Expense Audit ML Model for Comment-Based Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated systems are inadequate in ensuring the correct expense type is selected for transactions, as they can only detect errors in a small percentage of expenses and do not account for expenses without receipts or those automatically approved.
Innovation Solution
A machine learning modeling system is trained to predict the correct expense type by analyzing unstructured comments entered by users during the expense submission process, allowing for real-time flagging or automatic updating of incorrect expense types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review of each expense is performed to ensure correct expense type selection, then accuracy is improved, but time consumption and labor requirements increase significantly
Solution Approach 1:
The patent replaces manual mechanical review processes with an automated machine learning classification system. The ML model processes expense descriptions and metadata to automatically predict correct expense types, eliminating the need for manual inspection of each expense item while maintaining high classification accuracy through trained neural networks.
Solution Approach 2:
The system enables self-service by automatically classifying expenses without requiring manual intervention. The machine learning model independently analyzes expense data, makes classifications, and can even correct previously misclassified expenses, allowing the system to serve itself rather than requiring continuous human oversight for routine classification tasks.
2Productivity
If automated processes are used to detect expense errors, then productivity is improved, but detection coverage is limited to only 2-10% of expenses
Solution Approach 1:
The patent changes the detection parameters from traditional rule-based checks (which only caught 2-10% of errors) to machine learning-based classification. The system uses trained neural networks that analyze multiple parameters including expense descriptions, amounts, and metadata to achieve comprehensive error detection across all expense types, not just a subset.
Solution Approach 2:
The machine learning system provides universal error detection capability across all expense types and scenarios. Unlike specialized rules that only work for specific expense categories, the ML model can generalize to detect errors in any expense type, making the system universally applicable to the entire expense reporting population.
3Ease of operation
If existing automated detection rules are applied, then ease of operation is improved, but they fail to detect errors in expenses without receipts or automatically approved expenses
Solution Approach 1:
The machine learning model acts as an intermediary that processes expense data in a unified manner regardless of whether receipts are present or expenses are automatically approved. The ML system analyzes the expense description and metadata as intermediate representations that capture the essence of the expense, enabling reliable detection across all expense categories without requiring traditional receipt-based validation.
Data Source
AI summary
Systems and methods are provided for training a machine learning model to use comments entered by a user submitting an expense to determine a correct expense type. The trained machine learning model is used to predict an expense type by analyzing submitted text comments corresponding to a submitted expense. The expense can be flagged if a mismatch is determined between the expense type of the submitted expense and the predicted expense type, or the submitted expense can be automatically updated to the predicted expense type.