FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Aipedia: Actuaries
Actuaries can use machine learning to set mortality and longevity assumptions, but they must understand the model, check its data for bias and drift, and explain the uncertainty clearly. Small changes in life expectancy move liabilities for pension schemes, annuity writers and life insurers a long way.
Part of: AI in Actuaries
In short
Machine learning models can combine postcode, socio-economic, medical and lifestyle data to predict mortality for individuals or schemes. They may be used in pension valuations, bulk annuity pricing and life underwriting. TAS 100 Principle 5 applies, and the FRC Models guidance expects bias testing by sub-group and monitoring of drift in self-updating models. Health data is special category data, and automated underwriting decisions engage the new UK GDPR safeguards. A mis-estimated assumption can produce a large claim, which actuarial PI usually responds to subject to the policy terms, but the assumption is a judgement and must be shown to be a reasonable one.
Mortality work has always been data-driven, but machine learning lets actuaries model many more rating factors at once. Uses include member-level longevity models for defined benefit schemes, postcode and socio-economic models for bulk annuity pricing, medical underwriting models for individual annuities and protection products, and analysis of how mortality improvements differ between groups.
These models can be sharper than traditional base tables plus adjustments. They can also be harder to explain, more sensitive to the period of data they were trained on, and more likely to pick up patterns that reflect how data was collected rather than how long people live. A model that fits recent data well can still give poor answers about the next thirty years.
Health data is special category data. Where a model drives a solely automated underwriting decision with significant effect, the safeguards in UK GDPR Articles 22A to 22D, inserted by section 80 of the Data (Use and Access) Act 2025, apply.
Illustrative scenario: a consultancy builds a longevity model for a scheme preparing for a buy-in. It uses postcode and pension amount as key factors and is trained on a large pooled data set. The scheme's members are concentrated in a few areas and occupations that are thinly represented in that data, so the model's estimates for them are less reliable than the report suggests. The trustees rely on the result in negotiating the transaction and later question whether the uncertainty was adequately explained.
Other risks: a model trained on a period of unusual mortality experience, factors that act as proxies for protected characteristics, and improvement assumptions that the model cannot inform at all. See bias and fairness testing and, for the data feeding the model, AI for pension scheme data.
A claim alleging that a negligent mortality or longevity assumption caused loss would usually fall within actuarial PI, subject to the policy terms. Assumption setting involves judgement, and claims commonly turn on whether the assumption was within the range a competent actuary could reasonably have chosen at the time, and whether its limits and uncertainty were properly communicated. A clear explanation of uncertainty in the report is therefore a defence as well as a professional duty. Because the sums can be large, check your limit against the size of the schemes and books you advise on; see PI insurance for actuaries.
Apex arranges PI for life and pensions actuaries; a broker can check that your declared work, including longevity modelling for transactions, matches the cover. See the AI in actuarial work overview.
It can inform it, but the actuary sets the assumption and must be able to justify it. TAS 100 requires understanding of the model and its limitations, and the communication principles require the uncertainty to be explained. Comparing machine learning results with standard tables before relying on them is a sensible check.
Yes. Health data is special category data under UK GDPR, so a lawful basis and an additional condition are needed, and its use should be limited to the purpose. Where automated underwriting decisions have significant effects, the safeguards added by the Data (Use and Access) Act 2025 also apply.
That a mis-estimated assumption affects a large liability or a transaction price, and the client says the uncertainty was not explained. Claims usually turn on whether the assumption was reasonable at the time and properly communicated, so documentation and sensitivity testing matter.
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.
