AI agent competency matrix for collections team

A collections team's AI agent competency matrix lists seven skills collectors need when an AI agent proposes payment plans, sends reminders and holds settlements, names the evidence that proves each level, and shows who holds what. Because hardship and dispute cases arrive at any hour, the matrix ends with a shift cover check. Template and example below.

What do collectors need to be good at now?

The AI agent does much of the routine contact. In a typical deployment it sends reminders within contact rules at depth 3, drafts payment plans at depth 2 from the customer's stated income and spending, and holds settlements, plan changes outside policy and anything touching a dispute for a named collector. It refuses to discuss an account with anyone but the account holder or an authorised representative.

That moves the collector's skill from making contact to judging it. The hardest cases, a customer who mentions losing a job, a dispute that was never resolved, a relative offering to pay, are exactly the ones the AI agent holds. A collector who is excellent on the phone but weak at spotting a hardship signal in a written note is now the weak link. The matrix makes that visible.

How do you fill it in?

• Rate each person on each skill as none, developing, competent or coach.
• "Competent" requires the evidence named in the template, not a manager's impression.
• Record each person's credential and expiry.
• Complete the shift cover check every time the rota changes.

The competency matrix template

Skills and the evidence for "competent"

1. Hardship and vulnerability signals. Evidence: a practicum log with every signal in the queue caught over two weeks.
2. Disputes. Evidence: correctly pauses collection on every disputed item in a sample of 20.
3. Affordability provenance. Evidence: explains, for five plans, which figures are declared and which are verified, and whether that matters for the plan.
4. Settlement and plan authority. Evidence: no decisions above authority in the practicum log; reasons recorded for every exception.
5. Third-party contact. Evidence: upholds the AI agent's refusals correctly and takes payments from third parties without discussing the account.
6. Reading verdicts. Evidence: explains the deciding bound on ten held items.
7. Teaching the AI agent. Evidence: at least two corrections submitted with sources, accepted by the agent's owner.

Who holds what

Shift cover check

A worked example

Take a sample telecoms provider's collections team of seven. Values are illustrative. N = none, D = developing, C = competent, Co = coach.

The Saturday line is the finding. No one on shift can approve a settlement or plan outside policy, so held items wait until Monday, or someone approves outside their competence. The fix is a rota change, not a training plan: move S. Reid to alternate Saturdays and make sure the reviewer rota for the collections team checks skill 4 cover. The longer-term plan moves E. Cruz to competent on skill 1 through a supervised fortnight.

The approver rule for the collections team should require "competent" on skills 1, 2 and 4 plus a current certificate for anyone releasing settlements. For how a specific market shapes these skills, see AI agent training for collections agents in telecoms in Australia. Other functions use the same layout; compare the matrix for an operations team.

Where the Academy fits

Pryme Intelligence Academy's Certified Pryme Operator requires two weeks of real review decisions with at least 95 in every 100 correct and no missed red line, which matches the evidence for skills 1, 2 and 4. Its module tests are set in the learner's own industry and role. The admin view shows certificates and expiry dates, which fill the last two columns. More in AI agent skills by role.