How long does it take to train KYC analysts to supervise AI agents?

Expect about six weeks. KYC analysts need roughly 18 hours of learning, then two weeks of supervised decisions on live onboarding and refresh files, and a final exam and sign-off. The concepts come quickly to analysts who already think in terms of document strength. The time goes on learning where the agent's confident checks stop.

Why KYC analysts have a head start

KYC work already rests on the idea that not all evidence is equal. A passport checked against a chip is stronger than a scanned copy; a registry extract beats a client's own statement of ownership. That maps closely onto how a governed agent weighs its facts: sealed, signed, declared, external.

The gap is in reading the agent's work. An agent can extract data from documents, run screening, match names and draft a risk rating in minutes. What the analyst must learn is which of those steps the agent did on strong sources, which on weak ones, and where a held item needs real investigation rather than a second look at the same file.

How the six weeks split

Choose a supervised fortnight with a mix of retail and business customers, so analysts see beneficial ownership cases, not only individuals.

What the agent does and what stays with the analyst

• Depth 1: the agent reads identity documents, registry extracts and screening results.
• Depth 2: it drafts the customer profile, risk rating and a recommendation to approve, hold or decline.
• Depth 3, where earned: it approves low-risk retail customers whose identity checks rest on sealed or signed sources and whose screening is clear.

Stays with the KYC analyst:

1. Any possible screening match, including near-matches on sanctions or politically exposed person lists.
2. High-risk and business customers, especially complex ownership structures.
3. Enhanced due diligence and source-of-wealth judgements.
4. Declining or exiting a customer.

The FATF Recommendations set the standard for customer due diligence and a risk-based approach. The agent supports that work; it does not decide what the business's risk appetite is.

An illustrative example

Take a digital bank's onboarding team of ten analysts. Foundations in week one, then Operator learning over three weeks at about five hours a week. Supervised review starts in week four, with a senior analyst checking each decision.

Calibration on the same twenty files, run as in calibration sessions for KYC analysts, shows that several analysts accepted an ownership chart the customer had uploaded, which is declared, without asking for the registry extract the agent had flagged as missing. The agent had held those files correctly. That gap is fixed before the exam in week six.

Applicants sometimes submit documents designed to steer an agent, or press for fast approval. See training KYC analysts to spot prompt injection and pressure.

Controls while analysts learn

• Keep every business customer at depth 2 until analysts are signed off.
• Hold any screening near-match, whatever the agent's confidence.
• Sample the agent's depth 3 approvals weekly once that depth is earned.
• Know the agent's certificate date. A lapse after 90 days, or a failed run, puts every routine approval back in the queue.

Fraud analysts work at higher speed on transactions rather than identities; compare how long it takes to train fraud analysts. Credit analysts take longer; see the credit analyst timeline. The overall programme is in training teams for AI agents.

Where the Academy fits

KYC analysts who review held onboarding files every day fit the Operator track of Pryme Intelligence Academy. Module tests are set in the learner's own industry and role, so an analyst at a bank meets bank onboarding cases. Certified Pryme Operator requires 32 of 40 on the written exam and a practicum of real review decisions with no missed red line.