Attention on AI in professional services has focused on generative tools, but the older and more embedded technology is algorithmic decision-making. Credit scoring engines, suitability calculators used by financial advisers, tax-planning optimisers, portfolio-construction models, medical triage tools and conveyancing risk scorers already sit inside the daily work of many regulated professionals. Where a professional relies on algorithmic output, the professional indemnity implications deserve careful thought.
An algorithmic decision system is any set of rules or a statistical model that produces an output intended to guide, support or determine a decision. In professional practice this includes credit-risk scoring, suitability engines that map a client's attitude to risk against a model portfolio, tax-planning tools that select allowances and reliefs, algorithmic conveyancing checks and clinical triage tools that grade urgency. The systems are not new. What has changed is the depth of reliance, the opacity of some models and the speed at which drift and bias can creep into the output.
The standard of care has not moved. The Bolam test (Bolam v Friern Hospital Management Committee [1957] 1 WLR 583) still asks whether the professional acted in accordance with a practice accepted as proper by a responsible body of practitioners in the same field. That duty extends to the selection, use and monitoring of any tool the professional relies on. A solicitor using an algorithmic search product, an accountant using a tax-planning engine, an IFA using a portfolio-construction model — each remains responsible for the reasonableness of the output they act on.
The SRA's guidance on technology in legal services, the ICAEW Technical Release on algorithms and equivalent guidance from other bodies reinforce the same point: the tool does not carry the duty, the practitioner does.
Where a decision is solely automated and produces legal or similarly significant effects on an individual, Article 22 of the UK GDPR gives the data subject the right not to be subject to that decision, subject to limited exceptions, and the right to obtain human intervention, to express a point of view and to contest the outcome. The Data Protection Act 2018 supplements this in domestic law. The ICO's guidance on automated decision-making sets out how firms should identify solely-automated decisions, provide meaningful information about the logic involved and offer the human-review route. For an adviser this typically means building a review step into any process where the algorithm's output would otherwise be determinative.
Algorithms are not static. Models drift as the input data drifts. Edge cases — the client whose circumstances sit outside the training distribution — are where risk concentrates. Bias, whether in training data or feature choice, can produce systematically poor outcomes for particular groups. A reasonable professional is expected to understand the tool's limits, to test output against professional judgement and to review performance periodically. Blind reliance on a black-box output is unlikely to meet the standard of a responsible body of practitioners.
A standard PI policy responds to claims arising from negligent acts, errors or omissions in the provision of professional services. Where a claim arises because the professional failed to check, question or override an algorithmic output that a competent practitioner would have flagged, the claim is a conventional negligence claim and cover typically engages in the ordinary way. Cyber elements — where harm arises from a security failure in the algorithm's data — may fall to a cyber policy rather than the PI wording.
An IFA in 2024 uses a portfolio-construction algorithm to generate a recommended allocation for a retired client with an unusual income need. The algorithm outputs a growth-tilted portfolio that would suit a longer horizon. The IFA does not scrutinise the output against the client's actual profile and issues the recommendation. Markets fall, the client crystallises a loss and complains. The IFA's PI insurer treats the matter as a conventional negligence claim: the duty was to check the algorithm's output for suitability, and that duty was not discharged. Separately, the IFA may have contractual recovery against the algorithm vendor under warranty or indemnity terms, but that is a commercial question, not a substitute for the primary PI response.
The building blocks are familiar: keep an audit trail of what the algorithm returned and what the professional did with it; keep a human in the loop on any decision that would otherwise be solely automated and materially affect a client; test the algorithm on known scenarios at intervals; document the professional's own review of edge cases; and review the vendor contract for warranty scope, indemnity and data-security terms. Where a firm depends heavily on an algorithmic service, that dependency should be recorded on the risk register and reflected in PI renewal disclosure.
See the companion entry on AI-generated advice and professional negligence for the generative-AI counterpart. Sector guidance for the professions most likely to rely on algorithmic tools is set out in the pillar pages for independent financial advisers, accountants, solicitors and surveyors.
Apex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Firm reference number 724952. This entry is general information, not advice on any particular policy.