On 16 October 2019, the Bank of England published the results of a joint survey, conducted with the Financial Conduct Authority, researching the use of machine learning (ML) in UK financial services.
Following technological advances, including software and hardware developments, and increased availability of alternative datasets, ML methods have become a popular tool to improve internal business processes, as well as delivering better outcomes for customers.~
However, the BoE believes that if governance and controls are not updated in-line with technological developments then existing risks within the financial sector could worsen. Additionally, The FCA have recently stated their intention to evaluate whether current Treating Customers Fairly (TCF) rules extend to ethical data usage.
Key findings of the survey are as follows:
- The use of ML methods is increasing within the financial sector
1. Two thirds of respondents report that they already use ML to support client interaction, business decisions or transactions (‘live’ use)
2. Insurance and banking are the sectors within the survey sample with the most live cases
3. The median respondent expects the number of ML applications more than double over the next three years, with even larger expected growth (triple) in banking and insurance
- Development of ML methods has passed initial development phase and is in more advanced stages of deployment (‘live’ use), especially in banking and insurance
- ML methods are used across an array of functions, from back office to front office, but the most popular, and most advanced use, is within risk management and compliance (Regtech), for tasks such as AML and fraud monitoring
- Firms do not think regulation is a barrier to use of ML methods but need additional guidance on how to interpret current regulation. However, firms believe that legacy IT systems and data limitations provide hurdles to the implementation of ML methods
- Firms agree that ML could be an amplifier of existing risks, which may occur if governance frameworks do not keep in-line with technological advancements
- The most common safeguards against ML-associated risks are alert systems and ‘human-in-the-loop’ mechanisms
- Structured data sources are the most popular, but alternative datasets are being increasingly used
- Tree-based models are the most popular among ML methods, a highly flexible model that can model non-linear relationships, although natural language processing (NLP) is the second most popular
- 76% of the ML use cases are designed and developed in-house, with the remaining implemented by third-party providers. Insurance has the most in-house implementations, while non-bank lending has the least, an area that has been largely disrupted by ML methods and whose respondents were of a smaller size than insurance firms
Firms are advised to consider these developments and discuss them internally, including updating Directors and Senior Management Team, with a Board Briefing, if applicable.
While there is no specific regulation proposed, the use of algorithmic decision making in the financial sector is on the radar of Regulators, with the FCA stating their intention to undertake discovery work (which this survey forms a part of). It is advised that firms keep this in mind and ensure that customer data is not being misused such that firms remain compliant with GDPR.
To read more, please follow this link:
Contact us here
Please Note: This publication is not intended to be a comprehensive review of all developments in the law and practice, or to cover all aspects of those referred to. Readers should take legal advice before applying the information contained in this publication to specific issues or transactions.
- A message from Met Facilities CEO –Danny Kessler - 17th April 2020
- Financial Conduct Authority Statement on Assessing Suitability Review 2 - 16th March 2020
- European Systemic Risk Board letter on issues raised with Alternative Investment Fund Managers Directive - 13th March 2020