Credit decisions about people are among the most tightly governed uses of AI, which is why lending teams need a competency matrix. Once agents gather documents, check income and draft decision packs, loan officers must verify the agent's evidence, explain every decline in specific reasons, and keep the credit decision itself with a person who can defend it.
Why does lending need a stricter matrix than most teams?
The stakes and the rules are both higher. Under the EU AI Act, creditworthiness assessment and credit scoring of natural persons are listed as high-risk in Annex III, with obligations around human oversight (Article 14) and a duty on providers and deployers to ensure AI literacy among their staff (Article 4). In the US, ECOA and Regulation B require specific reasons when credit is denied, including when a model is used. Other markets set their own expectations.
Each of those rules lands on people. Human oversight only works if the overseer knows what the agent did and can disagree with it. Specific reasons only work if the officer can trace each reason to a fact. A competency matrix turns those expectations into skills you can test. It sits within the broader guide to AI agent skills by role and follows the logic of a training matrix: skills, people, targets, evidence.
How do you fill it in?
Use one entry per skill, with five fields. Levels run from 1 (understands) to 3 (decides and coaches).
- Skill: what the person can do, written as a verb.
- Why it matters in lending: the error it prevents.
- Target level by role: loan officer, underwriter, credit manager, agent owner.
- Evidence: how the level is proven.
- Refresh trigger: what forces a re-check, such as a policy change or a new data source.
The template
- Skill: _ / Why it matters: _ / Loan officer: __ / Underwriter: __ / Credit manager: __ / Agent owner: __ / Evidence: _ / Refresh: _
- Repeat for each skill below: - Explains depth, the four verdicts and provenance to an applicant - Checks income and identity evidence against its provenance class - Spots a document that arrived from outside and is being treated as fact - Decides held applications with written reasons - Writes decline reasons that trace to specific facts - Recognises when an applicant may be vulnerable - Reconstructs a decision from the record for a complaint - Reads a certification result and challenges the agent's depth
A worked example
A hypothetical consumer lender with ten loan officers, four underwriters, two credit managers and one agent owner. Its agent collects documents at depth 1 and drafts decision packs at depth 2. No credit decision is made by the agent.
| Skill | Officer | Underwriter | Manager | Owner | Evidence | Refresh |
|---|---|---|---|---|---|---|
| Explains verdicts to applicants | 2 | 1 | 2 | 2 | Call sample | Yearly |
| Checks evidence provenance | 2 | 3 | 3 | 2 | 40 files re-checked | New source |
| Spots external documents treated as fact | 2 | 3 | 3 | 2 | Scenario test | New source |
| Decides held applications | 1 | 3 | 3 | 1 | Two-week live sample | Policy change |
| Writes traceable decline reasons | 2 | 3 | 3 | 1 | Reason audit | Policy change |
| Recognises vulnerability | 3 | 2 | 3 | 1 | Case review | Yearly |
| Reconstructs for a complaint | 1 | 2 | 3 | 3 | Walkthrough | Yearly |
| Reads certification results | 1 | 1 | 2 | 3 | Owner review | Each certification |
Loan officers are at level 3 on vulnerability, above underwriters, because they talk to applicants and hear what an agent reading documents cannot. Underwriters lead on provenance: a payslip uploaded by the applicant is declared, while an income figure from a verified payroll provider is signed. Treating the first like the second is the commonest way a draft pack misleads.
Decline reasons get their own row because the agent drafts them. An officer at level 2 checks that each reason in the draft points to a fact in the file, and rewrites any that are generic. The person who sends the decline owns its reasons.
The agent owner sits at level 1 on deciding applications, by design: the owner answers for the agent but should not be one of its approvers. The approver rule for the lending team records that separation, and the authority matrix sets the limits.
Where the Academy fits
Pryme Intelligence Academy gives the whole lending team Foundations in about four hours, which covers the literacy expected of everyone. Underwriters and credit managers take the Operator track, whose practicum allows no missed red line. A lender can adopt the practice of granting approval rights only to people with a current certificate, depth by depth.
Questions people ask
Does the agent owner need to be a credit specialist?
They need enough credit knowledge to judge whether the agent's bounds match the lending policy. Many lenders appoint the credit manager who owns the policy, supported by a builder who writes the brief.
How does this compare with the IT team's matrix?
The IT team's competency matrix covers the connections and sources the agent uses. Lending relies on IT to register each data source with the right provenance class; the lending team then relies on that class when weighing evidence.
What changes for compliance?
Compliance tests whether the oversight and reasons are real. The compliance team's competency matrix adds skills such as sampling declines and checking reasons against the file.