Base the claims approver rule on handling authority, the same bands your claims team already uses, then add two things agents need: every decline is approved by a named handler, and every settlement payment is approved by someone whose certificate is current. Above a set value, a second approver. Suspected fraud leaves the agent's queue entirely.
What is the claims approver rule, and who owns it?
Claims teams already have authority levels: a handler can settle up to one amount, a senior handler up to another, a claims manager above that. An agent does not replace those levels. It prepares claims, drafts letters and holds payments for a handler with the right authority. The approver rule adds the agent's depths to the authority table you already run.
The claims manager owns the rule. Compliance or conduct risk signs it off. The agent owner configures the holds. Our explainer on what an agent approver is describes the role, and our guide to AI agent skills by role covers what handlers need to know.
What are the steps?
- Claims manager: lists the agent's actions by depth. Depth 1: reads the first notice of loss, the policy wording and the claim history. Depth 2: drafts acknowledgement, information requests, decline and settlement letters. Depth 3: requests documents, appoints a loss adjuster from a panel, updates the claim record. Depth 4: issues a settlement payment.
- Claims manager: maps depth 4 to existing authority bands. The agent holds each payment for a handler whose band covers the amount.
- Claims manager: puts every decline at depth 2. The agent drafts the decline with the policy wording it relied on; a handler decides and sends.
- Claims manager: adds a second approver above a set value, and for any payment to a payee other than the policyholder.
- Claims manager: ties each band to a current certificate. An approver whose certificate has lapsed drops out of the routing until renewed.
- Second line: signs off. Agent owner: configures and files.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Read claim | Checks cover and history | 1 | Nothing |
| Request documents | Sends from approved templates | 3 | Nothing |
| Appoint adjuster | Instructs from the panel | 3 | Choice outside the panel |
| Draft decline | Drafts with wording cited | 2 | Handler decides and sends |
| Settle | Prepares and holds payment | 4 | Handler within band; second above set value |
| Suspected fraud | Refers out | 1 | Investigations team |
Why are declines held at depth 2 even when the agent is accurate?
A decline is the decision customers challenge, complain about and take to an ombudsman. The handler who sends it needs to own the reasoning, not inherit it. Keeping declines at depth 2 means a person reads the policy wording the agent cited, checks it against the facts, and decides. The agent saves the handler the drafting, not the judgement.
Settlements are different. A settlement within the policy and within the band is often routine, and the hold is about money control, not judgement. That is why the approver for a settlement is defined by amount, while the approver for a decline is defined by the decision itself.
Fraud is different again. If the agent sees indicators, it should not draft a decline or a settlement. It refers the claim to investigations and stops. A handler who sees fraud indicators in a held payment should do the same.
Which controls make the rule work, and what evidence do they leave?
Bands match existing authority. Evidence: the approver authority matrix for the claims team, reconciled to the handling authority table.
Declines never sent by the agent. Evidence: every decline in the verdict record shows a handler as sender.
Third-party payees need two approvers. Repairers, solicitors and other payees are a common fraud route. Evidence: two approvers on each such payment.
Certificates checked at approval. Evidence: approver certificate status on the date of each release.
Payee details are sealed or signed. Bank details from an emailed repair invoice are external and cannot authorise a payment. Evidence: provenance stamps on the held payment.
What should you measure?
- Declines overturned on complaint, traced to the agent's draft or the handler's decision.
- Payments above the second-approver threshold with one approver. The target is none.
- Time from complete claim to settlement, by band.
- Fraud referrals from the agent, and how many investigations confirmed.
- Handlers approving outside their band. The target is none.
Where does it usually go wrong?
Bands drift upward. A busy month leads to temporary authority increases that never come back down. Make any change to bands a dated change with an end date.
The agent's confidence is mistaken for authority. A well-written settlement letter still needs the right handler's approval. Fluency is not evidence.
Investigations are bypassed. A handler releases a held payment despite indicators because the customer is chasing. The rule should say plainly that indicators stop the payment.
IT and operations rules are reused. Approvals for system changes or operations work look different; compare the approver rule for the IT team and the competency matrix for the claims team when designing your own.
Where the Academy fits
Pryme Intelligence Academy's Operator track is for the handlers and team leads who review held claims actions every day. Its module tests are set in the learner's own industry and role. Certified Pryme Operator is valid for 12 months, so a claims business can tie authority bands to a current certificate and see expiry dates in the admin view.
Questions people ask
Can a handler approve a payment above their band if the agent recommends it?
No. The agent's recommendation does not change authority. The payment is held for someone whose band covers it.
Should small settlements be fully automated?
Some insurers allow small, clearly covered settlements to run at depth 3 within bounds once the agent is certified for it. Declines and third-party payments should still be held.
Who owns the policy wording the agent cites?
The product or underwriting team, through the policy library. The agent reads it as a sealed source; it should never rely on a policy summary from an email.