How to run a two-week practicum for contact centre managers

A contact centre manager's practicum should test the decisions only managers make: escalated refunds, complaints about the agent, held-queue cover when the agent's depth changes, and what to do with a poor quality sample. Run ten working days on live decisions, with a certified peer deciding each independently. The bar is at least 95 in every 100 right and no missed red line.

What is the practicum, and who owns it?

Advisers and team leads clear most of the agent's held work. Managers see what rises above them, and they make operational calls that shape the whole floor: when to activate the fallback rota, whether a refund outside policy is justified, how to respond when a customer complains that the agent misled them. The practicum checks those calls on real days.

The head of customer operations owns the practicum. A peer manager with a current certificate, ideally from another site or shift, reviews. The customer operations director sets the business's own limits and the red lines. The manager owns their decisions and reasons.

How do you build the queue?

Manager decisions are fewer and heavier. Around ten a day gives 100 over the fortnight. Include:

• refunds and credits escalated above team leads, held at depth 4;
• complaints about something the agent said or did, where the manager reads the record and decides the response;
• quality samples of agent-resolved conversations flagged by analysts, where the manager decides on remediation;
• cover decisions: a shift without enough certified approvers for the held queue;
• requests from team leads to change the agent's scope or handover triggers, to support, reject or send to the builder;
• disclosure wording changes for the agent's opening message.

Seed anonymised past items for anything thin. For example, a seeded complaint might say the agent told a customer their refund was approved when it was in fact held. The manager should read the record, find the verdict and the exact wording, apologise, clear or decline the refund on its merits, and send the wording issue to the builder. Offering a goodwill credit without reading the record is the wrong order.

The steps

1. Person (head of customer operations): brief the manager on the bar, red lines and log.
2. Agent, depth 1: route items to the practicum queue, tagged by type.
3. Agent, depth 2: present each item with the record extract, the deciding bound and, for complaints, the full conversation.
4. Person (manager): decide, with a reason.
5. Person (peer reviewer): decide independently, then compare.
6. Person (peer reviewer): countersign refunds above limit, scope requests and disclosure changes during the practicum.
7. Agent, depth 1: log both decisions, the match and any red-line flag.
8. Person (peer reviewer, then assessor): review the log at the end.

How do you set and apply the bar?

A decision is correct if it matches the peer's or the peer agrees it was sound. For complaints, correctness includes reading the record before responding.

The red lines: approving a refund outside the business's policy and the manager's authority; leaving a shift's held queue without a certified approver; changing or removing the disclosure that customers are talking to an AI agent; closing a complaint without addressing what the agent did. The EU AI Act's Article 50 requires people to be told when they are interacting with an AI system, which is why the disclosure is a red line. One miss fails the practicum.

Cover decisions are judged on outcome and timing. If the manager spots a night shift with one certified approver at four in the afternoon and fixes it before the shift starts, that is correct. If the gap is found when held refunds are already waiting, the decision is logged as late, even if the fix was right.

Which controls matter, and what evidence do they leave?

• Peer countersignature. Evidence: both names on each countersigned item in the append-only record.
• Independent peer decision. Evidence: logged before comparison.
• Rota changes. Evidence: the rota version after each cover decision.
• Daily debrief. Evidence: a line per mismatch.
• Full log. Evidence: the practicum log template for contact centre managers.

Pryme Intelligence Academy's Certified Pryme Operator uses the same practicum bar, signed off by a Pryme assessor.

What should you measure?

Correct decisions out of the total, by type. Red-line misses (zero). Complaints where the manager read the record before responding. And cover gaps found and closed within the shift. The last two say more about readiness than the score alone.

Where does it usually go wrong?

The practicum runs in a quiet month with no complaints about the agent. Seed them.

The peer reviewer is the manager's own deputy. Agreement then means little; choose someone outside the line.

The manager fixes complaints with credits rather than causes. Every complaint about the agent should end with either "the agent was right" or a change request.

Approval rights follow the job title. Whether managers need a certificate to approve is the business's choice; see do contact centre managers need a certificate to approve AI agent actions and, on responsibility, are contact centre managers accountable for an AI agent's mistakes. Operations managers run a similar practicum; compare running a practicum for operations managers and the guide to AI literacy and certification for people who work with agents.