Key points
- Most European insurers now use generative AI somewhere, but mostly in back-office work and mostly still at proof-of-concept stage.
- In Norway, Finanstilsynet surveyed 31 insurers in mid-2024. Twelve used little or no advanced technology and eight used it extensively.
- AI helps most where the work is repetitive and document-heavy: extracting data, searching, drafting and supporting staff.
- It helps least where a decision must be explained to a customer, where the data is poor, or where the process itself is unclear.
- Finanstilsynet expects the governance framework to be in place before advanced methods are used, people to oversee the results, and prices to be explainable to the customer.
- For a smaller insurer, the safest start is one well-understood process, a clear owner and a person who stays responsible for the decision.
Picture a mid-sized Norwegian non-life insurer. The board has asked for an AI plan by the next meeting. The claims team wants help with the inbox. IT is still running a core system that two people fully understand. A vendor has offered a chatbot. Everyone agrees AI matters. Nobody agrees where to start.
That situation is common, and it is a reasonable place to be. The question is not whether to use AI. It is which work to give it first, and which work to leave alone for now.
How far have insurers actually got with AI?
Further than the headlines about pilots suggest, but not as far as the vendor brochures claim.
Across Europe, EIOPA surveyed 347 insurers in 25 countries on generative AI and published the results in February 2026. Nearly two-thirds were already using the technology, but most were still at proof-of-concept stage. Almost half (49%) had a dedicated AI policy, up from about a quarter in 2023.
Norway shows a wider spread. Finanstilsynet surveyed 31 insurers in mid-2024. The survey covered advanced technology in general, including machine learning, not only generative AI. It found:
- 12 used little or no advanced technology
- about five used it only for simple efficiency and automation tasks
- eight used it extensively, across all or several parts of the value chain
- the remaining firms fell somewhere in between
The survey was qualitative, so these are counts, not a market statistic. The data is also more than two years old, and the picture has likely moved since. Still, the pattern is clear: a handful of insurers are well ahead, and a large group has barely started.
Where does AI help today?
EIOPA's survey shows where most of the real use is. Of the generative AI use cases reported, 64% were back-end productivity tools. Examples include extracting data from invoices, audio and reports, generating content, and assistants for coding and underwriting staff. Separately, 36% of respondents reported developing customer-facing applications such as chatbots and voice assistants. Most of those were still being tested.
Norway shows a similar mix, but with more customer-facing use than the European figures suggest. Insurers reported using advanced technology for internal work such as coding, writing, claims handling, HR, and classifying and searching documents. Chatbots in customer communication were also common, and many firms used the technology in distribution, advice and sales. Several used it for fraud detection, and seven used it in pricing.
Across both surveys, AI earns its place where the work has three features:
- It is repetitive. The same kind of document or question arrives many times a day.
- It is text-heavy. Someone has to read, sort or summarise before they can act.
- A person still checks the result. The AI prepares the work and an employee decides.
Claims intake is a good example. A claim arrives with photos, receipts and an email. Reading and sorting that material is slow. Deciding what to pay is a judgement. AI can do much of the first part well. The second part should stay with a claims handler.
Where does AI not help yet?
AI disappoints in three situations.
When the decision has to be explained. Finanstilsynet expects insurers to keep pricing transparent and to be able to explain a price to the customer. It also expects insurers not to develop a practice of maximising the price based on how much a customer is willing to pay. A model that cannot show why it reached a result is a poor fit for any decision that a customer, a complaints board or a supervisor may question.
When the data is poor. AI does not fix missing or inconsistent data. It hides the gaps behind a confident answer. EIOPA's respondents named inaccurate outputs, or hallucinations, as the top risk of generative AI. In Norway, the smaller insurers that used little advanced technology gave three main reasons: lack of competence, not seeing enough business value, and too little data or data of poor quality.
When the process is unclear. If two experienced employees would handle the same case differently, an AI system will not settle the question. It will copy the inconsistency faster. The process has to be agreed before it can be automated.
This is why fully automated, customer-facing AI is still rare. The technology can produce an answer. The insurer still has to stand behind it.
What does Finanstilsynet expect?
Finanstilsynet's 2024 survey set out its expectations clearly. The most important is simple: the framework must be in place before advanced methods are used ("Rammeverket må vere etablert før slike metodar blir tekne i bruk"). The regulator also expects insurers to ensure sufficient human oversight of systems, processes and results, and to carry out regular audits to check that systems work as intended.
In our reading, that means an insurer should be able to answer four questions before an AI tool touches real cases:
- Who owns the tool and the decisions it supports?
- What data does it use, and is that data allowed for this purpose?
- How are its results checked, and how often?
- How is a decision explained to the customer if they ask?
The survey also found that very few insurers had carried out specific controls of their advanced technology. Finanstilsynet noted a further risk: more detailed data and more advanced analysis can, in practice, shut some customers or groups out of insurance altogether.
This is a description of the regulator's stated expectations, not legal advice. Each insurer should confirm its own obligations with a qualified adviser.
What new risks come with AI?
Finanstilsynet's 2026 risk and vulnerability analysis covers the financial sector as a whole. Three of its themes matter directly for insurers.
Supplier dependence. The regulator calls supplier and value-chain risk one of the most critical vulnerabilities in the financial sector. EIOPA found the same at European level: insurers rely heavily on third-party providers for generative AI. An AI project is also a supplier decision.
Shadow AI. Staff who use generative AI tools outside approved systems can send company data into services that store or reuse it. A clear policy and an approved tool cost less than finding out later.
AI-enabled fraud. Finanstilsynet notes that fraudsters across the financial sector increasingly use AI to make social-engineering attacks more efficient and more tailored to their targets.
The analysis also reports that some financial firms see risk in core systems built on older or in-house solutions, which can create challenges around knowledge and competence. For many smaller insurers, that is the real constraint on AI. The limit is not the model, but the system it has to connect to.
How should a smaller insurer start?
At Hundred Solutions we work by a simple rule: solve first, automate later. Applied to insurance, it gives a short plan.
- Pick one process, not a strategy. Choose work that is repetitive, text-heavy and easy to check, such as sorting claims documents or answering routine broker questions.
- Write the process down as it really works. Note where people disagree, and settle that first.
- Check the data. Find out where the inputs come from, how complete they are, and whether you may use them for this purpose.
- Put the framework in place. Name an owner, decide how results are checked, and decide how a customer gets an explanation.
- Keep a person responsible for the decision. Let AI prepare the case and let an employee decide. This is also what Finanstilsynet's expectation of human oversight asks for.
- Measure, then widen. Compare time and quality before and after. Only then move to the next process.
This is slower than buying a platform and switching it on. It is also how a small team avoids a pilot that never reaches production.
Where AI fits in an insurer: a quick view
| Area | Where AI helps now | Where to be careful |
|---|---|---|
| Claims | Reading, sorting and summarising documents and photos | Deciding what to pay |
| Customer service | Drafting replies for staff to check; internal search | Unsupervised chatbots giving advice |
| Pricing | Analysis that supports actuaries | Prices that cannot be explained to the customer |
| Fraud | Flagging patterns for an investigator | Treating a flag as proof |
| IT and operations | Coding support, documentation, testing | Connecting AI to old core systems without a plan |
| Compliance | Finding and summarising internal documents | Replacing the control function's judgement |
Thinking about where AI fits in your claims or customer-service work? Hundred Solutions helps insurers map one process, check the data and put governance in place before anything is automated. Talk to us
Sources
- EIOPA, press release on its survey of generative AI use among European insurers (2 February 2026): https://www.eiopa.europa.eu/eiopa-survey-generative-ai-shows-swift-cautious-adoption-among-europes-insurers-2026-02-02_en
- Finanstilsynet, "Kartlegging av forsikringsføretaka sin bruk av avansert teknologi" (2024): https://www.finanstilsynet.no/publikasjoner-og-analyser/andre-rapporter/forsikringsforetakenes-bruk-av-avansert-teknologi/kartlegging/kartlegging-av-forsikringsforetakenes-bruk-av-avansert-teknologi/
- Finanstilsynet, "Risiko- og sårbarhetsanalyse 2026" (April 2026): https://www.finanstilsynet.no/publikasjoner-og-analyser/risiko--og-sarbarhetsanalyse/risiko--og-sarbarhetsanalyse-2026/risiko--og-sarbarhetsanalyse-2026/risiko--og-sarbarhetsanalyse-2026
