Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
Status of the Claims
Claims 1-20 are pending.
Claims 1, 5-8, 12-15 and 18-20 are currently amended.
The rejection under 35 USC 101 is maintained.
Response to Applicant Remarks
Applicant’s well-articulated remarks have been considered but are unpersuasive for the reasons below.
Regarding the rejection under 35 USC 101, Applicant argues that the claimed invention cannot be practically performed in the human mind. (Applicant’s 8/4/26 remarks, p.11, “Applicant submits that claims 1-
20, as amended, recite limitations that cannot practically be performed in the human
mind. As recited by claim 1, as amended, the method is performed "responsive to a
notification that a predetermined action has been performed on an online compensation
management system of an organization" and requires "automatically preventing, by the
processing device, the requested change to the digital record from being completed."
Automatically preventing a change to a digital record maintained by an online
compensation management system is not an act of observation, evaluation, judgment, or
opinion, and it cannot practically be performed in the human mind. Similarly, the
amended claims recite that the determination is performed using "a set of AI models"
comprising "a first AI model trained to detect an indication of a first type of fraud and to
produce a first output and a second AI model trained to detect an indication of a second
type of fraud and to produce a second output," where the set of AI models produces a
final output that is subject to a confidence-score comparison. The coordinated operation
of multiple fraud-type-specific AI models, each trained to detect a different type of fraud,
producing outputs that are combined into a final output and then filtered by a confidence-
score threshold, is not a process that can practically be performed in the human mind.”) The examiner respectfully disagrees.
The courts have held, in the context of computer-assisted methods, that claims are not made patent eligible under § 101 simply because they speed up human activity. See, e.g., Content Extraction, 776 F.3d at 1347; DealerTrack, 674 F.3d at 1333. The increased speed and efficiency resulting from use of computers (with no improved computer techniques) do not themselves create eligibility. The examiner respectfully suggests that detection of payroll or transaction fraud, or better yet catching a fraudster “in the act” to avert damage is an idea that predates computing. The courts have held that software claims do not become eligible by limiting their application to a particular field or environment – eligibility does not result from taking an abstract idea and adding in “do it on a computer” or “do it over the internet.” See, e.g., Intell. Ventures I LLC v. Capital One Bank (USA), 792 F.3d 1363, 1366 (Fed. Cir. 2015); Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205 (Fed. Cir. 2025). Although applicant has clarified that the claimed invention employs of a set of AI models and a confidence threshold, the examiner respectfully suggests that these are fairly well known architectures in the field. (See e.g., “Panel-of-Experts”, 1991, https://www.envisioning.com/vocab/panel-of-experts; Zewe, “A technique to improve both fairness and accuracy in artificial intelligence”, 7/2022, https://news.mit.edu/2022/fairness-accuracy-ai-models-0720)
Applicant also argues that the claimed invention is a practical application of an abstract idea. (Applicant’s 8/4/26 remarks, pp.12-13, “Specifically, the practical application of the claims lies in protecting the security of the online compensation management system by automatically preventing completion of a fraudulent change to a digital record, and in improving the accuracy of fraud detection by suppressing low-confidence fraud findings to reduce false positives… The USPTO's own Subject Matter Eligibility Example 47 (Anomaly Detection), issued as part of the 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence, confirms this analysis. In Example 47, claim 3, which recited training an ANN, detecting anomalies in network traffic, determining that a detected anomaly is associated with malicious network packets, detecting the source address, dropping the malicious packets in real time, and blocking future traffic from the source address, was held eligible because the claim as a whole integrates the judicial exception into a practical application by improving network security.”) The examiner respectfully disagrees.
The examiner notes that example 47 is distinguishable at least because it is directed toward computing networking technologies rather than fraud detection. The courts have consistently held that this type of invention, directed towards collecting and analyzing data, may not be patent eligible. (See e.g. Bozeman Financial LLC v. Federal Reserve Bank of Atlanta, 955 F.3d 971 (Fed. Cir. 2020), The Federal Circuit held that financial transaction software for detecting fraud was directed to the abstract idea of collecting and analyzing data/information to detect fraud or errors. The fact that preventative steps were involved in the method (“sending a notification to said payee bank participant to not process said electronic financial transaction when said parameters do not match”) would not save the method from being abstract.)
Applicant also argues that the claimed invention is significantly more than an abstract idea. (Applicant’s 8/4/26 remarks, p.15, “Specifically, obtaining digital records responsive to a notification that a predetermined action comprising a request to change a digital record has been performed, the use of a set of AI models comprising a first AI model trained to detect an indication of a first type of fraud and a second AI model trained to detect an indication of a second type of fraud whose outputs yield a final output, modification of the final output when an associated confidence is below a threshold confidence value, and the automatic prevention of the requested change to the digital record from being completed, taken together, go well beyond the generic computer functions of collecting, analyzing, and displaying information or receiving and transmitting data over a network that the Examiner has alleged to be conventional.”) The examiner respectfully disagrees.
The examiner does not assert that the claimed steps are necessarily generic. However, the steps are implemented on generic computing elements. At the level claimed, the examiner understands that the AI model does not represent an advancement in AI techniques, so much as applying an abstract idea using AI as a tool on a generic computer. Even if the claimed fraud prevention technique is unconventional, it is nevertheless an abstract idea. ““We may assume that the techniques claimed are “[g]roundbreaking, innovative, or even brilliant,” but that is not enough for eligibility… The claims here are ineligible because their innovation is an innovation in ineligible subject matter. ... No matter how much of an advance … the claims recite, the advance lies entirely in the realm of abstract ideas, with no plausibly alleged innovation in the nonabstract application realm. An advance of that nature is ineligible for patenting.” (SAP America, Inc. v. InvestPic, LLC, 898 F.3d 1161 (Fed. Cir. 2018))
Applicant’s amendments are addressed by the newly cited art.
Claim Rejections - 35 USC § 101
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.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Regarding independent claims 1, 8, 15 the claimed invention recites an abstract idea without significantly more. The claims recite the abstract idea of fraud detection which is a mental process. Other than reciting a processor and AI model nothing in the claims precludes the steps from being performed mentally. But for the processor and AI model the limitations on obtaining records responsive to notification of an action, generate model input, determine fraudulent action using a set of AI models, perform corrective action is a process that under its broadest reasonable interpretation could be performed by mentally but for the recitation of generic computer elements. If claim limitations, under the broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Further the above limitations related to fraud detection stripped of the identified additional and insignificant elements could also be considered a “Method of Organizing Human Activity” relating to the managing human behavior and interactions. Thus, the claims recite an abstract idea.
The judicial exception is not integrated into a practical application. The computers are recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using generic computer components. The additional element(s) does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. Simply implementing the abstract idea on a generic computer environment is not a practical application of the abstract idea and does not take the claim out of the mental process or method of organizing human activity grouping.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above, with respect to integration of the abstract idea into a practical application, the additional element of a processor and AI model amounts to no more than mere instructions to apply the exception using a generic computer components. Mere instructions to apply an exception using generic computer components cannot provide an inventive concept. Collecting, analyzing and displaying information, and receiving and transmitting over a network are conventional in the computing arts. (MPEP 2106.05h; See also MPEP 2106.05, Alice v. CLS, “. Nearly every computer will include a ‘communications controller’ and ‘data storage unit’ capable of performing the basic calculation, storage, and transmission functions required by the method claims.”). The claims are not patent eligible.
Regarding the dependent claims, these claims are directed to limitations which serve to limit the fraud detection steps. The subject matter of claims 2/9/16( records are generated by the compensation management system within a threshold time and prior), 3/10/17 (features comprise bank account data, login data or payroll setup data), 4 /11(features comprise payment history or timekeeping), 5/12/18 (model comprises decision trees), 6/13/19 (training on known fraud and fraud target), 7/14/20 (corrective action is prevention of change of record or fraud alert) appear to add additional steps to the abstract idea, implemented by generic computers. To the extent that the claims recite decision trees or training AI, the examiner understands these to be generally mathematical algorithms employed in the AI field, a type of abstract idea. These claims neither introduce a new abstract idea nor additional limitations which are significantly more than an abstract idea. They provide descriptive details that offer helpful context, but have no impact on statutory subject matter eligibility.
Therefore the limitations on the invention, when viewed individually and in ordered combination are directed to in-eligible subject matter.
Claim Rejections - 35 USC § 103
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.
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.
Claim(s) 1,2,3,4,6,7, 8,9,10,11,13,14,15,16,17,18,19,20 is/are rejected under 35 U.S.C. 103 as being unpatentable over
Schrage US20210004915A1 in view of
Baker 20070106582 in view of
Zewe, “A technique to improve both fairness and accuracy in artificial intelligence”, 7/2022, https://news.mit.edu/2022/fairness-accuracy-ai-models-0720
Regarding Claim 1,
on an online compensation management system of an organization
obtaining, by a processing device, a plurality of digital records, wherein each digital record comprises data indicating an action performed by a user associated with the online compensation management system of the organization,
and wherein the predetermined action comprises a request to change a digital record of the plurality of digital records
Schrage is directed to a payroll fraud detection machine learning system. (Schrage, abstract). Schrage discloses collecting data such as changes to payroll information. (Schrage, para 0052-55, “[0052] As shown, system 700 collects/receives a data set 702 which can include, among other things, exceptions of the following types: [0053] Employees with salary increases of 10% and over [0054] Employees who have changed their own account/payment data [0055] Employees with multiple bank account changes”)
generating, by the processing device, artificial intelligence (AI) model input data based on the plurality of digital records, wherein the AI model input data comprises a plurality of compensation management data features indicated by the plurality of digital records;
Schrage discloses training an AI model. (Schrage, para 0042, “To address the foregoing issues, FIG. 6 depicts a workflow 600 that may be executed by system 100 of FIG. 1 to initially train final evaluation engine 112 using an indirect training approach according to certain embodiments. One advantage of this approach is that it automatically labels data points for training using a set of rules, thereby creating a training data set without requiring the time-consuming task of manual labeling. More importantly, the set of rules that are used in this approach are chosen in a manner that enables final evaluation engine 112 to learn broader data patterns, or in other words data patterns that are not specifically encoded in the rules. This allows engine 112 to identify and flag employees that do not strictly conform to those rules but nonetheless appear suspicious by virtue of certain correlations between their data and the data in the training data set.”)
responsive to the plurality of digital records indicating that the at least one action performed by the at least one user is fraudulent, performing by the processing device a corrective action to protect the online compensation management system of the organization,
(Schrage, abstract, “The results of these ML algorithms (or chains of algorithms) are fed into a neural network-based final evaluation engine that outputs an indication of whether a given employee is suspicious and should be audited as a potential payroll fraud case.”)
Schrage does not explicitly disclose
responsive to a notification that a predetermined action has been performed
Baker is directed to a system for detecting transactional fraud. (Baker, abstract). Baker discloses obtaining transactional data in real time for fraud analysis. (Baker, para 0022, “ [0022] FIG. 1 is a functional block diagram illustrating a fraud detection system 100. The fraud detection system 100 may be configured to detect fraud in a "real time" mode, e.g., prior to, or as part of, the authorization process for the transaction.”)
determining, using a set of AI models and using the AI model input
data as input, whether the plurality of digital records indicate that at least one action
performed by at least one user is fraudulent, wherein the set of AI models comprises a
first AI model trained to detect an indication of a first type of fraud and to produce a first
output and a second AI model trained to detect an indication of a second type of fraud
and to produce a second output, wherein the set of AI models produces a final output
based on the first output and the second output, and wherein, responsive to the final
output indicating that the at least one action is fraudulent and a confidence score associated with the final output…
Baker discloses that a panel of experts may be trained on different fraud types and their output may be combined to evaluate fraud. (Baker, para 0037-39, “[0037] In one embodiment, at least a portion of the models 106 may be configured to define a "committee" or "panels of experts." A committee comprises a group of the models 106 that are optimized in terms of input variables and TRAITS for a particular type of fraud detection, particular type of transaction, particular type of entity or entity state, or suitable for processing any other particular data configurations. In one embodiment, each model 106 is configured to identify its area of expertise in terms of transaction or entity data to allow the model expert selector 204 to select the particular model 106 when suitable transaction data is being processed. By providing such configuration data, the system 100 can be configured more easily, with less configuration management, etc. to include new or different models 106 over time or in a particular instance of the system 100.
[0038] For example, the models 106 may comprise models 106 configured to process foreign transactions, jewelry purchases, and or young accounts ("thin files"). Thus, a young account (thin file) transacting at a jewelry store outside the United States may employ all three of these example models 106 to score the transaction. Each model 106 may provide a score and confidence level to the score summarizer 206 to be combined to generate a combined score and appropriate risk indicators.
[0039] Models 106 may be based on any suitable modeling technique including, for example, neural networks, cascades (which may significantly improve performance for real-time referrals if trained on fraud transaction tags (identifying training transactions as fraudulent) rather than fraud account tags (identifying accounts as being used fraudulently)), support vector machines (SVM), genetic algorithms (GA), fuzzy logic, fraud case-based reasoning, decision trees, naive Bayesian, logistic regression, and scorecards.”)
wherein the corrective action comprises automatically preventing, by the
processing device, the requested change to the digital record from being completed.
(Baker, para 0044, “[0044] In one embodiment, the models 106 may output various types of data indicative of fraud to the score summarize 206. In addition to fraud scores in a particular range, e.g., 1-999, secondary scores may be generated to provide additional ways of detecting or quantifying risks such as first-party fraud, identity theft, application fraud, and merchant risk. Such scores for the models 106 may depend on the characteristics of the transaction, any triggers in place for the account, and TRAITS of entities associated with particular transaction data. One or more of the models may also generate a "bust-out" score (e.g., in the range of 1-999) indicative of a cardholder who maintains good credit for a time period then suddenly begins spending larger amounts (without paying). One or more of the models may also generate a "mass compromise" score (e.g., in the range of 1-999) indicative of a multi-account compromise. It has been found that the use of TRAITS for groups of accountholders or a portfolio of accounts desirably improves the detection of mass compromise events by the models 106. Model scores may also include a point-of-sale (POS) Action Score (Refer Score) (1-999), for example, for real time identification and referral to the vendor or issuer of accounts or transactions that appear to be fraudulent prior to approval.”) It would have been obvious to one of ordinary skill in the art before the filing date of the invention to combine Schrage with the AI fraud detection of Baker with the motivation of preventing fraudulent transactions from occurring. Id.
Schrage does not explicitly disclose
… being below a threshold confidence value, the final output is modified to not indicate that the at least one action is fraudulent; and
Zewe is an article discussing accuracy in machine learning predictions. (Zewe, p.1). Zewe discloses that it is known to for an artificial intelligence to make a prediction, but the prediction may be rejected if it is below a confidence threshold. (Zewe, p.1, “For workers who use machine-learning models to help them make decisions, knowing when to trust a model’s predictions is not always an easy task, especially since these models are often so complex that their inner workings remain a mystery.
Users sometimes employ a technique, known as selective regression, in which the model estimates its confidence level for each prediction and will reject predictions when its confidence is too low. Then a human can examine those cases, gather additional information, and make a decision about each one manually.”) It would have been obvious to one of ordinary skill in the art before the filing date of the invention to combine Schrage and Baker with the threshold of Zewe with the motivation of evaluating AI decision making. Id.
Regarding Claim 2, Schrage, Baker and Zewe disclose the method of claim 1.
wherein the plurality of digital records comprises:
a first digital record generated by the online compensation management system within a threshold amount of time; and
one or more second digital records generated by the online compensation management system prior to generating the first digital record.
(Schrage, para 0069, “Input:
Salary increase data for all employees on a monthly basis for some time period (e.g., past twelve months)”)
Regarding Claim 3, Schrage, Baker and Zewe disclose the method of claim 1.
wherein the plurality of compensation management data features
comprises at least one of:
bank account data;
employee login data; or
employee payroll setup data.
See prior art rejection of claim 1.
Regarding Claim 4, Schrage discloses the method of claim 1.
wherein the plurality of compensation management data features comprises at least one of employee payment history data or employee timekeeping data.
(Schrage, para 0013, “As shown, system 100 receives a data set 102 from a number of external data sources 104(1)-(N) (e.g., external payroll systems, external HR systems, external banking systems, etc.). Data set 102 can comprise many different types of data, each of which may be relevant to payroll fraud detection. Examples of such data include (but are not limited to):
Employee data
Family data (i.e., family members of employees)
Employee bank data
Payments data (i.e., payments made to employees)
Payments over time
Employee timesheets”)
Regarding Claim 6, Schrage, Baker and Zewe disclose the method of claim 1.
further comprising training the set of AI models on a plurality of items
of a training dataset, wherein a portion of the plurality of items comprises:
values for at least a portion of the plurality of compensation management data features indicating known instances of fraudulent actions; and
a target output indicating the fraudulent actions.
(Schrage, para 0011, “Further, because this ML-based approach makes use of a trainable neural network for final evaluation, the neural network can be refined over time via feedback from auditors, resulting in relatively high reliability and precision in flagging potential fraud cases.”)
Regarding Claim 7, Schrage, Baker and Zewe disclose the method of claim 1.
wherein performing the corrective action comprises
causing the online compensation management system to generate a fraudulent action alert
presentable to an administrator of the online compensation management system on a user
interface (UI).
See prior art rejection of claim 1.
Regarding Claims 8,9,10,11,13,14,15,16,17,18,19,20
See prior art rejection of claims 1,2,3,4,6,7,8,1,2,3,6,7
Claims 5,12,18 are rejected under 35 U.S.C. 103 as being unpatentable over Schrage in view of Baker in view of Zewe in view of Yeri 20210097605
Regarding Claim 5, Schrage, Baker and Zewe disclose the method of claim 1.
Schrage does not explicitly disclose
the first AI model and the second AI model comprise a random forest comprising a plurality of decision trees;
a first tree of the random forest performs a first inference using a first subset of the
plurality of compensation management data features; and
a second tree of the random forest performs a second inference using a second subset of
the plurality of compensation management data features.
Yeri is directed to a system for determining fraudulent behavior by analyzing communications with a machine learning model. (Yeri, abstract). Yeri discloses that a fraud detection model could be based upon random forest decision trees. (Yeri, para 0066-67, “[0066] A type of skipgram modeling that the emerging component 506 can perform is Random Forest, which is a non-probabilistic classifier which itself generalizes the popular decision tree algorithm. For a given (binary) feature set, a decision tree iteratively applies splitting criterion (Entropy, Gini coefficient) to determine which feature provides the most discriminative partition over the training data at each step. The best feature is assigned to the current node and each binary value of that feature is assigned to one of the two branches from that node, with the data partitioned to each node according to the value of the connecting branch. This procedure is then iterated on each sub-node so that the decision tree grows and continues to sub-partition the data, until either each node of the tree corresponds to one label (providing a total classification for each possible feature path) or another stopping criterion (in what is often called a ‘pruned’ tree). To apply a decision tree to predict the class of a test item, the feature vector is ‘read off’ the tree from root node to frontier, following the path down the tree to the leaf with the corresponding label for that class.
[0067] A random forest generalizes this concept by building a set of decision trees, where each decision tree is given a random subset of the features and a random subset of the data, and returning the majority prediction from this ensemble. The key advantage of the random forest algorithm is that the random partitioning of the data to the trees makes the ensemble more tolerant of noise, because the trees which overfit noise in the data will tend to ‘cancel’ each other, due to the Central Limit Theorem. A key reason to utilize the random forest algorithm with skipgram is that it is robust to correlation of features, which is a challenging aspect of generating skipgram features.”) It would have been obvious to one of ordinary skill in the art before the filing date of the invention to combine Schrage, Baker and Zewe with the decision tree of Yeri with the motivation of detecting fraud. Id.
Regarding Claim 12, 18,
See prior art rejection of claim 5.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALLEN C CHEIN whose telephone number is (571)270-7985. The examiner can normally be reached Monday-Friday 8am -5pm.
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, Florian Zeender can be reached at (571) 272-6790. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/ALLEN C CHEIN/Primary Examiner, Art Unit 3627
