Insurers process a constant stream of documents: claim notifications by email, photos, repair estimates, medical invoices, address changes. The rules are set in tariffs and policy terms, but the inputs arrive unstructured. Agents read, match and prepare; the benefit decision in disputed cases stays with people.
Industry 02/103 example agentsAutonomy level L3–L4
Example agents
Three agents that take over work here
Examples from typical workflows, not client references. Which agents pay off for you is shown by assessing your own processes.
01
The claims intake agent
Claims · first notice of loss
Takes claim notifications from email, portal and call notes, identifies line and policy, opens the claim, requests missing documents such as photos or a police report and routes the case with a summary to the right team.
Claim creation and document requests at L4; coverage checks in unclear cases stay at L3.
Systems
Policy admin systemClaims systemEmail · portalDMS
Guardrails
No benefit promise, no payout, customer communication labelled as AI, health data only for authorised roles.
Human in the loop
Claims handlers take every case with a coverage question or fraud indicator; the agent supplies its reasoning.
What to measure
Time to claim creation · complete files at first contact · misrouted cases
02
The invoice review agent
Claims · motor and property
Checks garage and contractor invoices against the estimate, agreements and reference rates, flags inflated items, drafts the reasoning for a reduction and prepares payment for unremarkable invoices.
Release limits per line, no reduction without reasoning and source, payment only to registered accounts.
Human in the loop
Assessors and claims handlers confirm reductions and handle objections.
What to measure
Review time per invoice · savings from reductions · successful objections
03
The policy service agent
Policy admin · customer service
Handles policy changes such as a new address, payment frequency or vehicle from email and portal, has the rating engine recalculate the premium, issues the endorsement and confirms to the customer.
No bank detail changes without confirmation, no acceptance of new risks, premiums come only from the rating engine, never from the model.
Human in the loop
Underwriting handles anything that needs risk assessment; complaints go straight to people.
What to measure
Turnaround per change · straight-through rate · complaints after changes
Framework
What to factor in for this industry
AI Act: AI used for risk assessment and pricing in life and health insurance is high-risk; the obligations apply from 2 December 2027.
DORA: Insurers have also been subject to the requirements on ICT risk management and ICT third-party oversight since 17 January 2025.
Customer contact: People writing to an agent must be able to tell (Art. 50 AI Act). Health data is a special category under Art. 9 GDPR and needs strict access rules.
Note, not legal advice: whether a specific use falls under these rules is for you and your legal counsel to decide. We provide the technical facts for it, from the agent profile to the audit trail.
Claim notifications arrive in large numbers and very mixed quality. An agent that produces complete files relieves claims handling at once, without deciding on benefits itself.
