Map your existing authority levels to depth before anyone trains. Insurers already limit what each handler may settle; agents should fit that structure. Certify the claims handlers and underwriters who will approve the agent's held work first, train the people who set authority next, and let no agent settle a claim until both sides have evidence.
Why do authority levels come first?
Insurers run on delegated authority. A junior claims handler may settle up to one amount, a senior up to a higher one, a claims manager beyond that. Underwriters work the same way with referral limits. These limits are already a control, and the people inside them already understand them.
An agent should slot into that structure rather than sit beside it. Drafting a claim decision for a handler is depth 2. Settling small, clear claims within the insurer's own limits is depth 3 at most for the decision, and paying out is depth 4 because money leaves. If training starts before this mapping, handlers learn an abstract vocabulary they cannot connect to their authority. If it starts after, they learn the agent's place in a structure they already respect.
What order should training follow?
- Claims and underwriting leadership. Map authority levels to depth, decide which capabilities the agent will target first, and learn what the record will show.
- Claims handlers on the first queue. The reviewer path: reading held claims, checking provenance, declining with reasons, red lines. A practicum during shadow.
- Team leaders and technical claims managers. Calibration across teams, drift, what happens when the agent's certificate lapses and it drops to depth 1 the same day.
- Underwriters, if the agent will prepare referrals or quotes.
- Builders. Whoever designs and certifies the agents, after they have spent time on the held queue.
- Complaints, fraud and audit. How to reconstruct a decision and how to spot patterns across held and refused cases.
The argument for putting reviewers this early, even before procurement is finished, is made in train the reviewers before you buy the agent.
What must claims reviewers learn?
Evidence by class. A claimant's photos and description are declared. A loss adjuster's report from a verified firm is signed. Policy terms from the policy administration system are sealed. A repair invoice emailed by a garage is external and cannot authorise a payment. Handlers who sort evidence this way spot inflated and fabricated claims faster.
The refusals that protect customers. The insurer's limits will refuse some actions outright, such as declining a claim on grounds the agent cannot verify. Reviewers must know that a refused decline is the system working, and that the decision is theirs.
Fairness in outcomes. For life and health insurance in the EU, risk assessment and pricing are high-risk under the AI Act. Even outside that, regulators expect consistent treatment. Calibration on the same claims across teams is how insurers find inconsistency before a regulator does.
What may the agent do, and what stays with a person?
| The agent may | Stays with a person |
|---|---|
| Read the policy, claim history and documents (depth 1) | Declining a claim or voiding a policy |
| Draft first notice of loss files and claim decisions (depth 2) | Claims involving injury, bereavement or vulnerability |
| Settle small, clear claims within limits, once certified (depth 3) | Any claim with a fraud indicator |
| Prepare payments for release (depth 2) | Releasing settlement payments above the insurer's limit |
What should be proven before widening?
Before a capability moves from drafting to settling, insurers should see four things: the agent certified at that depth on their own closed claims; handlers' practicum results with no missed red line; a team leader's calibration showing consistent outcomes; and a named claims manager's signed decision. Banks run a similar four-part test before payments, as how banks should train staff describes.
An example
Take a hypothetical home insurer. Leadership maps the junior handler limit to the agent's ceiling for settling glass and lock claims. Twenty handlers become certified reviewers while the agent drafts decisions in shadow for three weeks. Once calibration shows consistent outcomes and the agent is certified on past claims, it settles glass claims at depth 3. Payment release stays with a handler. The AI agent training plan template for insurers sets this out week by week.
Where the Academy fits
Pryme Intelligence Academy's Operator track certifies claims handlers through a written exam and a two-week practicum of real review decisions signed off by a Pryme assessor. An insurer can adopt the practice of making approval rights depend on a current certificate, depth by depth, the same way it already ties settlement limits to grade.
Questions people ask
Can an insurer pilot an agent before handlers are certified?
Yes, at depth 1 or 2, where handlers decide every claim. See can insurers use AI agents before staff are trained.
Should third-party administrators be trained too?
If their staff approve the agent's held actions on your book, yes, to the same standard.
What do insurers get wrong?
Training underwriters and claims as one group. More in what insurers get wrong about AI agent training and the guide to training teams for AI agents.