How to keep contact centre managers ready for an AI agent’s 90-day recertification

Plan each recertification like a forecast change. Every 90 days the customer-facing agent is recertified on your own conversations, and the set of requests it resolves, drafts or hands over can shift. Contact centre managers need the change note early enough to re-forecast staffing, update quality scorecards, brief team leaders and keep certified people on the held queue every shift.

What does recertification change on the floor?

A contact centre agent typically resolves routine requests end to end at depth 3: order status, address changes, appointment moves, password help within policy. It drafts replies at depth 2 for human advisers to send on harder cases. Refunds and goodwill credits that move money are depth 4 and held above the business's own limit. Complaints, signs of vulnerability and anything outside scope are handed to a person.

Recertification re-runs the agent's cases on recent conversations. The result can widen or narrow the set of requests the agent resolves, change the refund value it may pay without a person, or add escalation triggers. Each of those moves contacts between the agent and your human advisers, so each one changes your staffing model.

A run that clears no depth is the hard case. The agent drops to depth 1 the same day: it can read the account but resolves nothing. Every contact it would have handled reaches a human queue, often within the hour.

Who owns the cycle?

The contact centre manager owns readiness across shifts: forecast, rota, scorecards and briefings. The builder who certifies the agent owns recertification. The customer operations director owns the business's own limits, such as refund values. Workforce planning should see the change note at the same time as the manager.

Steps for each cycle

1. Person (contact centre manager): set the date away from seasonal peaks and launches.
2. Agent, depth 1: report the cycle. Contacts resolved, drafts sent with or without edits, refunds paid and held, handovers, refusals, each with the deciding bound.
3. Person (quality lead): review a sample of agent-resolved contacts and handovers, weighted to complaints.
4. Person (builder): recertify on recent conversations.
5. Person (manager and builder): write the change note: request types added or removed, refund limits, new handover triggers.
6. Person (workforce planner): re-forecast the human queue for the next 90 days, including a fallback forecast if the agent drops to depth 1.
7. Agent, depth 2: draft scorecard updates and practice conversations, checked by the quality lead.
8. Person (manager): brief team leaders, who brief their teams by shift, and sample the first fortnight.

What should the manager retrain, and how?

Team leaders first. They run the held queue and coach advisers. They need the full change note and two worked examples per change.

Advisers by shift. Give each shift a five-minute huddle on what is newly arriving from the agent. The key skill is the warm handover: reading the agent's summary so the customer does not repeat themselves, and checking which facts were stated by the customer, which are declared, versus which came from the account, which is sealed.

Quality analysts. If the agent now resolves a new request type, the scorecard for that type must change. Scoring human advisers on contacts the agent now handles makes no sense.

The manager's own skill. Reading the record when a customer complains about the agent: which verdict, which bound, who cleared any hold. The EU AI Act's Article 50 requires people to be told when they are interacting with an AI system, so the manager should also confirm the disclosure wording survived any change.

Controls and the evidence they leave

• Fallback forecast. Evidence: the dated forecast and the rota it produced.
• Certified cover on the held queue each shift. Evidence: the rota matched to current certificates.
• Shift huddles. Evidence: attendance per shift.
• Updated scorecards. Evidence: scorecard version per cycle.
• Practicum for new approvers. Evidence: the practicum log template for contact centre managers.

What should you measure?

Three things move first after a cycle. The volume reaching human queues against forecast. The time held refunds wait for approval. And the repeat-contact rate on handovers, which shows whether advisers are reading the agent's summary. Complaints that mention the agent by type are the fourth; track them weekly for a month after each change.

Where does it go wrong?

The forecast ignores the change. If the agent stops resolving a request type, human queues grow and service levels fall the same week.

The held queue has no certified person on nights. Refunds wait until morning, and customers call back. Whether people who clear these need a certificate is a business choice, set out in do contact centre managers need a certificate to approve AI agent actions.

Accountability is unclear when the agent gets it wrong. The manager answers for the operation, not for each sentence, as are contact centre managers accountable for an AI agent's mistakes explains. Risk managers run a parallel cycle; see keeping risk managers ready for recertification.

Where the Academy fits

Pryme Intelligence Academy's Operator track is built for team leads and operations managers who run a live workspace. Modules on reading the workspace, usage and billing, and running the rhythm map directly onto forecasting and the held queue. For teams, seats come with paid Pryme Intelligence plans, and the admin view shows each person's progress, results and certificate expiry. More in the guide to AI literacy and certification for people who work with agents.