Run the practicum on live credit files the agent has spread and drafted, and count each judgement as a decision: each spread adjustment, memo section, covenant flag and decline reason. A senior analyst with a current certificate decides independently and nothing reaches committee without them. The bar is at least 95 decisions in every 100 right and no figure reaching committee that does not trace to its source.
What is the practicum, and who owns it?
Credit analysts working with an agent receive spreads, ratios, covenant checks and drafted memo sections. Their job is to judge what the draft gets wrong, what it leaves out and what it rests on. A written test can check that they know the policy; the practicum checks that they catch errors in real files, under real deadlines.
The head of credit analysis owns the practicum and names the senior analyst who supervises. Credit risk sets the red lines. The analyst owns each decision and writes a short reason.
How do you build the queue?
Credit files are long, so define the unit of decision carefully. Each of these counts as one:
- accepting or adjusting a line in the agent's spread, such as an add-back or a reclassification;
- accepting or rewriting a drafted memo section;
- confirming or rejecting a covenant flag;
- tracing a key figure to its source;
- wording decline reasons from a decision already made.
Four to six files a day yields about 12 to 18 decisions, and 120 to 180 across the fortnight. The files should mix borrower sizes, statement formats and outcomes. Seed anonymised past files for anything thin: group accounts, a change of accounting standard, a borrower whose management accounts disagree with the audited ones.
For example, a seeded file might show management accounts, declared by the borrower, reporting earnings well above the last audited year, which is signed by a named auditor. The agent's draft uses the management figure for debt service cover. The analyst should flag it, show both, and base the view on what the evidence supports.
The steps
- Person (head of credit analysis): brief the analyst on the decision unit, the bar and the red lines.
- Agent, depth 1: route files for the analyst's portfolio to the practicum queue.
- Agent, depth 2: present spreads and drafts with every figure linked to its source and marked by provenance.
- Person (analyst): decide each item and write a reason.
- Person (senior analyst): decide independently, then compare.
- Person (senior analyst): sign off each memo before it goes to committee.
- Agent, depth 1: log decisions, matches and red-line flags.
- Person (senior analyst, then assessor): review the log at the end.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Spreading | Spreads into the template | 2 | Analyst adjusts; senior checks |
| Ratios and covenants | Calculates and flags | 2 | Interpreting flags |
| Memo sections | Drafts with sources | 2 | Analyst rewrites; senior signs |
| Tracing | Links figures to sources | 1 | Checking the source says it |
| Decline reasons | Drafts from the decision | 2 | Analyst confirms they are the real reasons |
| Log | Records decisions | 1 | Confirming |
How do you set and apply the bar?
A decision is correct if it matches the senior's or the senior agrees it was sound. An adjustment within a defensible range counts as correct.
The red lines: a memo figure going to committee that does not trace to its source; presenting declared figures as verified; decline reasons that are not the real reasons; missing a covenant breach the agent did not flag. Where lending is to individuals, creditworthiness assessment is also listed as high-risk in Annex III of the EU AI Act, which raises the bar for human oversight of exactly the decisions this practicum tests. Decline reasons matter in law too: in the US, ECOA and Regulation B require specific reasons when credit is denied, including when a model is used. One miss fails the practicum.
Which controls matter, and what evidence do they leave?
- Senior sign-off before committee. Evidence: the senior's name on each memo.
- Source links on every figure. Evidence: the memo with its links.
- Independent senior decisions. Evidence: logged before comparison.
- Daily debrief. Evidence: a line per mismatch.
- Full log. Evidence: the practicum log template for credit analysts.
What should you measure?
Correct decisions out of the total, by unit type. Red-line misses (zero). Errors the analyst caught that the agent made, compared with errors the senior caught that the analyst missed. And time per file, watched for corner-cutting rather than as a target.
Where does it usually go wrong?
The decision unit is the whole file. Then a fortnight yields too few decisions for the bar to mean anything, and one error is decisive. Count sections.
Every file is a clean renewal. Seed the hard ones.
The senior edits the memo directly instead of logging a separate decision. The log then shows agreement that never happened.
The practicum ends at quarter-end, when committee is busiest. Move it. Credit control uses a similar log; compare the practicum log template for credit controllers and running a practicum for fraud analysts.
Where the Academy fits
Pryme Intelligence Academy's Certified Pryme Operator practicum follows the same bar: two weeks of real review decisions, at least 95 in every 100 correct and no missed red line, signed off by a Pryme assessor. Module tests before it are set in the learner's own industry and role, so an analyst's preparation uses credit cases. More in the guide to AI literacy and certification for people who work with agents.
Questions people ask
Should the analyst see the agent's earlier drafts on the same borrower?
Yes, as context. Prior memos often show what the last analyst had to correct.
Can the analyst's recommendation be scored?
Yes, as one decision per file, judged on whether the reasoning supports it. Credit views can differ; reasoning should not be thin.
What if committee rejects a practicum memo?
Log the reason and discuss it at the debrief. It is evidence about the analyst and the senior alike.