Summary
We were inspired by ASIC's Regtech webinar on responsible lending, and carried out a proof of concept build of AI applications for ASIC's regulation of credit. The objective was to implement a demo App that uses Natural Language Processing to find the most relevant answers to questions related to ASIC's regulatory guides.
The demo App is available at this link.
Demo: Q&A - Responsible lending Conduct
Demo: Q&A - General conduct obligations
Demo: Q&A - Approval and oversight of external dispute resolution schemes
Introduction
There has been a considerable amount of attention focused on the conduct of financial services providers in Australia recently, particularly in the area of responsible lending. As a result, the industry is experiencing a regulatory wave of investigations, monitoring and remediation programs.
Regulation in finance is vast and complex. There are four bodies that regulate the finance industry. The Australian Securities and Investments Commission (ASIC) alone has over 250 regulatory guides. How do companies remain informed and maintain regulatory compliance? The industry has found the answer in Regtech.
Regtech is the management of regulatory monitoring, reporting, and compliance within the financial industry through technology. It disrupts the regulatory landscape by providing technologically advanced solutions to the ever increasing demands of compliance within the financial industry.
Some examples of Regtech:
Verifier - A company that provides services to verify income for consumers applying for loans, credit, and rental.
Alessa - Platform as a Service (PaaS) used for sanctions screening of clients, transaction monitoring, retail store monitoring and taxation reporting and customs compliance
6clicks - Combines regulation, risk assessment and risk management into a compliance collaboration platform.
Objective
The objective was to implement a demo App that uses Natural Language Processing to find the most relevant answers to questions related to ASIC's regulatory guides.
How to Use the App
Question text box: Enter your question here.
Number of relevant paragraphs: Number of top relevant paragraphs to search for the question.
Number of relevant answers: Number of top relevant answers related to the question.
Get Answers button: Button to submit question to the App to obtain the answers
Regulatory Guide: The regulatory guide document identifier related to the paragraph associated with the question.
Page: The regulatory guide page related to the paragraph associated with the question.
Section: The regulatory guide section related to the paragraph associated with the question.
Clause: The clause or regulation identifier related to the paragraph associated with the question.
Paragraph: The paragraph relevant with the question.
Highlighted Answers: The answers found by the App in the paragraph associated to the question.
Paragraph Relevance Score: A score of 0-100 on the relevance of the paragraph with the question. A higher score indicates a higher relevance.
Answers: A list of possible answers in the paragraph associated with the question.
Answer Relevance Probability: The probabilities of the answers' relevance in relation to the question (Score of 0-1). A higher score indicates a higher relevance.
Data
The following regulatory guides were curated into the App:
REGULATORY GUIDE 139: Approval and oversight of external dispute resolution schemes
REGULATORY GUIDE 165: Licensing: Internal and external dispute resolution
REGULATORY GUIDE 201: Unsolicited credit cards and debit cards
REGULATORY GUIDE 203: Do I need a credit licence?
REGULATORY GUIDE 204: Applying for and varying a credit licence
REGULATORY GUIDE 205: Credit licensing: General conduct obligations
REGULATORY GUIDE 206: Credit licensing: Competence and training
REGULATORY GUIDE 207: Credit licensing: Financial requirements
REGULATORY GUIDE 208: How ASIC charges fees for credit relief applications
REGULATORY GUIDE 209: Credit licensing: Responsible lending conduct
REGULATORY GUIDE 210: Compensation and insurance arrangements for credit licensees
REGULATORY GUIDE 218: Licensing: Administrative action against persons engaging in credit activities
REGULATORY GUIDE 234: Advertising financial products and services (including credit): Good practice guidance
REGULATORY GUIDE 267: Oversight of the Australian Financial Complaints Authority
REGULATORY GUIDE 270: Whistleblower policies
REGULATORY GUIDE 271: Internal dispute resolution
REGULATORY GUIDE 272: Product intervention power
REGULATORY GUIDE 273: Mortgage brokers: Best interests duty
REGULATORY GUIDE 51: Applications for relief
Machine Learning
The App uses a 2 step process to retrieve the answer from the documents:
Document Retrieval- A ranking function used by search engines to estimate the relevance of documents to a given search query. This step retrieves the most relevant regulatory guides associated with the question.
Closed Domain Q&A - A task of automatically answering a correct answer to the questions asked by human in natural language using either a pre-structured database or a collection of natural language documents. This step extracts answers from the most relevant regulatory guides associated with the question.
This published whitepaper contains further information of the techniques used for the Q&A engine.
The machine learning model was trained on the following regulatory guides:
REGULATORY GUIDE 139: Approval and oversight of external dispute resolution schemes
REGULATORY GUIDE 205: Credit licensing: General conduct obligations
REGULATORY GUIDE 209: Credit licensing: Responsible lending conduct
Conclusion
The results were pretty eye opening for a model trained on such a small training dataset.
More information about our company, platforms and what we do: