ML Anomaly Detection for Expense Fraud
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing expense management systems face challenges in accurately and efficiently auditing travel and reimbursable expenses, particularly in detecting fraudulent submissions such as duplicate or machine-generated receipts, which can lead to resource wastage and revenue loss.
Innovation Solution
Implementing machine learning models for automated expense report auditing, including OCR, receipt token extraction, and policy compliance verification, to replace or augment human auditors, providing real-time feedback and reducing human error.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning models are implemented for automated expense report auditing, then auditing efficiency and accuracy are improved, but system complexity increases
Solution Approach 1:
The auditing system is divided into multiple specialized machine learning models, each trained to detect specific types of fraud (duplicate receipts, machine-generated receipts, altered receipts). This segmentation allows the system to handle different fraud detection tasks independently, improving overall efficiency while managing complexity through modular architecture.
Solution Approach 2:
The patent introduces an intermediary processing layer that receives expense reports, routes them to appropriate ML models based on detected fraud indicators, and aggregates results. This intermediary manages the complexity of coordinating multiple ML models while maintaining high auditing throughput and accuracy.
2Measurement precision
If multiple machine learning models are used for comprehensive fraud detection, then detection accuracy improves, but processing time increases
Solution Approach 1:
The system performs preliminary analysis using lightweight ML models first to identify suspicious patterns or indicators of fraud. Only reports showing fraud indicators are then subjected to more comprehensive analysis by additional specialized models. This preliminary filtering reduces the number of reports requiring full multi-model analysis, maintaining high accuracy while reducing average processing time.
Solution Approach 2:
The patent implements a continuous auditing pipeline where ML models process expense reports in real-time as they are submitted, rather than batch processing. This continuous action allows the system to provide immediate feedback and maintain high detection accuracy across all reports without significant delays, as the workflow remains uninterrupted and efficiently utilized.
3Reliability
If automated ML-based auditing is implemented, then human error is reduced, but initial implementation cost and complexity increase
Solution Approach 1:
The ML models are trained using historical expense report data and fraud cases from the organization's own past experiences. The system learns from its own historical data to detect fraud patterns specific to the organization, reducing the need for external expertise and simplifying implementation while maintaining high reliability through organization-specific learning.
Solution Approach 2:
The system incorporates feedback mechanisms where audit results (both automated ML decisions and manual reviewer corrections) are fed back into the training data. This continuous feedback loop allows the ML models to improve their accuracy over time based on actual performance, reducing human error while the incremental learning approach makes implementation more manageable by starting with baseline models that improve progressively.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
The present disclosure involves systems, software, and computer implemented methods for transaction auditing. One example method includes training at least one machine learning model to determine features that can be used to determine whether an image is an authentic image of a document or an automatically generated document image, using a training set of authentic images and a training set of automatically generated document images. A request to classify an image as either an authentic image of a document or an automatically generated document image is received. The machine learning model(s) are used to classify the image as either an authentic image of a document or an automatically generated document image, based on features included in the image that are identified by the machine learning model(s). A classification of the image is provided. The machine learning model(s) are updated based on the image and the classification of the image.