Pinpoint and examine AI use cases now in place or emerging in financial services, from both industry and the regulatory and supervisory side.
The idea
AI use cases split between the industry and the authorities. In financial firms the established uses are fraud detection and anti-money-laundering monitoring, credit scoring and insurance underwriting, and algorithmic trading and portfolio signals. Risk management and stress testing, and customer service chatbots, are also established. Newer generative uses are mostly internal. They include summarising documents, drafting and checking code, and supporting compliance and research, usually with a human reviewing the output before anything reaches a client. On the official side, supervisory technology, or suptech, applies machine learning and language processing to collect and analyse data, flag anomalies, scan disclosures and monitor markets. Regtech is the firm-side automation of compliance. Do not confuse the two labels. Adoption tends to start with lower-materiality internal tasks and move toward client-facing and high-impact decisions as controls mature.
A practice question
Which of the following is an example of AI use by supervisors rather than by financial firms?
- Deploying a chatbot to answer retail customers' account questionsWhy not: Customer service is an industry use case, not a supervisory one.
- Machine learning applied to supervisory returns to flag anomalies - correct
- Generating algorithmic trading signals from market dataWhy not: Trading signals are an industry use case; supervisors do not trade on them.
- Scoring loan applicants using alternative dataWhy not: Credit scoring is carried out by lenders, not supervisors.
Supervisory technology (suptech) uses AI to collect, analyse and monitor data held by authorities, such as anomaly detection in regulatory returns. The other options are firm-side uses.
A harder one
A bank gives staff a large language model tool to summarise regulatory documents and draft and review software code, with every output checked by an employee. How is this use best described?
- An emerging generative AI use for internal productivity and support, with a human in the loop - correct
- A client-facing automated credit decisioning system that approves or declines loan applicationsWhy not: Wrong category: the tool is internal and its outputs are reviewed, not an automated decision on customers.
- A supervisory technology application used by the regulatorWhy not: Wrong user: the bank is the user, so this is a firm-side application, not suptech.
- An algorithmic trading strategyWhy not: Wrong function: summarising documents and drafting code is unrelated to executing trades.
Document summarisation and code assistance with human review are typical early generative AI uses: internal, lower materiality and supervised. They are not client decisions or supervisory tools.
The hardest tier
An insurer uses AI for underwriting and a bank scores consumer loans with alternative data. Which supervisory concern is most specific to these two use cases?
- Correlated selling in stressed markets from many firms using similar trading modelsWhy not: This concern belongs to trading and portfolio use cases, not individual underwriting or credit decisions.
- Biased or unexplainable outcomes for customers and related data protection issues - correct
- Prompt injection against the supervisor's own analytical tools used to read firms' returnsWhy not: A supervisor-side cyber issue that has nothing to do with decisions made about customers.
- Concentration of settlement activity in a single financial market infrastructureWhy not: An infrastructure concentration issue, unrelated to AI underwriting or credit scoring models.
Decisions about individual customers raise fairness, explainability and data-protection questions. Correlated trading and infrastructure concentration relate to other activities.
In the app: 6 practice questions on this outcome, spaced review that brings it back before you forget it.
Start free →More in The Financial Stability Implications of Artificial Intelligence
- Account for the main AI advances made since 2017, separating those pushed by supply from those pulled by demand.
- Weigh up how AI could affect financial stability, concentrating on third-party reliance and provider concentration, market correlations, cyber risk, model risk with data and governance, and others like fraud, disinformation and misalignment.