Sector
AI leadership for insurance
Last updated
Justas Butkus is an AI expert and fractional AI officer based in Vilnius, Lithuania, working with mid-market companies and scale-ups across the UK, EU and US at two to four days a month. He builds and operates the production AI systems he advises on, rather than only advising on them.
Short answer
Insurance runs on document-heavy, repetitive processes with qualified people in the middle of them – which is exactly the shape AI suits. The constraint is that pricing and underwriting decisions carry regulatory obligations that operational automation does not.
Where does AI actually pay for itself in insurance?
The processes worth automating in any sector share three properties: they repeat constantly, they consume qualified people, and nobody enjoys them. In this sector that usually means:
- First notification of loss – intake, classification and routing at volume.
- Claims document handling – extraction, completeness checking and chasing missing items.
- Renewal and lapse contact – high volume, time-sensitive, usually under-resourced.
- Broker and customer query triage – the same questions, answered repeatedly.
- Policy wording retrieval – staff searching documents for answers that already exist.
None of these are exciting. That is the point – the value comes from volume, not sophistication, and the exciting use case is usually the one that fails.
What makes this sector different
- Risk pricing for life and health insurance is named in Annex III of the EU AI Act as high risk. Claims triage and customer contact generally are not – but the classification has to be done deliberately.
- Automated decisions affecting customers carry data protection obligations independent of the AI Act.
- If AI speaks to customers, Article 50 transparency applies to how the system was designed, and that duty sits with whoever provides it.
The pattern I see most often is a firm that has automated something genuinely low-risk, and cannot demonstrate that it is low-risk because nobody documented the classification.
How an engagement here typically runs
- Inventory what is already runningIncluding the unapproved tools. In regulated environments this is usually the most uncomfortable and most valuable step.
- Map the repeating processes against costFrequency multiplied by fully loaded cost per occurrence. The ranking usually surprises people.
- Establish governance proportionate to the riskWho approves what, what happens when a system is wrong, and how a decision is reconstructed months later.
- Ship the unglamorous one firstProve the pipeline end to end before attempting the ambitious thing.
Frequently asked questions
Is AI in insurance high risk under the EU AI Act?
Some of it. Risk assessment and pricing in life and health insurance appear in Annex III. Claims intake, document handling and customer contact usually do not, but the classification should be documented rather than assumed.
What is the highest-value first project in insurance?
Usually document handling in claims: high volume, highly repetitive, currently consuming qualified assessors, and low regulatory exposure compared with anything touching pricing.
Can AI handle customer contact in insurance?
Technically yes, and it is deployed widely. The obligations are that people must be informed they are dealing with an AI system, and that there is a real route to a human when the case needs one.
If this describes your operation
The useful first conversation is about one specific process rather than about AI in general.