KYC analysts own the onboarding decisions they approve, including every screening match they clear. An AI agent can extract identity data, run sanctions and politically exposed person checks, and draft a risk rating, but the analyst who accepts its recommendation owns that acceptance. Errors hidden from the analyst, or built into the agent's rules, sit elsewhere.
Where does the KYC analyst sit in the decision chain?
Customer due diligence is one of the foundations of the FATF Recommendations, and every firm in scope has to show it knows who its customers are. An onboarding agent does much of the preparation: it reads the identity document, checks it against the application, screens the name, pulls company registry data for business customers, and proposes a risk rating.
Opening an account is the action that matters. Most firms hold it for a person whenever anything is less than clean: a possible screening match, a missing document, a high-risk country, a complex ownership chain. The analyst's approval turns the agent's draft into a customer relationship.
What may the agent do, and what stays with the analyst?
- Allowed within bounds: onboarding low-risk individual customers whose identity is verified against a signed source and whose screening is clean.
- Held for an analyst: any possible screening match, any business with layered ownership, any customer the agent rates above low risk, and any mismatch between documents.
- Refused: onboarding on a document the agent cannot verify, or clearing a sanctions match.
- The analyst keeps: deciding whether a match is a false positive, judging source of wealth, and escalating to enhanced due diligence.
What does the record show about who decided?
The record shows each fact and where it came from. An identity check returned by a verified document service is signed. A company's ownership from an official registry is signed. The applicant's own statement about their job is declared. A screening list result is signed by its publisher.
For a screening match the analyst cleared, the record shows the match, the reason the agent gave for considering it a false positive, the evidence behind that reason, and the analyst's decision. If the agent proposed clearing a match on a date-of-birth mismatch and the analyst accepted it without checking the list entry, the clearance is theirs.
An example: the match cleared on one field
Take a digital bank onboarding small business owners. An applicant's name partially matches a sanctions list entry. The agent notes the date of birth on the list differs by one year and proposes treating it as a false positive, but holds the case. The analyst clears it.
Later review finds the list entry gave an approximate year of birth, and the applicant was the listed person. The record shows the agent's proposal, the one-field reasoning and the analyst's clearance. The analyst is accountable for clearing a match on a single weak field. The agent's rules should also never present one mismatch as enough, and that is a design fix.
What controls make this fair?
- No match cleared on a single field. Require at least two independent differences, both shown on screen.
- Declared facts cannot lower a risk rating without signed support.
- Analysts practise on real files through a two-week practicum for KYC analysts, recorded in a practicum log template for KYC analysts.
- Review at each renewal. Agent certificates lapse after 90 days, and screening rules change; see how to prepare KYC analysts for 90-day recertification.
Later in the customer's life, other teams face the same question; compare fraud analysts and AI agent mistakes and credit analysts and AI agent mistakes. The wider topic is in AI literacy and certification for people who work with agents.
Where the Academy fits
The Operator track at Pryme Intelligence Academy teaches how a governed agent decides and how to read the workspace, including provenance and held actions. Its practicum, two weeks of real review decisions signed off by a Pryme assessor, fails outright on a missed red line, and clearing a sanctions match on one weak field is exactly that.
Questions people ask
Is the analyst accountable for a customer the agent onboarded alone?
Not directly. If the agent onboarded within bounds the firm set, the people who set those bounds own the outcome, and the record names them.
What if the identity document was a high-quality forgery?
If the agent's verification source passed it and the analyst had no visible reason to doubt it, the analyst acted reasonably. The record shows the check and its result.
Should analysts see the agent's reasoning or just its rating?
The reasoning, always. A rating without reasons cannot be reviewed, only accepted.