A claims team's AI agent competency matrix works in three tables: the level each role should reach on each skill, the level each person holds today, and the gaps between them. The skills are the ones that matter when an agent drafts offers, declines and document requests: cover wording, settlement release, fraud indicators and evidence provenance. An example follows.
Why three tables instead of one?
Because a claims team has different roles touching the same agent, and they should not all reach the same level. A claims handler who releases held settlement offers needs to be strong on wording and fraud indicators. A claims assistant who handles document chasing needs to read verdicts well but will never release a payment. Rating everyone against one standard either over-trains assistants or under-trains handlers.
So the first table sets targets by role, the second records reality by person, and the third lists the gaps. It is the gap list that drives training plans and approval rights. The authority matrix then takes over: it says who may approve what, and should never grant a right where the competency matrix shows a gap.
How do you fill it in?
- Agree the skills. Eight is a good number for claims.
- Set role targets (Table 1) on a scale of 0 (not needed) to 3 (does alone and can be relied on).
- Rate people (Table 2) using evidence: practicum logs, quality samples, assessments.
- List gaps (Table 3) where a person is below their role's target, with an action and a date.
- Repeat quarterly, and whenever a person's credential lapses or the agent's depth changes.
In a typical claims deployment, the agent records first notice of loss at depth 3, requests documents at depth 2, and drafts offers and declines for a handler; releasing a settlement is depth 4 and held for a named handler.
The competency matrix template
Table 1: target level by role
| Skill | Claims assistant | Claims handler | Team leader | Technical claims manager |
|---|---|---|---|---|
| K1 Reading verdicts and deciding bounds | ||||
| K2 Cover and policy wording | ||||
| K3 Evidence provenance | ||||
| K4 Fraud indicators | ||||
| K5 Releasing held settlements | ||||
| K6 Declines and the reasons given | ||||
| K7 Complaints and vulnerability | ||||
| K8 Teaching the agent |
Table 2: actual level by person
| Person | Role | K1 | K2 | K3 | K4 | K5 | K6 | K7 | K8 | Credential | Expires |
|---|---|---|---|---|---|---|---|---|---|---|---|
Table 3: gaps
| Person | Skill | Target | Actual | Action | Due |
|---|---|---|---|---|---|
A worked example
Take a sample motor insurer's claims team. Values are illustrative.
Table 1 (targets). Claims assistants: K1 2, K3 2, K7 2, all others 1 or 0. Claims handlers: 3 on K1 to K7, 2 on K8. Team leaders: 3 on everything. Technical manager: 3 on everything, plus responsibility for assessing others.
Table 2 (extract).
| Person | Role | K1 | K2 | K3 | K4 | K5 | K6 | K7 | K8 | Credential | Expires |
|---|---|---|---|---|---|---|---|---|---|---|---|
| A. Fox | Handler | 3 | 3 | 3 | 3 | 3 | 3 | 2 | 2 | Operator certificate | 31 May |
| R. Sims | Handler | 3 | 3 | 2 | 2 | 3 | 3 | 3 | 1 | Operator certificate | 31 Dec |
| J. Webb | Handler | 3 | 2 | 2 | 2 | 2 | 2 | 2 | 1 | None | |
| T. Pike | Assistant | 2 | 1 | 2 | 1 | 0 | 0 | 2 | 0 | Foundations badge | No expiry |
Table 3 (gaps).
| Person | Skill | Target | Actual | Action | Due |
|---|---|---|---|---|---|
| A. Fox | K7 | 3 | 2 | Shadow complaints handler for a week | End of month |
| R. Sims | K3, K4 | 3 | 2 | Practicum entries focused on third-party estimates and fraud referrals | Next quarter |
| J. Webb | K2 to K6 | 3 | 2 | Two-week practicum; no settlement release until complete | Next quarter |
R. Sims holds a certificate and releases settlements, yet the matrix shows K3 and K4 at 2. That is the most important line in the example: a credential is evidence, not the whole picture. Until the gap closes, the approver rule for the claims team could cap R. Sims's release limit, and the reviewer rota should pair R. Sims with A. Fox on fraud-heavy days.
The same three-table approach suits other functions with mixed roles; compare the competency matrix for an operations team and the matrix for an IT team.
Where the Academy fits
Pryme Intelligence Academy's Foundations badge gives assistants the vocabulary for K1. Certified Pryme Operator, a written exam plus a two-week practicum on real review decisions with module tests set in the learner's own industry and role, is good evidence for level 3 on K1 to K6 for handlers. Insurers can make settlement release depend on a current certificate, depth by depth. See AI agent skills by role.
Questions people ask
Should targets differ between motor and property claims?
The skills are the same; the targets for K2 and K4 may differ, because wording and fraud patterns differ by product. Keep one matrix per team.
Who rates the technical claims manager?
The head of claims, using the same evidence standard. Managers who set others' ratings should have their own on record.
What if a handler's certificate lapses?
Their K5 rating should drop to 2 the same day, because releasing settlements depends on it. It returns when the renewal shows as current at source.