Claims approval · Claims agent v4

Sponsor: COO · Teams: Claims, Fraud, Finance · Blueprint version 1, endorsed 21 Nov

At a glance

Decision gaps

Found in week 1: what the agent and existing controls missed, across 9 decisions and 3 teams.

Named owners

9 / 9

Every decision now has a named owner. Before the sprint: 6.

Conflicts resolved

2 double owners, 1 step nobody answered for, 1 conflict between teams on criteria.

Replay

14 / 200

Past agent decisions the endorsed Blueprint would have flagged or sent to an owner.

Week 1: decision gap assessment

How the agent decides today, what existing controls cover, and what they miss.

How the agent decides

Counts claims per claimant over 60 days; never checks other policies at the same address.

AI policy & AI register

The agent is registered and in policy scope. Neither says what a claims decision must weigh.

RACI

Names teams per process, not per decision: two teams believe they approve adjustments; nobody answers for payout release.

Monitoring & evals

Show what the agent did and how accurate the fraud model is, not whether a decision considered what it should.

Who answers for what

Decision	Can intervene	Answers for it	Keeps watch	Before the sprint
Score claim risk	Fraud lead	Fraud lead	Risk	Criteria in conflict with Claims
Approve / adjust claim	Senior handler	Head of Claims	Fraud lead	Two owners
Release payout	Finance ops	Finance controller	Head of Claims	No owner
… remaining decisions				Unchanged

Conflicts, and how they were resolved

Double owner

Claims and Fraud both believed they approved adjustments above €2,000. Resolved: Head of Claims answers; Fraud lead keeps watch and can hold.

No owner

Payout release had no one who answered for it once the agent approved. Resolved: Finance controller answers for it; payee changes always go to them.

Conflicting criteria

Risk wanted escalation at a third claim on one incident; Business wanted pass under €2,000. Resolved: escalate at the third claim, whatever the amount.

Replay: what the Blueprint would have changed

200 past decisions, sampled from 4 weeks, run against the endorsed Blueprint. No integration with the agent. Your experts then judged each one.

Sent to an owner

9 · mostly recurrent claims on the same incident, to the Fraud lead

Flagged

5 · payouts after a payee change, for the Finance controller

Judged wrong by your experts

6 of the 14

Passed unchanged

186

Open risks and next steps

Open risks

One criterion still informal: what counts as the same incident. Owner: Head of Claims, by 15 Dec.

Next workflow

Supplier qualification, by your team with the standard from this sprint. Your teams can update and re-endorse this Blueprint without MeaningStack.

Steward pilot

Optional: Steward runs the endorsed Blueprint on every live decision, watching first and stepping in when the owners say so, with every decision on record.

MeaningStack · Maastricht · [email protected] · Purple fields are filled from your sprint; values shown are illustrative.