AI adoption in insurance: which use case to start from

In short. In an insurance company the choice of a first use case carries a constraint that doesn't exist elsewhere: the highest-value process, life and health pricing, is also the one classified as high risk under Annex III. The cases that can reach production immediately sit in claims, network support and back office. What separates those who reach production from those stuck at pilot is the system's owner, not the technology.

In insurers, the AI conversation almost always starts from pricing, because that's where results are measured. It's also the point not to start from, for reasons about sequencing rather than merit.

Two measures to keep apart

Assessing an insurance use case takes two distinct measures, and overlapping them is what stretches projects by months.

The first is economic value, concentrated in pricing, claims handling and fraud detection. The second is regulatory friction, meaning the governance work required before production.

Life and health pricing sits high on both. Claims document handling sits mid-range on value and low on friction, which is why almost every successful path starts there.

The map of candidate use cases

Use case	Value	Regulatory friction	Suitable as first
Reading and classifying claims documents	High in efficiency	Low	Yes
Network support on products and terms	Medium	Low, Article 50	Yes
Search across regulation and internal circulars	Medium, in time saved	Low	Yes
Claims fraud detection	High	Medium, depends on effect on the individual	As a second step
Automated settlement of simple claims	High	Medium-high, affects entitlement to a benefit	Only with mature controls
Life and health pricing	The highest	The highest, Annex III	No, not as a first

The row worth attention is the fifth. Automated settlement of low-value claims looks low risk because the amounts are small, and the criterion isn't the amount: it's whether the system affects recognition of entitlement to a benefit.

The advantage insurers don't use

Insurers hold a model governance framework few other organisations possess: independent validation, model documentation, second-line controls, an actuarial function that validates.

That framework covers much of what the AI Act asks for high-risk systems, and in most cases it goes unused. The reason is organisational: new AI systems arrive from different functions, often digital or operations, and never pass through the validation process because nobody classified them as models.

Extending the perimeter of existing model governance to AI systems costs less than building a parallel framework, and produces a more solid result, because that process has already been tested by a supervisory authority.

What must exist before calling it a pilot

Three elements need defining in writing before starting, and the first is the measured baseline of the current process: how many files per day, at what average time, with how much rework. Without that number the final assessment becomes an impression.

The acceptance criteria, set before seeing results, because written afterwards they accommodate any outcome.

The exit plan, meaning what happens if it works. In an insurer, moving to production requires integrations with policy and claims systems that were rarely estimated at pilot stage.

The owner, not the technology

The error blocking most projects isn't the choice of case, it's who answers for it. An AI system in production needs a named person accountable for its outcomes, and in an insurer that person sits in the technical function or in claims, not in IT.

When ownership stays with IT the system gets maintained rather than governed: nobody decides when it's out of scope, nobody detects drift, and the control exists only in procedures. It's also the condition that weakens the company's position if that system ends up in a dispute.

Where to start

The sequence begins with the AI system inventory, built while also looking at models already validated, because the regulation's definition is broader than the one insurers use internally.

The selection rules are in AI adoption: how to choose your first use case, and the cost model from pilot to production in AI project budgeting.

The sector's regulatory picture is on the AI governance for insurance companies page.

To assess which use cases make sense in your company, you can book an assessment meeting.