance, and register the best model to the ML Flow Model repository.), by the one or more hardware processors (see at least Breed-Allison: ¶ [0046].) , the one or more forecasts based on one or more parameters (see at least Breed-Allison: ¶ [0041 & Figs. 2K-2W].) comprising at least one of: weighted accuracy, R2-score, a first absolute error percentile for the pre-determined time interval, and a second absolute error percentile for the pre-determined time interval (see at least Breed-Allison: ¶ [0102] & ¶ [0117] & ¶ [0121].)
- assessing (see at least Breed-Allison: Figs. 1-2), by the one or more hardware processors (see at least Breed-Allison: ¶ [0046].), the accuracy of the stacked AI model (see at least Breed-Allison: Figs. 2T-2W].) based on the ranking of the one or more forecasts (see at least Breed-Allison: ¶ [0089] & Figs. 2K-2W.);
- selecting (see at least Breed-Allison: Figs. 1-2), by the one or more hardware processors (see at least Breed-Allison: ¶ [0046].), an optimized stacked AI model for generation of the one or more forecasts based on the one or more parameters and the accuracy of the one or more forecasting models (see at least Breed-Allison: Figs. 2K-2W.).
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: ranking, by the one or more hardware processors, the one or more forecasts based on one or more parameters comprising at least one of: weighted accuracy, R2- score, a first absolute error percentile for the pre-determined time interval, and a second absolute error percentile for the pre-determined time interval; and assessing, by the one or more hardware processors, the accuracy of the stacked AI model based on the ranking of the one or more forecasts; and selecting, by the one or more hardware processors, an optimized stacked AI model for generation of the one or more forecasts based on the one or more parameters and the accuracy of the one or more forecasting models, and in further view of Breed-Allison, in order to automatically learn and adapt from new data, continuously improving its predictive accuracy. By capturing cyclic trends and latent variables, the system may provide a granular, dynamic view of the demand for each product, enabling precise allocation of production resources to reduce surpluses and mitigate shortages effectively. The system may integrate time series modeling and machine-learning models for capturing complex temporal patterns and interdependencies inherent in the data. By combining historical data with data streams, the prediction models may generate forecasts that adapt dynamically to changing market dynamics. This integrated approach may enhance the accuracy of predictions but also facilitates agile decision-making in resource allocation, effectively optimizing productions to minimize surpluses and alleviate shortages (see at least Breed-Allison: ¶ [0031].).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Breed-Allison, the results of the combination were predictable.
10. Claims 13 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over US PG Pub (US 2025/0103605 A1) hereinafter Gianelle, et. al., in view of US PG Pub (US 2024/0037370 A1) hereinafter Chan, et. al., in further view of NPL Document: Predictive analytics and AI-driven strategies for enhanced cash flow forecasting. In Intelligent Systems Conference (pp. 296-315). Cham: Springer Nature Switzerland, hereinafter Mehra, et. al., and in further view of US PG Pub (US 2021/0303970 A1) hereinafter Ma, et. al, and in further view of US Patent # (US 12,210,949 B1) hereinafter Silver, et. al.
Regarding Dependent Claims 13 and 18, Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method / system for forecasting cash flow does not explicitly disclose, but Silver in the analogous art for machine-learning based (ML-based) method for forecasting cash flow teaches the following limitations:
- monitoring, by the one or more hardware processors, performance of the stacked AI model for a time duration (see at least Silver: Col. 20, Lns. 1-67 & Col. 44, Lns. 1-4 & Figs. 5-7);
- determining, by the one or more hardware processors, whether the performance of the stacked AI model with the data, is below a threshold value (see at least Silver: Col. 31, Lns. 60-67 & Figs. 5-7 & (Claim 52 of Silver).);
- fine-tuning, by the one or more hardware processors, the stacked AI model with at least one of: a plurality of parameters, one or more optimized features and horizon combinations (see at least Silver: Col. 7, Lns. 5-29 & Col. 27, Lns. 12-24.)
- evaluating, by the one or more hardware processors, the performance of the fine-tuned stacked AI model using a weighted average accuracy across the one or more forecasts for one or more horizons (see at least Silver: Col. 55, Lns. 60-67 & Col. 60, Lns. 22-33)
It would have been obvious for one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the teachings of Gianelle / Chan / Mehra / Ma machine-learning based (ML-based) method for forecasting cash flow with the aforementioned teachings of: monitoring, by the one or more hardware processors, performance of the stacked AI model for a time duration; determining, by the one or more hardware processors, whether the performance of the stacked AI model with the data, is below a threshold value; fine-tuning, by the one or more hardware processors, the stacked AI model with at least one of: a plurality of parameters, one or more optimized features and horizon combinations; and evaluating, by the one or more hardware processors, the performance of the fine-tuned stacked AI model using a weighted average accuracy across the one or more forecasts for one or more horizons, and in further view of Silver, whereby once errors are defined and identified, the system aggregates these errors to provide an overall error count. This count is then used to calculate error rates or other performance metrics like accuracy, precision, recall, and F1-score, which provide insights into the model's effectiveness and areas that may require improvement. The analysis might also involve deeper dives into the types of errors, their patterns, and potential causes, which can inform further model tuning, feature engineering, or even adjustments to the model architecture. This thorough evaluation helps ensure that the model is robust, reliable, and aligned with its application goals, thereby supporting better decision-making based on its outputs (see at least Silver: Col. 49, Lns. 13-25).
Further, the claimed invention is merely a combination of old elements in a similar field of a machine-learning based (ML-based) method for forecasting cash flow, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that, given the existing technical ability to combine the elements as evidenced by Silver, the results of the combination were predictable.
Examining Claims with Respect to Prior Art
11. Examiner deems that Dependent Claims 4, 15 and 20 are allowable over the prior art only. Please note that Claims 1-20 are still rejected under 35 U.S.C. § 101. Regarding Dependent Claim 4, there is no disclosure in the existing prior art or any new art that either teaches and/or discloses the sequence operation of features either individually or in combination relating to:
- wherein pre-processing the data further comprises performing, by the one or more hardware processors, a noise removal process for the data corresponding to each input Al model of the plurality of input Al models, wherein performing the noise removal process comprises at least one of:
- filtering, by the one or more hardware processors, the data associated with the one or more invoices posted and cleared within a same week, across training data to restrict adding noise to one or more user payment patterns;
- filtering, by the one or more hardware processors, the data associated with at least one of: one or more manually adjusted invoices, one or more invoice reversals with at least one of: a credit and debit note, and one or more prepaid invoices;
- filtering, by the one or more hardware processors, the data associated with horizons with abnormal deduction claim rates in history during computing projected deduction claim rate for a forecast period;
- filtering, by the one or more hardware processors, the data associated with at least one of: one or more prepaid sales orders and one or more purchase orders, along with clearance of one or more corresponding invoices occurring in a week; and
- filtering, by the one or more hardware processors, the data associated with at least one of: one or more credits that are blocked and the one or more credits that are posted and cleared within a same week, across the training data to restrict adding the noise to one or more user credit claim patterns.
The closest prior arts are as follows:
#1) (US 2025/0103605 A1) - Systems and methods for aggregating time-series data streams based on potential state characteristics following aggregation, hereinafter Gianelle, et. al.
#2) (US 2024/0037370 A1) - Automated data forecasting using machine learning, hereinafter Chan, et. al.,
#3) NPL Document: Predictive analytics and AI-driven strategies for enhanced cash flow forecasting. In Intelligent Systems Conference (pp. 296-315). Cham: Springer Nature Switzerland, hereinafter Mehra, et. al;
#4) (US 2021/0303970 A1) - Processing data using multiple neural networks, hereinafter Ma, et. al.
Regarding the Gianelle reference, Gianelle teaches and/or discloses the sequence operation of features of:
- data obtaining subsystem (see at least Gianelle: Fig. 3B & ¶ [0117-0119].) configured to obtain data associated with one or more business units (see at least Gianella: ¶ [0195] & ¶ [0199-0200]. Gianelle teaches that each user profile may have specific investment requirements. These requirements may be based on the investment itself (e.g., corresponding to a specific company, type of company, etc.) or a state of the portfolio (e.g., a required rate of return, level of risk, etc.). See also Gianelle at ¶ [0199-0200]: The system may analyze multiple sources to discern whether a user has any preferences regarding the types of companies associated with the bonds or equities that make up a given structured note. The system may employ web crawlers to determine whether the user has publicly expressed specific interests either for or against specific companies, technologies, ecological interests, social interests, governmental entities, high-profile persons, etc. ) from one or more data sources (see at least Gianelle: ¶ [0057-0058] & ¶ [0211] & ¶ [0218] & ¶ [0224]. Gianelle teaches that the system may determine potential data streams for the time-series data stream clusters based on the user's interests, respective cohorts, and/or other information about the user using data sourced from multiple sources such as account transaction data, portfolio management experiential data, external public data sources, market data, customer social media interests, user-specified directions, etc. to match a user with a structured note (e.g., represented by a plurality of time-series data streams and/or clusters thereof).) See also Gianelle at [0057-0058]: This procedure often requires data cleaning, normalization, and transformation to ensure consistency and accuracy across different data sources before aggregation. See also Gianelle at [0218]: For example, the system may determine that different data types reflect different aspects of data (e.g., the same time period, sampling frequency, data sources, data formats, etc.). See also Gianelle at [0224]: The system may receive a first dataset by first interfacing with a data source or repository that provides access to non-synthetic, real-world data.));
- a data pre-processing subsystem (see at least Gianelle: Fig. 3B & ¶ [0117-0119].) configured to pre-process the data to generate pre-processed data (see at least Gianelle: ¶ [0058]. Gianelle teaches that after ingestion, the data often requires preprocessing, including cleaning, deduplication, and normalization, to ensure consistency across the different data streams. This step may involve handling missing data, resolving conflicts in data types, and applying transformations like date formatting or unit conversion. The cleaned data is then typically stored in a staging area, such as a data warehouse or a data lake.).
Regarding the Chan reference, Chan teaches and/or discloses the sequence operation of features of:
- a feature determining subsystem configured (see at least Chan: Figs. 1-2 & Figs. 6A-6B.) to determine one or more features (see at least Chan: ¶ [0042] & ¶ [0101].) associated with one or more financial information for one or more horizons (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].) based on the pre-processed data (see at least Chan: ¶ [0085-0089] & ¶ [0101-0102] & ¶ [0109]. Chan notes generating the training data set may include pre-processing module 208 deriving additional data from the user-level information. For example, pre-processing module 208 derive a transaction type for various transactions included in the user-level data. Exemplary transaction type inputs may include, but are not limited to, rent payments, loan payments, one-time loan funds, and the like. These transactions may be further classified as usual and/or unusual payments. See also Chan at ¶ [0101-0102]: The system may then be configured to pre-process the historical account activity data. Pre-processing may include a combination of data imputation, data filtering, feature engineering and/or data scaling. Data imputation may comprise identifying data among the historical account activity data having missing values, and replacing the missing values with system-generated imputed values. See also Chan at ¶ [0109]: The system may be configured to perform further pre-processing operations on the current inflow activity data, the current outflow activity data, and the current balance information associated with the target account(s). The further pre-processing operations may include, for example, data imputation, feature engineering and data scaling.) using a plurality of input artificial intelligence (AI) models (see at least Chan: Fig. 2 noting “ML models 216” & ¶ [0019] & ¶ [0056]. Chan notes forecasting cash flow for one or more customers using one or more artificial intelligence processes. The forecasting may be implemented on an account level (e.g., for selected account(s) from among any number of accounts associated with a particular customer), on the customer level (e.g., across all accounts associated with the particular customer), and/or across a user base clusters (e.g., across multiple customers' accounts). See also Chan at ¶ [0056]: Machine learning model 216 may be representative of a long short-term memory (LSTM). A LSTM model is a specific artificial recurrent neural network architecture capable of learning long term dependencies. LSTM model may learn to capture and/or determine multiple periodic patterns in a user's or client's data, such as weekly, semi-weekly, monthly, and the like. Using an LSTM model may allow machine learning platform 116 to model inflow, outflow, balance and/or any number of other variables together in a combined model, for any number of time steps, and for any number of users, all at once.);
- a feature combination generating subsystem configured to (see at least Chan: Figs. 1-2 & Figs. 6A-6B.) generate one or more feature combinations by integrating the one or more features associated with the one or more financial information, for each horizon of the one or more horizons (see at least Chan: ¶ [0005] & ¶ [0025] & ¶ [0033] & [0051] & ¶ [0064] & ¶ [0104]. Chan notes that the future inflow forecast may comprise an inflow forecast for each of a plurality of time steps in the multi-step time horizon, and the future outflow forecast may comprise an outflow forecast for each of a plurality of time steps in the multi-step time horizon. The system may then determine, based on the future inflow forecast and the future outflow forecast, a future balance forecast for the one or more target accounts that comprises a balance forecast for each of a plurality of time steps in the multi-step time horizon. See also Chan at ¶ [0025]: Chan is able to learn from historical data from across multiple accounts, intake multiple variable inputs, and generate multi-variable, multi-step forecasts (e.g., multiple days in a time horizon) for multiple entities, all at once. In this regard, the deep learning prediction model(s) described below may be characterized as a centralized, multi-entity, multi-step and multi-variable time series forecasting model(s). The prediction model(s) of the present disclosure may also be customizable, insofar as parameters such as look back period (e.g., period of historical data that may be used to train the prediction model(s)), time horizon (e.g., forecast time-steps), number of accounts, number of entities for which forecasts are generated, number of variables modeled, etc. may all be customized for the particular implementation. See also Chan at ¶ [0033]: Back-end computing system 104 may dynamically receive or retrieve daily financial data (e.g., cash balance, inflow data, outflow data, etc.) from one or more third-party systems 106 for use with machine learning platform 116. See also Chan at ¶ [0051]: Training module 210 may train machine learning model 216 to generate a single day's forecast. Training module 210 may train machine learning model 216 to forecast a longer horizon, i.e., more than a single day's forecast. For example, training module 210 may utilize forecasted values, as well as historical inflow and outflow data, to train machine learning model 216 to generate forecasts beyond a single day. See also Chan at ¶ [0064]: Machine learning model(s) 216 are deployed, the forecast module 212 may be configured to generate forecasts (e.g., future inflow data, future outflow data, future balance data, etc., as discussed above)) for any number of entities for any multi-step time horizon (e.g., 30 days, 90 days, etc.), all at once. See also Chan at ¶ [0104].).
Regarding the Mehra reference, Mehra teaches and/or discloses the sequence operation of features of:
- a forecast generating subsystem configured to generate one or more forecasts for each horizon for a pre-determined time interval, using a stacked AI model comprising one or more forecasting models (see at least Mehra: (Under Proposed Methodology pages 299-300) & (Pages 309-310). Mehra teaches that Fig. 1, is to combine the predictive powers of a Random Forest model with a Custom Neural Net model to create an advanced forecasting system. Our goal is to improve the precision and dependability of upcoming three-month cash flow forecasts by utilizing the advantages of both models. By calculating the average of these models’ outputs, it will be possible to integrate them and provide a thorough and reliable method of financial forecasting. This study uses a unique strategy that combines the strengths of a Random Forest (RF) model with a Custom Neural Net model to improve the accuracy of future three-month forecasting. See also Mehra at Pages 302-303: “Using daily user datasets, the models are retrained independently by adjusting hyperparameters, adding fresh data, and removing the oldest data from the training set. The goal of reinforcement learning is to predict the inflow and outflow over the upcoming three months. A weighted average, with the weights derived from each model’s past performance, is used to combine the forecasts of the two models. This approach of dynamic reinforcement learning makes sure that the forecasting models are always able to adjust to the evolving patterns of user behavior.” See also Mehra at Page 309: The model unfolds in a systematic three-step process, each geared towards optimizing financial forecasting through the use of pretrained custom models. In the first step, the pretrained LSTM model is utilized to forecast the inflow and outflow for the next three months. Following the initial LSTM forecasting, the methodology seamlessly transitions into the second step, employing a pretrained Random Forest model. The Random Forest model is applied to forecast inflow and outflow for the same three-month horizon. The next critical phase involves parameter selection for both the LSTM and Random Forest models. In the third and final step, the predictions from the LSTM and Random Forest models are combined to form a unified forecast. Taking a weighted average of the individual model predictions allows for a balanced and comprehensive outlook on future inflow and outflow trends. The weighted averaging approach ensures that the unique strengths of each model contribute meaningfully to the final forecast. The outcomes of this combined prediction are then plotted, providing a visual representation of the forecasted financial trends over the next three months as per Fig. 5.), wherein each feature combination of the one or more feature combinations is analyzed as an individual input for each forecasting model of the one or more forecasting models for generating the one or more forecasts for each horizon (see at least Mehra: Fig. 1 & Fig. 5 & (Pages 302-303) & (Page 309-310). Mehra teaches that LSTM and random forest models that have already been trained are used in this stage. Using daily user datasets, the models are retrained independently by adjusting hyperparameters, adding fresh data, and removing the oldest data from the training set. The goal of reinforcement learning is to predict the inflow and outflow over the upcoming three months. A weighted average, with the weights derived from each model’s past performance, is used to combine the forecasts of the two models. This approach of dynamic reinforcement learning makes sure that the forecasting models are always able to adjust to the evolving patterns of user behavior. The models are kept current and adaptable to changing trends by incorporating daily user input into the training process. See also Page 309 of Mehra. See also Page 310 of Mehra: “The combination harnesses the power of LSTM’s sequential memory and Random Forest’s ensemble learning, resulting in a more nuanced understanding of the underlying patterns in cashflow data.”).
Regarding the Ma reference, Ma teaches and/or discloses the sequence operation of features of:
- an output subsystem (see at least Ma: Fig. 1 & Figs. 9-10) configured to provide the generated one or more forecasts for the cash flow of the one or more business units (see at least Ma: ¶ [0015] & ¶ [0071] & Figs. 3-4. Ma teaches that liquidity forecasting is performed directly on corporate cash flow data, the results have high noise and low accuracy. By considering additional data instead of only the aggregate corporate cash flow, greater accuracy results. Separate models are used for each division or subsidiary of the corporation, and the final model 740 uses the results from those models to generate the final result 750 for the cash flow of the corporation. A second hierarchy, within each division or subsidiary, may be by data source (e.g., Associated Press, Bureau of Labor Statistics, World Bank, International Monetary Fund, Yahoo! ® Finance, Google® Finance, or any suitable combination thereof), by currency (e.g., dollar, euro, yen, yuan, or any suitable combination thereof), by political region (e.g., the European Union, the United States, China, or any suitable combination thereof). See also Ma at ¶ [0015]: “Predict liquidity using neural networks would be to use the historical cash position of a business over a period of time to train a neural network. The neural network is used to predict future liquidity based on the cash position to date.” See also Ma at Figs. 3-4.), as an output (see at least Ma: Fig. 5 & ¶ [0033-0035] & ¶ [0039] & ¶ [0043]. Ma notes that machine-learning algorithms operate by building an ML model 516 from example training data 512 in order to make data-driven predictions or decisions expressed as outputs or assessments 520. See also Ma at ¶ [0039]: The machine-learning algorithms utilize the training data 512 to find correlations among identified features 502 that affect the outcome. A feature 502 is an individual measurable property of a phenomenon being observed. The concept of a feature is related to that of an explanatory variable used in statistical techniques such as linear regression. Further, deep features represent the output of nodes in hidden layers of the deep neural network. See also Ma at ¶ [0043]: The training data 512 is time-series data comprising a sequence of values and the output of the machine learning model 516 is a predicted next value of the sequence. For example, 256 previous values may be used as an input feature and the 257th value used as the labeled output. ), to the one or more users on one or more user interfaces associated with one or more electronic devices associated with the one or more users (see at least Ma: ¶ [0021] & Fig. 1 & Figs. 9-10. Ma notes client devices 160A and 160B in Fig. 1 and causing the predicted liquidity to be presented on a user interface of a client device at step 930 of Fig. 9. See also Ma at ¶ [0021]: “The application server 120 causes the trained neural network 150 to process the business data 140 to generate a liquidity forecast. The liquidity forecast is provided by the application server 120 to a client device 160 via the network 190 for display to a user.”).
Therefore, when taken as a whole, the claims are not rendered obvious as the available prior art does not suggest or otherwise render obvious the noted features nor does the available art suggest or otherwise render obvious further modification of the evidence at hand. Such modification would require substantial reconstruction relying solely on improper hindsight bias, and thus would not be obvious.
Similarly, Regarding Dependent Claims 15 and 20, there is no disclosure in the existing prior art or any new art that either teaches and/or discloses the sequence operation of features either individually or in combination relating to:
- determine a payment date of one or more open invoices at a time of forecast based on one or more historical payment patterns, using a first input AI model of the plurality of input AI models;
- compute an expected amount from the one or more open invoices across n number of weeks from a week of forecast run, based on the determined payment date of the one or more invoices, using the first input AI model of the plurality of input AI models;
- determine at least one of: an expected creation and clearance of one or more invoices, one or more credit memos, and pre-deductions across the one or more horizons, using a second input AI model;
- determine a deduction to be claimed for the n number of weeks from the week of the forecast run, using a third input AI model, wherein a value associated with the n number of weeks depends on a week cutoff from the first input Al model;
- compute a deduction claim rate using at least one of: one or more closed accounts receivable (AR) invoices, and one or more actuals from bank statements, based on an optimum accuracy in historical time period, using the third input AI model;
- determine an expected payment against at least one of: one or more sales orders and one or more purchase orders, based on an average time taken from posting of at least one of: the one or more sales orders and the one or more purchase orders, to clearance of the one or more invoices linked to the at least one of: the one or more sales orders and the one or more purchase orders, on a user level, using a fourth input AI model;
- compute an expected amount from at least one of: the one or more sales orders and the one or more purchase orders, across n number of weeks from a week of forecast run, based on the determined payment date of the one or more invoices, using the fourth input AI model, wherein a value of the n number of weeks depends on a distribution of an actual clearance of at least one of: the one or more sales orders and the one or more purchase orders, across one or more weeks;
- determine an expected credit application against one or more open invoices at a time of the forecast based on at least one of: one or more historical credit application patterns and a claim rate of the one or more users, using a fifth input AI model;
- determine cash flow for a predetermined horizon based on at least one of: one or more historical trends and seasonality with time series data, using a sixth input AI model.
Therefore, when taken as a whole, the claims are not rendered obvious as the available prior art does not suggest or otherwise render obvious the noted features nor does the available art suggest or otherwise render obvious further modification of the evidence at hand. Such modification would require substantial reconstruction relying solely on improper hindsight bias, and thus would not be obvious.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
NPL Documents
Dadteev, Kazbek, Boris Shchukin, and Sergey Nemeshaev. "Using artificial intelligence technologies to predict cash flow." Procedia Computer Science 169 (2020): 264-268. (Year: 2020). See abstract of NPL Document: Financial planning of large commercial banks. This paper describes methods of forecasting cash flow volumes using regression model, ARIMA model and MLP neural network model.
Foreign Documents
WO-2023159115-A1 - System and method for aggregating and enriching data, hereinafter Leach, et. al. Leach at [abstract]: “An intelligent forecasting system that includes sources of financial and non-financial input data and a user interface generator for generating user interfaces having window elements that display the input data. A Model Prediction soft button for selecting one of the statistical forecasting models to apply to the input data, and a Simple Prediction soft button for automatically generating forecasts.”
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DERICK HOLZMACHER whose telephone number is (571) 270-7853. The examiner can normally be reached on Monday-Friday 9:00 AM – 6:30 PM EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, Applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Brian Epstein can be reached on 571-270-5389. The fax phone number for the organization where this application or proceeding is assigned is 571-270-8853.
Information regarding the status of an application may be obtained from Patent Center. Status information for published applications may be obtained from Patent Center. Status information for unpublished applications is available through Patent Center for authorized users only. Should you have questions about access to Patent Center, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/DERICK J HOLZMACHER/ Patent Examiner, Art Unit 3625A
/SARA GRACE BROWN/Primary Examiner, Art Unit 3625
Prosecution Insights