On the 21 November 2024, the Bank of England and the UK Financial Conduct Authority ("FCA") published the results from their third survey of artificial intelligence ("AI") and machine learning in UK financial services. The aim of the survey is to help the Bank of England and the FCA build on their existing work on understanding AI in financial services.
The survey had 118 respondents across six different sectors (UK banks, international banks, insurance, non-bank lending, investments and capital markets, financial market infrastructures, payments and others).
For the purposes of the survey, artificial intelligence was defined as "the simulation of human intelligence by machines, including the use of computer systems, which have the ability to perform tasks that demonstrate learning, decision-making, problem solving, and other tasks which previously required human intelligence".
Below is a breakdown of the main sections of the survey results, focusing on key findings and take aways.
The first section of the survey to be covered focused on the uptake of AI across the respondents and how it is being used.
There has been an increase in uptake of AI, with 75% of respondents saying that they are already using AI, up from 58% in 2022. Additionally, 10% of respondents said that they planned to use AI over the next three years (which is a slight decrease from the 14% reported in 2022).
The insurance sector has the highest percentage of firms using AI (95%) while financial market infrastructure firms have the lowest percentage of firms using AI (57%).
In terms of the use of AI, operations and IT was found to be the largest area of AI use cases (making up 22% of all use cases). Foundation models account for 17% of use cases, with operations and IT accounting for 30% of all foundation model use cases. The area with the highest percentage of respondents using AI is optimisation of internal processes (41%) and the area over the next three years expected to have the biggest growth is AI for customer support, with 36% of respondents expecting to use AI for that.
When asked to rate the number of use cases by level of materiality (a rating of the use case's impact) 62% of cases were rated as of low materiality, 22% as medium and 16% as high. High materiality was most common in use cases in general insurance, risk and compliance and retail banking. Low materiality was most common in operations and IT. 12% of foundation model use cases and 16% of third-party implementations were found to be of high materiality.
There seems to be a large uptake in automated decision making, with 55% of all AI use cases used by respondents containing such decision making to some degree. Additionally, 81% of respondents who use AI employ an explainability method. The most popular explainability methods were Feature importance (which provides general ranking of features) and Shapely (which considers all possible combinations of features to assess the impact of each one).
A third of all current AI use cases that are deployed by respondents are third party implementations (up from 17% in 2022). The top three third-party providers of cloud, model and data accounted for 73%, 44% and 33% of all name providers respectively.
The next section of the survey focused on the strategies and governance approaches the respondents were taking in regard to AI.
The most commonly used governance framework, control or process specific to AI was to have an accountable person/persons with responsibility for the AI framework (according to 84% of respondents currently using AI).
When it comes to accountability for AI use cases and their outputs, 72% of respondents which are using, or planning to use, AI said that they allocate this to executive leadership.
In regard to data management practices, most practices are not AI-specific. For example, 87% respondents discussed change management practices, with 71% being not AI-specific and 16% being AI-specific.
The most common metrics being used by respondents to monitor model effectiveness are accuracy, precision, recall and sensitivity (reported by 88% of firms using AI).
However, there is still some way to go with increasing understanding of AI across the financial sector. The survey found that 46% of respondents (who use or are planning to use AI) reported having a "partial understanding" of the AI technologies they use, and only 34% of respondents claimed to have "complete understanding".
The final section of the survey focused on what the respondents perceived to be the benefits and risks of implementing AI.
When it comes to benefits, the areas with the highest perceived current benefits from AI are data and analytical insights, anti-money laundering, combating fraud, and cybersecurity. The areas of benefit expected to have the largest increase in the next three years are operational efficiency, productivity and cost base. Overall, the increase in expected benefits in the next three years is greater than the increase in average expected risk (21% compared to 9%).
In terms of risks, respondents identified the top 3 risks relating to AI to be data privacy and protection, data quality and data security. The risks expected to increase the most in the next three years are third-party dependencies, model complexity and embedded or "hidden" models. Cybersecurity was ranked as the highest potential systematic risk, and it is expected to be to be so in three years' time.
Results show that data protection and privacy was found to be the greatest regulatory constraint (23% identified it as a large constraint). High regulatory burden was found to be the main type of regulatory constraint (with 33% of firms noting it for data protection and privacy). The top three non-regulatory constraints to AI adoption were found to be safety, security and robustness; insufficient talent/access to skills; and appropriate transparency and explainability.
The results of the survey show that AI is being increasingly adopted by the financial sector, a trend that is expected to continue, and that attitudes towards AI are mostly positive – with firms expecting it to supply more benefits than risk.
Nonetheless, firms and regulators cannot become complacent and must monitor and mitigate the risks found to be a concern through this survey. There should also be a focus on increasing understanding AI technologies across the sector.
