Aipedia: Actuaries

AI in experience and claims analytics

Aipedia, by the Apex broking team · Last reviewed 2026-10-27

Actuaries can use AI to analyse lapse, claims and fraud experience, but the assumptions drawn from it remain the actuary's responsibility and must rest on sound, lawfully used data. Experience analysis feeds pricing, reserving and capital, so one bad finding can spread a long way.

Part of: AI in Actuaries

In short

Machine learning can find patterns in lapse behaviour, claims development and suspected fraud that traditional experience studies miss. The same techniques can overfit, mistake data artefacts for real effects, and reproduce bias. TAS 100 covers data, assumptions and models, and the IFoA guide asks for bias to be identified and responsibility to be agreed in multidisciplinary teams. Claims and policyholder data is personal data, often including health, and automated decisions such as fraud flags engage UK GDPR safeguards. Actuarial PI usually responds to claims that wrong assumptions caused loss, subject to the policy terms, but data protection penalties are a separate matter.

How actuaries use AI for experience analysis

Experience analysis compares what happened with what was assumed. AI tools can help with predictive lapse and persistency models, claim-level analysis of development and severity, text mining of claims notes and adjuster reports, clustering of policyholders by behaviour, and fraud detection scores. Results feed into pricing, reserving, product design and capital assumptions.

The work is often shared between actuaries, data scientists and claims or operations teams. That brings skills together but can blur who decides what the analysis means and who signs off the resulting assumption. It also means data moves between systems and teams, which raises data protection questions.

The duties that apply

TAS 100 P2.3, P3 and P4: data and assumptions; P5.1 to P5.4: models, including bias (P5.3).
The FRC Models guidance says algorithm and data bias are "especially significant" for opaque models and that outputs should be tested by sub-group (5.18 to 5.20).
The IFoA data science and AI guide asks for bias to be identified, communicated and mitigated (2.17), for clear governance and agreed responsibility in multidisciplinary teams (2.11, 2.29), and for attention to consent and the purpose for which data was collected (2.10 to 2.11, 2.20).
Claims data often includes health information, which is special category data. Where analytics drive solely automated significant decisions, such as declining a claim on a fraud score, UK GDPR Articles 22A to 22D apply under section 80 of the Data (Use and Access) Act 2025. The ICO AI guidance is relevant.

What can go wrong

Illustrative scenario: an actuarial team uses a machine learning model to analyse lapse experience on a protection product. The model finds that lapses have fallen sharply and the pricing basis is updated. The fall was driven by a temporary change in how lapses were recorded during a systems migration, which the model treated as genuine behaviour. Profitability suffers, and the insurer asks why the data change was not identified.

In a second illustration, a fraud model trained on past investigations flags claims from certain groups more often because of how cases were historically selected for investigation. Complaints and a regulatory review follow. See bias and fairness testing.

Responsibility and how PI usually responds

Where an actuary or consultancy recommends an assumption based on AI analytics, a claim that the assumption was negligently derived would usually fall within actuarial PI, subject to the policy terms. Claims commonly turn on whether reasonable data checks were done and whether limitations were explained. Where a data scientist built the model and the actuary signed off the assumption, agreed responsibility matters for both the professional position and the claim. Regulatory fines imposed on an insurer client for data protection or conduct failings are generally separate and may not be insurable; see GDPR and data breach cover within PI. Assumptions then flow into pricing models and reserving, so the loss may surface elsewhere. See PI for actuaries.

Practical controls

Reconcile AI findings with a simple traditional experience study before changing assumptions.
Check for changes in data recording, systems or processes over the period analysed.
Test results by sub-group and look for proxies for protected characteristics.
Agree in writing who owns the model, who interprets it and who signs off the assumption.
Confirm the lawful basis and purpose for using claims and health data; see client data confidentiality.
Keep a human in fraud and claims decisions with significant effects.

How Apex helps

Apex arranges PI for actuaries working on experience and claims analytics; a broker can review whether cyber or data protection cover is needed alongside PI. See the AI in actuarial work overview.

Sources

Frequently asked

Can I set an assumption directly from a machine learning model?

You can use the model to inform it, but TAS 100 expects you to understand the model, check the data and communicate limitations. Reconciling the result with a simpler analysis and checking for data recording changes are sensible steps before relying on it.

Do fraud models used with claims data raise data protection issues?

Yes. Claims data is personal data and often includes health information. If a fraud model drives solely automated decisions with significant effects, the safeguards in UK GDPR Articles 22A to 22D apply, and bias in the model can lead to complaints and regulatory scrutiny.

Who is responsible when a data scientist builds the model?

It depends on what was agreed. The IFoA guide stresses clear governance and agreed responsibility in multidisciplinary teams. An actuary who signs off an assumption based on the model remains responsible for that professional judgement under the Actuaries' Code and TAS 100.

Check your PI covers how you use AI

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 callback

Apex 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.