Regulatory Question Management System Using Vector Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Financial institutions face challenges in providing timely, accurate, and consistent responses to regulatory questions from internal audit entities and external regulators, such as the Federal Reserve Bank and the Office of the Comptroller of the Currency, due to the complexity and variability of these inquiries.
Innovation Solution
A regulatory management system that receives questions, identifies keywords, converts them into vectors, calculates similarity scores with stored questions, ranks and selects the most relevant answers from a knowledgebase, and submits them to regulators, while utilizing machine learning to improve answer selection and identify emerging risks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If questions are answered on a question-by-question basis by representatives, then responses can be tailored to specific entities, but response time is extended and consistency across responses becomes difficult to maintain
Solution Approach 1:
The system pre-processes and stores questions and answers in a knowledge base before they are needed during regulatory examinations. By anticipating and preparing potential questions and their answers in advance, the system enables rapid retrieval and consistent delivery of responses without requiring time-consuming manual analysis during the actual examination process
Solution Approach 2:
The system creates and maintains a reusable knowledge base of questions and answers that can be copied and applied across multiple regulatory examinations and different entities. This allows standardized responses to be efficiently replicated while maintaining consistency, reducing the time required to answer similar questions repeatedly
2Reliability
If multiple regulators ask questions about the same model, then comprehensive review is achieved, but the same questions may be answered differently, reducing credibility
Solution Approach 1:
The knowledge base serves as a universal repository that stores questions and answers applicable across multiple regulators, models, and examination contexts. This single centralized system enables consistent responses to be delivered to different regulators regarding the same or similar models, ensuring reliability and credibility while simplifying the management of cross-regulator questionnaires
Solution Approach 2:
The system incorporates feedback mechanisms that track which questions have been asked by which regulators and what answers have been provided. This feedback loop ensures that the same questions are answered consistently across different regulatory examinations and allows the system to learn from and adapt to regulatory questioning patterns, maintaining reliability
3Productivity
If regulators request short time to receive answers, then examination efficiency is improved, but accuracy and completeness of responses may be compromised
Solution Approach 1:
The system performs preliminary processing of questions and preparation of answers in advance, storing them in a ready-to-use knowledge base. This pre-computation and pre-organization of information enables the system to deliver accurate and complete responses within the short timeframes requested by regulators, as the analytical work has already been completed beforehand
Solution Approach 2:
The system replaces manual mechanical processes of question analysis and answer formulation with automated computational processes. By using algorithms and machine learning to analyze questions and generate or retrieve answers, the system achieves both high speed (meeting short deadlines) and high accuracy (maintaining response quality) without relying on manual human effort
Data Source
AI summary
The innovation disclosed and claimed herein, in one aspect thereof, comprises systems and methods of managing regulatory questions. The systems and methods receive a question having words and phrases. The systems and methods identify keywords in the question using a knowledgebase. The systems and methods determine closely related questions based on the identification, the closely related questions having answers associated with each question of the closely related questions. The systems and methods perform machine learning on the answers of the determined closely related questions.