Half the reading time is not half the judgement

In November 2025 Aviva launched what it called the first AI underwriting summarisation tool in the UK protection market, for individual life insurance. In March 2026 it extended it to critical illness cover, meaning it can now summarise “most individual protection applications where a medical report is required” (Aviva).

The tool uses generative AI to turn “lengthy and complex reports” into short summaries “containing the relevant information for underwriters to review.” Aviva says this has cut the time underwriters spend reviewing each case “by around half.” At the start of October 2026 it added income protection, completing the rollout across its individual protection range (Insurance Edge).

“By bringing AI summarisation to the protection market, we’re helping more customers get the protection they need more quickly.”Robert Morrison, chief underwriting officer, Aviva

This is the sensible end of AI underwriting

Credit where it is due. The tool sits at the front of the journey and the human still makes the decision. In the FCA’s Mills Review vocabulary, that is “Consultant” mode, where the regulator judges existing rules to work well. Applicants waiting weeks for a life policy because a GP report sits in a queue is a real harm. Halving that wait is a real benefit.

Our view: a summary is an editor

Every summary is a decision about what to leave out. When an underwriter reads a 30-page medical report, the odd phrase on page 22 can change a decision. When they read a summary, they only see what the model judged relevant.

Aviva says it has seen “high levels of accuracy in extensive testing.” We do not doubt that. But accuracy of what is included is a different test from completeness of what is excluded. The second is harder to measure and matters more on the cases that end in a loading or a decline.

The Mills Review makes a related point: as AI shapes how products are delivered, it may become “harder to ensure outcomes remain appropriate and fair across different consumer groups, particularly where models may embed historic patterns of bias.”

What we would like to see

• An omission audit. On a sample of cases, have an underwriter read the full report and record what the summary missed. Publish the rate.
• Full report on adverse decisions. Where the outcome is a decline or a loading, the underwriter reads the source, not just the summary.
• Outcome checks by group. Compare acceptance and loading rates before and after, by age, sex and condition type.

Aviva says its capability “also extends to post-application auditing”, so some of this may already exist. If it does, publishing it would turn an efficiency story into a trust story. Half the reading time is a good result. It just is not the same thing as half the judgement.