FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Aipedia: Wealth managers and discretionary fund managers
Wealth managers can use machine-learning signals and AI research in investment decisions, but they must understand and test the models they rely on. Whether a loss becomes a PI claim depends on showing negligence, not simply poor performance.
Part of: AI in Wealth managers and discretionary fund managers
In short
Machine-learning signals and AI research summaries may inform discretionary investment decisions. The duties are unchanged: COCON requires due skill, care and diligence, COBS 9A requires suitable decisions, and the FCA's October 2026 good-practice page expects owners, auditable records and challenge. The main risks are overfitted models, data errors, regime changes and creeping reliance. PI covers negligence, not market loss, so a client must usually show the firm fell below a reasonable standard. A tested, documented and challenged model leaves the firm far better placed.
Larger discretionary firms build machine-learning models that generate signals on asset classes, sectors or individual securities, and use generative AI to summarise company reports, earnings calls and broker research. Smaller firms more often buy signals or AI research summaries from third parties. Either way, the output feeds the investment committee's view or the stock selection for discretionary portfolios.
The Bank of England and FCA 2024 survey found 46% of financial firms reported only a partial understanding of the AI they use. For investment signals, partial understanding is a real problem: a manager who cannot explain why a model favours an asset cannot judge when it is likely to fail.
Model risk takes several forms. A model may be overfitted to historical data and look excellent in backtests but fail in live markets. Data feeds may change format or contain errors. A regime change, such as a shift in interest rates, can invalidate relationships the model relied on. Generative AI research summaries can misstate figures or omit material risks. And reliance creeps: a signal used as one input gradually becomes the decision.
Third-party signals add a further layer. The firm may not see the model's design or training data, yet SYSC 8.1.6R keeps it fully responsible for outsourced functions. Due diligence on how a vendor builds, tests and changes its signals, and the right to be told of material changes, matter as much as the signal's past performance.
Illustrative scenario: a firm's in-house model, built on several years of backtests, recommends a concentrated position in a sector it has scored as low risk. The investment committee adopts it across its higher-risk mandates without stress testing beyond the backtest period. Conditions change and the sector falls sharply. Clients complain that the firm took a concentrated bet based on a model it did not properly test or explain.
This is the central PI question. PI covers liability for negligent professional services, not investment losses that happen when a reasonable decision turns out badly. A client must usually show that the firm fell below the standard of a reasonably competent discretionary manager, or breached the mandate, and that this caused the loss. The standard is discussed in the Bolam test for financial advisers.
AI changes the evidence rather than the test. A firm that tested the model, documented its limits, applied concentration limits and recorded the committee's challenge is in a stronger position. A firm that followed a black box it could not explain may find it harder to show it acted with reasonable care. Where a claim is made, PI usually responds subject to the wording; see does PI cover wealth managers' AI errors and algorithmic decisions and PI.
Apex arranges PI for wealth managers and discretionary fund managers. A broker can review how your policy treats claims involving model-driven investment decisions and what insurers will ask about your model governance. See AI in wealth management.
Only if the client alleges the firm was negligent or breached the mandate, for example by relying on an untested model. Ordinary market losses from reasonable decisions are not negligence. PI usually responds to negligence claims, subject to the wording.
No specific rule requires full disclosure, but the FCA's good-practice page encourages telling consumers how AI is used and its limits. Any description must be accurate and must not overstate the model's role or reliability.
Backtests alone are weak evidence. Overfitting can make a model look reliable in history and fail live. Out-of-sample testing, stress testing and limits on position size give a firm a much stronger record of reasonable care.
Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.
Start a PI quote Request a callbackApex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.
