Credit analysts should hold a current certificate before agents go live because the agent will build the credit case and the analyst will sign it. Spreading, ratio analysis and the first draft of the memo move to the agent. The recommendation stays with the analyst, and a recommendation built on figures the analyst cannot trace is a recommendation nobody made.
Where does the agent sit in a credit analyst's workflow?
A credit analyst's week is mostly preparation. Financial statements arrive in different formats. The analyst spreads them into the house template, calculates ratios, checks covenants on existing facilities, reads management commentary, pulls bureau data, and writes a memo with a recommendation for a credit committee or an approver.
A governed agent takes the preparation. At depth 1 it reads the statements, the bureau report and the existing facility terms. At depth 2 it spreads the figures, calculates the ratios and drafts the memo. It does not approve credit. The recommendation, and any decision to decline, stays with people.
That sounds like a clean split, and it is. The risk is in what the analyst does with a finished-looking draft.
Why can a tidy credit memo mislead?
An agent-drafted memo will be consistent, well formatted and arithmetically correct. That is exactly what makes it dangerous to an untrained reviewer. Most credit errors are not arithmetic. They are about which numbers went in and what they mean.
Take provenance. Under the Pryme Context Protocol every fact reaching the agent is stamped with where it came from. Audited accounts filed by the borrower's auditor might arrive signed. Management accounts sent by the borrower's finance director are declared. A spreadsheet attached to an email is external. The agent will use all of them, and it will show their class, but the ratios look the same on the page. A certified analyst checks the class before trusting the leverage figure. An uncertified one reads the ratio.
Then there is scope. The agent might not have been given the borrower's other facilities with other lenders, or the guarantees behind them. A memo can be complete with respect to its sources and incomplete with respect to the borrower. Analysts must know what the agent could and could not see.
Why does the regulatory picture raise the stakes?
Credit decisions about individuals are among the uses the EU AI Act lists in Annex III as high-risk: creditworthiness assessment and credit scoring of natural persons. Deployers of high-risk systems have obligations, including human oversight under Article 14, and the people providing that oversight need to understand the system well enough to do it. In the US, ECOA and Regulation B require specific reasons when credit is denied, including when a model is used, so the person signing a decline must be able to state reasons that are true for this applicant.
For commercial credit the rules differ, but the principle holds. The person whose name is on the recommendation must be able to explain it without the agent. That is the subject of whether credit analysts need a certificate to approve agent actions, and the short answer is that the business should require one wherever their approval relies on agent work.
What should the certificate test?
For credit analysts, a useful certificate demonstrates:
- Source reading. For any figure in the memo, the analyst can name its source and its provenance class.
- Gap finding. They notice what the agent did not have: missing facilities, an outdated year, a guarantor with no financials.
- Reasons that hold. For a decline, they can write reasons that are specific, true and their own.
- Knowing the floor. They understand that the agent will refuse to propose terms outside the lender's policy, and that a refusal is not something to work around.
A practicum is where this shows. An assessor watching two weeks of real decisions sees whether the analyst opens the sources or reads only the summary.
Why test the agent and the people together?
Before go-live, teams red-team the agent: feeding it manipulated statements, borrowers with hidden related-party debt, and declared figures that contradict sealed ones. That exercise doubles as reviewer training. The analysts who will review the agent's memos should see what the agent did with each trap and whether its draft would have fooled them. Where it would have, that is the training gap to close before certification.
What is the trade-off?
Credit teams are often small and specialised. Certifying every analyst before go-live means weeks during which the agent's time savings are partly unrealised. Senior analysts may resist, on the reasonable grounds that they have assessed credit for years without a certificate.
The response is to separate credit skill from agent skill. Nobody is questioning the senior analyst's credit judgement. The certificate tests whether they can tell when the agent's preparation has misled them. That is a new skill, and seniority does not confer it.
Meanwhile the agent can run at depth 1 and 2 from the first day, saving time on spreading while analysts check every figure. As analysts certify, they can rely on the agent's preparation with confidence rather than redo it.
The accountability question settles the matter for most lenders. When a facility goes bad, the record will show who signed the memo, and the question of whether credit analysts are accountable for an agent's mistakes is answered partly by whether the analyst was equipped to catch them. The same reasoning applies to credit controllers and fraud analysts further along the credit lifecycle.
What should a chief credit officer do now?
Require a current certificate for any analyst who signs a recommendation based on agent work. Use red-team cases as training material. Keep declines and any terms outside policy with people. And keep the certificates and practicum logs where a supervisor or auditor can see them. The guide to AI literacy and certification for people who work with agents sets out the levels.
Where the Academy fits
Pryme Intelligence Academy's Operator track trains analysts to read verdicts, check provenance and correct an agent well, with module tests set in their own industry and role. The certificate needs a written exam and a two-week practicum of real review decisions signed off by a Pryme assessor. Lenders can check each analyst's certificate at source and see its expiry on the public register.
Questions people ask
Can an agent recommend approval of a credit facility?
It can draft a memo with a proposed recommendation at depth 2. The decision to recommend, and the approval itself, stay with people under the lender's credit authority.
Does a credit committee member need certification?
Committee members who rely on agent-prepared memos benefit from at least a grounding in how those memos are built. Analysts who prepare and sign the memos need the full reviewer certificate.
What if the agent's ratios differ from the analyst's own?
That is a finding, not a nuisance. The analyst should trace the difference to its source, record which figure is right and why, and bank the correction so the agent's next draft improves.