Insurers tend to train claims handlers on the agent's screens and stop there. They confuse the agent's depth with a handler's delegated authority, skip teaching where claim evidence comes from, leave underwriting and pricing teams out, and forget the partners who handle claims on their behalf. Each gap shows in the claims record before it shows in a complaint.
How does a held claim expose the gaps?
Take an illustrative home insurance claim for water damage. The claims agent reads the policy, checks cover and drafts a settlement of 4,000 at depth 2, with the claimant's photos, a contractor's estimate and the policy wording attached. The settlement payment itself sits at depth 4 and is held for a named handler.
The handler who picks it up needs to know three things the screen tour did not teach. Whether they personally have authority to settle at 4,000. Which of the attached facts the agent's draft rests on, and how strong each is. And what to do if the draft is right on cover but wrong on amount. Insurers that train on screens alone leave all three to instinct.
Why is depth not the same as delegated authority?
Insurers already run on authority limits: a handler may settle up to a set amount, a senior handler more, a claims manager more again. It is natural to assume the agent's depth slots into that ladder.
It does not. Depth describes what the agent may do per capability, earned through evidence and enforced at the moment of action. Delegated authority describes what a person may approve. A held settlement must clear both: the agent's bounds put it in the queue, and the handler's authority decides whether they can release it or must pass it up. Training must teach the two side by side, with real cases where they differ, such as a payment the agent was entitled to draft but the handler on shift is not entitled to approve. An insurer can adopt the practice of tying approval rights to a current certificate as well as to the authority limit, depth by depth.
What should handlers know about claim evidence?
Claims run on documents of very different strength. The policy wording from the insurer's own policy system is sealed. A loss adjuster's report comes from a named, identifiable party and is signed. The claimant's description of the loss is declared. Photos and a contractor's estimate sent in by the claimant are external until checked.
An external estimate cannot authorise a payment on its own, and the agent's verdicts reflect that. Handlers who do not understand provenance read a hold on an estimate-backed settlement as the agent being awkward. Handlers who do understand it know to check the estimate against a validated rate or ask for an inspection. Training on real past claims, with their mix of evidence, teaches this faster than any slide.
Why do underwriting and pricing teams need training too?
Claims gets the attention because it has the volume. Underwriting and pricing get less, though their decisions can carry more regulatory weight. In the EU, the AI Act lists risk assessment and pricing for life and health insurance among high-risk uses in Annex III, with obligations for deployers including human oversight under Article 14 and AI literacy under Article 4.
An underwriting agent may gather application data and draft a risk summary at depth 2, with the underwriter deciding terms. The underwriter needs the same reviewer skills as a claims handler, applied to different cases: checking what the applicant declared against what can be verified, spotting where the draft leaned on a weak source, and writing the reason for any decision that departs from the draft. This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
What about partners who handle claims for you?
Many insurers use delegated partners and third-party administrators for parts of the claims process. When agents arrive, partner staff may approve held actions inside the insurer's workspace or run their own process that feeds it.
Their approvals sit in the same record as the insurer's own. If partner staff are untrained, the insurer's agent is being overseen by people it never assessed. Hold partners to the same assessment, the same approval rules and the same register of certificates, and check a sample of their decisions as part of normal oversight.
What should an insurer do instead?
- Teach depth and delegated authority side by side, on cases where they differ.
- Train handlers on claim evidence provenance using real past claims.
- Include underwriting and pricing teams from the start.
- Bring delegated partners into the same assessment and register.
- Recertify on a cycle, and whenever an agent's certificate lapses after 90 days or a run that clears no depth drops it to depth 1 the same day, brief the team on the extra held work coming their way.
The general order is in the guide to training teams for AI agents. For an insurer's version, see how insurers should train staff before deploying AI agents, the AI agent training plan template for insurers, and how much AI agent training costs for insurers. Banks make a related mistake by assuming maker-checker covers agent review, set out in what banks get wrong about AI agent training.
Where the Academy fits
Pryme Intelligence Academy sets module tests in the learner's own industry and role, so a claims handler is tested on claims and an underwriter on underwriting. Handlers and team leads take the Operator track, certified through a written exam and two weeks of real review decisions with no missed red line. Certificates are issued through VerifyDoc.ai, so partners' certificates can be checked at source too.
Questions people ask
Should loss adjusters be trained on the agent?
If they approve held actions in the insurer's workspace, yes, to the same standard as handlers. If they only supply reports the agent reads, they need to know their reports are treated as signed, named sources and must be accurate.
Can an agent decline a claim?
It can draft a declinature at depth 2 for a person to review and send. Declining cover has significant effects for the customer, so keep the decision with a trained handler and require a written reason.
How do we train handlers to spot vulnerable customers?
Use real past claims where vulnerability was present, including ones where it was missed. The agent can flag signals in what the claimant wrote, but the handler decides what changes, and must be trained to act on it.