How to keep HR advisers ready for an AI agent’s 90-day recertification

Line up policy changes with the agent's cycle and brief HR advisers on both together. Every 90 days the HR agent is recertified on your own employee queries and letters. What it answers alone, what it drafts and what it escalates can all move. Advisers need the delta, a few practice cases and a firm reminder of the decisions that never leave them.

What does recertification change for an HR adviser?

An HR agent usually answers employee questions about policy at depth 3, drawing on the policy library, which is a sealed source. It drafts letters for advisers at depth 2: absence follow-ups, flexible working responses, contract variation letters. It records case notes at depth 2. Any change that affects pay or contract terms is held for a named adviser. Disciplinary, grievance and dismissal decisions stay with people at every depth.

Recertification re-runs the agent's cases on recent queries and letters. The escalation gate matters most: an employee asking about leave may be describing harassment or a health crisis, and the agent must hand over. The grounded gate matters next, because an answer from an old policy version is wrong even if it reads well.

The most common change comes from policy, not from the agent. A new leave policy enters the library mid-cycle; the agent's answers on leave may be held for review until the agent is recertified on the new version. Advisers then see those queries arrive with drafts.

Who owns readiness?

The HR operations lead owns adviser readiness and the rota for escalations. The policy owner owns the library and the timing of changes. The builder owns recertification. An employee relations specialist should check any change to escalation triggers.

Steps for each cycle

1. Person (policy owner): list policy changes due in the next 90 days and agree with the builder when each will enter the library.
2. Agent, depth 1: report the cycle. Questions answered, drafts produced and edited, escalations and refusals, with the deciding bound for each.
3. Person (HR operations lead): read a sample of answered questions, weighted to leave, pay and conduct.
4. Person (builder): recertify on recent queries, including ones touched by new policy.
5. Person (HR operations lead and builder): write the change note: topics the agent now answers or only drafts, new escalation triggers, policy versions in use.
6. Person (employee relations specialist): approve escalation changes.
7. Agent, depth 2: draft practice queries, anonymised and checked by the operations lead.
8. Person (HR operations lead): brief advisers, then sample the first fortnight.

What should HR advisers practise?

Escalations. Work through three short employee messages that sound routine but are not. Advisers should say what they would do in the first hour.

Reading answers against the policy. If the agent answered a question about notice periods, the adviser should be able to check which policy version and clause it relied on.

Letters with consequences. A contract variation letter changes terms. Advisers should read every draft as if signing it, because in effect they are.

Decisions solely by automated means. Under GDPR, Article 22 restricts decisions based solely on automated processing that have legal or similarly significant effects on people. Keeping people decisions with advisers is a design choice that supports this, and advisers should understand why the agent drafts but never decides.

Controls and the evidence they leave

• Policy calendar aligned with cycles. Evidence: the calendar and library version history.
• Change note with employee relations approval. Evidence: dated, signed note.
• Adviser briefing. Evidence: attendance and practice answers.
• Named approval on held pay and contract changes. Evidence: approvals in the append-only record.
• Practicum for new advisers. Evidence: the practicum log template for HR advisers.

What to measure

Time from an escalation to an adviser's first contact. The edit rate on letters by type. And repeat queries on the same topic from the same employee, which show an answer that did not land. Employee data is sensitive, so measure through the record rather than by exporting conversations.

Where does readiness slip?

Policy changes without warning. The agent keeps answering from the version it was certified on, or its answers on that topic are held, and advisers do not know why. The policy calendar fixes this.

Advisers treat the agent's case notes as the record. Notes drafted from a conversation can miss what was said in tone or between the lines. Advisers confirm them.

Approval rights blur. Whether advisers need a certificate to approve held pay or contract changes is a decision HR makes, set out in do HR advisers need a certificate to approve AI agent actions. Who answers for mistakes is in are HR advisers accountable for an AI agent's mistakes. Wealth advisers face similar letter risks; compare keeping wealth advisers ready for recertification.

Where the Academy fits

Pryme Intelligence Academy suits HR teams at two levels. The Foundations track gives any adviser the basics in about 4 hours, free. Advisers who approve held changes or manage escalations can take the Operator track, whose module tests are set in the learner's own industry and role. Lessons can be listened to as audio, with highlights and notes. More in the guide to AI literacy and certification for people who work with agents.