KYC analysts log every onboarding or refresh file they decide during the practicum: the AI agent's draft risk rating and recommendation, each document and its provenance, the screening result, the ownership picture, the analyst's decision and reason, and a supervisor's independent call. Three red lines sit in the header. The template and four sample files follow.
What is the analyst checking that the agent cannot?
The agent is good at assembly. It extracts details from identity documents, runs screening, draws the ownership chart for a company, and drafts a risk rating at depth 2. What it cannot do is decide whether the file is enough. That is the analyst's call, and it rests on three questions the log makes explicit:
- Is each document what it claims to be? A certificate of incorporation pulled from a registry is signed. A utility bill the customer uploaded is declared. A PDF forwarded by an introducer is external.
- Is every screening match resolved? A potential match closed with one discriminating fact is weaker than one closed with three.
- Is every owner above the threshold identified and verified? Ownership charts with a trust or nominee in the middle are where people go missing.
Customer due diligence is a core FATF expectation, applied on a risk basis. The practicum tests whether the analyst applies that judgement to the agent's work. How it is set up is covered in how to run a two-week practicum for KYC analysts.
How do you fill it in?
The agent pre-fills fields 2 to 6 from its file at depth 1. The analyst completes 7 to 9 after reading the documents themselves, not only the agent's extraction. The supervisor completes 10 without seeing the analyst's answer. Fields 11 and 12 are completed at the daily debrief.
The practicum log template
Header
- Analyst and team.
- Supervisor and certificate number.
- Customer types in scope: individuals, sole traders, companies, trusts.
- Dates.
- Bar: at least 95 correct in every 100 files, no missed red line.
- Red lines: approving with an unresolved sanctions or politically exposed person match; treating a declared or external document as verified; missing an owner above the firm's threshold.
Per-file entry
- Entry number and date.
- Customer reference and type.
- Review type: onboarding, periodic refresh, trigger event.
- Documents, each with its provenance class.
- Screening results and how each match was resolved.
- Ownership and control, with owners above threshold listed.
- Agent's draft risk rating and recommendation, with deciding bound.
- Analyst's decision: approve, request more, escalate, decline.
- Analyst's reason.
- Supervisor's decision.
- Match: yes, no or defensible.
- Red line: none, caught or missed, and a debrief note.
Four sample files
Take a sample electronic money institution onboarding business customers. Values are illustrative.
File 52. A private limited company with two directors, each owning half. Registry extract (signed), directors' passports checked through the identity service (signed), proof of address uploaded (declared). Screening: clear. Agent: low risk, approve. Analyst: approve. Supervisor: approve. Match: yes.
File 53. A company owned by a holding company in another jurisdiction, itself owned by a trust. The agent's chart stops at the holding company and rates medium risk. Analyst: request the trust deed and identify the settlor, trustees and beneficiaries. Supervisor: same. Red line 3: caught.
File 54. A sole trader whose name partially matches a politically exposed person. The agent closes the match on a different date of birth. Analyst: approve. Supervisor: also confirm nationality and middle name before approving; date of birth on the screening record is incomplete. Match: no. Red line 1: missed, because the match was not fully resolved when the analyst approved.
File 55. A refresh for an existing customer. The new bank statement arrived from the introducer by email (external). Agent: approve refresh. Analyst: request the statement directly from the customer through the secure portal. Supervisor: same. Red line 2: caught.
File 54 fails the bar. It is a common pattern: the agent's reasoning was plausible, the analyst checked it rather than the underlying record, and the record was weaker than the reasoning.
How do KYC logs fit with credit and fraud logs?
The same customer often passes through three practicums. KYC checks who they are; the practicum log for credit analysts checks whether to lend; the log for fraud analysts checks whether a later instruction is really theirs. All three use the same provenance classes, which makes the logs easier to read side by side. Teams preparing for the Certified Pryme Operator practicum at Pryme Intelligence Academy can use this format as their record; the wider picture is in AI literacy and certification for people who work with agents.
Questions people ask
Should analysts open every document, or rely on the agent's extraction?
Open every document during the practicum. The point is to learn where extraction goes wrong, and that only happens by comparing the two.
What if the customer is lost because the analyst requested more?
Record it in the note. A practicum measures correctness, not conversion; a request for more that the supervisor agrees with is a correct decision.
Does the analyst need a certificate to approve onboarding after the practicum?
That is the firm's decision, depth by depth. See do KYC analysts need a certificate to approve AI agent actions and, on responsibility, are KYC analysts accountable for an AI agent's mistakes.