Industry
AI in insurance
Where AI creates value in insurance — claims, underwriting support, service, fraud signals and document handling — with regulatory constraints, ROI drivers and documented examples.
Miguel Torres, Founder, Merjora · Updated 3 September 2026
In short
In insurance, AI creates the most dependable value in document-heavy and communication-heavy steps: first notification of loss intake, claims triage, evidence extraction, underwriting submission preparation, service triage and proactive status updates. Pricing, entitlement and decline decisions remain human-owned, both because of supervisory expectations and because EU AI Act obligations attach to systems that materially affect access to insurance.
Context
Insurance runs on documents and correspondence, which is precisely the material modern models handle well. That makes the sector unusually well-suited to AI at the operational layer.
It is also one of the most heavily supervised sectors in Europe. The practical consequence is that design begins with the oversight question — who reviews what, and on what evidence — rather than ending with it.
Where the opportunities are
FNOL intake and claims triage
Turning notifications from email, forms and calls into structured claims, then segmenting by complexity so simple claims take a fast path and handlers concentrate on complex ones.
Complexity: Moderate
Evidence and document extraction
Extracting fields from claim forms, invoices, reports and correspondence with confidence-based routing to review. Usually the highest-volume, most measurable opportunity in the business.
Complexity: Moderate
Underwriting submission preparation
Structuring broker submissions, checking completeness against appetite and surfacing missing information before an underwriter opens the file. Support, not decisioning.
Complexity: Moderate
Customer service triage and drafting
Classifying and routing inbound contacts, retrieving policy wording and drafting replies grounded in the policy and case record for adviser approval.
Complexity: Low
Fraud and anomaly signals
Surfacing patterns for investigator review. Value should be claimed only against a measured control comparison, and fairness testing is mandatory.
Complexity: High
Renewal and retention communication
Prioritising renewal outreach and drafting the communication, with pricing and terms unchanged by the model.
Complexity: Low
Constraints that shape delivery
- EU AI Act classification must be recorded per use case; systems affecting access to insurance carry heavier obligations
- Special-category data appears routinely in health, injury and motor claims
- Explainability is a complaint-handling and supervisory requirement, not a nice-to-have
- Legacy policy administration systems often limit integration options and drive cost
- Fairness testing is required wherever outputs influence customer treatment
ROI considerations
- Handler minutes per claim × claim volume is the primary value driver
- Cycle-time reduction on the simple claim segment usually beats accuracy gains in value terms
- Include governance, documentation and fairness testing in first-year cost
- Model straight-through rates conservatively; a 15-point error moves the case materially
Documented examples
Publicly documented industry examples. These companies are not Merjora clients.
- Aviva →
Publicly documented AI use in claims and service operations.
- Mutua Madrileña →
Publicly documented AI and automation programme.
Frequently asked questions
- What are the best AI use cases in insurance?
- Claims intake and triage, document extraction, service triage and underwriting submission preparation. They are high-volume, text-based and measurable, and they keep a human at the decision point.
- Can insurers use AI in pricing and underwriting decisions?
- AI is widely used to support them, but decisions affecting access or price attract significant regulatory obligations around oversight, documentation and fairness. Treat those as human-owned decisions with AI support.
- Where do insurance AI projects usually fail?
- On integration with policy administration systems and on unmeasured baselines. Both are solvable, but neither is visible in a vendor demo.
Where would AI pay off in your insurance business?
Merjora maps your workflows, scores each opportunity and quantifies the value range — with the constraints named.
Discover your AI opportunitiesRelated reading
Editorial standard. Merjora publishes analysis, frameworks and publicly documented examples. We do not publish invented statistics, unattributed benchmarks or unverified customer stories.