PrymeIntelligence

Training teams for AI agents·Playbook

How to onboard a new contact centre manager to an AI agent workspace

The first two weeks for a new contact centre manager where AI agents answer and draft: read queues, staff review, plan for depth drops and certificates.

Hand a new contact centre manager three things in their first two weeks: a map of which AI agents handle which contacts and at what depth, a staffing plan that covers review of held work as well as calls, and a contingency for the day an agent drops to depth 1. Everything else in the role builds on those three.

What changes for a manager when agents handle contacts?

In a contact centre that runs AI agents, agents typically answer routine messages and chat at depth 3 within the business's limits, draft replies to more complex contacts for advisers to send at depth 2, and hand over anything outside their bounds: complaints, vulnerable customers, cancellations above a value, anything the policy reserves for people. Advisers spend less time on routine contacts and more on held drafts and hand-overs.

For the manager, the workload model changes. Forecasting calls is no longer enough; they must forecast review work too, which depends on how much each agent holds. And the agent's depth can change suddenly: if an agent's re-certification clears no depth, it drops to depth 1 the same day and everything it was handling alone comes back to people. How the day shifts is described in what changes in a contact centre manager's day when AI agents arrive. The head of customer operations owns onboarding.

What does the plan cover?

  1. Day 1. Person (head of customer operations): shares the agent map: each agent, channel, depth, certificate expiry, and the hand-over rules.
  2. Day 1 to 2. Agent, depth 1: produces a summary of the last month per queue: contacts resolved by agents, held for advisers, handed over, and corrections made.
  3. Day 2 to 3. Person (new manager): sits with advisers working held drafts and hand-overs, to see what review takes in practice.
  4. Day 4 to 5. Person (new manager): rebuilds the staffing model with the workforce planner, adding review work and hand-over volume per queue.
  5. Week 2. Person (new manager): writes the depth-drop contingency: if a given agent fell to depth 1 tomorrow, how many extra advisers would the queue need, and from where?
  6. Week 2. Person (new manager): runs the first weekly correction review with team leads and the agent owner.
  7. Week 2. Person (new manager): checks adviser certificates against the queues they review and books renewals.
Step The agent does Depth Stays with a person
Routine contacts Answers within limits 3 Out-of-bounds contacts
Complex contacts Drafts replies 2 Sending
Hand-overs Holds and passes with context none Complaints, vulnerability
Monthly summary Reads the record 1 Interpreting volumes
Staffing Nothing none Manager and planner

Who on the team needs training first?

Not everyone at once. Team leads and advisers who review held drafts on the highest-depth agents come first, because their decisions carry the most consequence. Advisers who mainly take hand-overs need to understand why the agent handed over and what the customer has already been told. Workforce planners need enough to model review work. The order is set out in who in a contact centre needs AI agent training first.

What red lines does the manager carry?

Hand-over rules must work. The manager should test them in week one: a message describing a bereavement, a formal complaint, a request to speak to a person. Each should reach an adviser with context.

Customers know when an AI is answering. Where the law requires it, as under the EU AI Act's transparency rules, and wherever the business's policy does, the channel says so.

No raising depth to clear a queue. On a bad day, letting an agent send more drafts on its own is tempting. Depth is earned through Attest evidence on the business's own cases, not granted to cover a staffing gap.

Which controls and evidence should onboarding leave?

  • An agent map. Evidence: the reviewed map with depth and expiry.
  • A staffing model including review. Evidence: the model, signed off by the head of operations.
  • A depth-drop contingency. Evidence: the written plan per agent.
  • Hand-over tests. Evidence: the test contacts and where they landed.
  • Adviser certificates mapped to queues. Evidence: the list with expiry dates.

What should the manager measure from then on?

Contacts resolved by agents that customers reopened. Held drafts waiting longer than the service level. Hand-overs where the customer had to repeat themselves. Corrections per queue after changes are banked. Advisers reviewing without a current certificate. And agent certificates due in the next month, so re-certification can be scheduled away from peaks.

Where does it usually go wrong?

Forecasting only calls. Review work goes unplanned and held drafts pile up, so customers wait longer for complex answers than they did before agents.

No contingency. An agent fails re-certification during a peak and the centre has no plan for the volume that returns to people.

Correction reviews dropped. Team leads correct drafts all day, but nobody turns the pattern into changes, so the agent never improves.

Managers in neighbouring roles face the same oversight questions; see the plans for a new operations manager and a new risk manager. The AI agent training plan template for contact centre managers sets the learning out, and the wider approach is in training teams for AI agents.

Where the Academy fits

Pryme Intelligence Academy's Operator track is for people who run a live workspace daily, contact centre managers included. Modules on reading the workspace, usage and billing, and running the rhythm cover the planning side of this role. For teams, a free readiness assessment for up to 25 people shows scores by concept and the gaps, which helps a new manager decide who to train first.

Questions people ask

How much review work should a manager plan for?

Use the last month of held drafts per queue from the record, and the average time to decide one. Re-check after any change to an agent's bounds.

Should advisers know which agent drafted a reply?

Yes. The record shows it, and knowing which agent and which bound decided helps advisers review well.

What if customers prefer people?

Offer the route. Hand-over rules should honour a request to speak to a person.

Train the people who run your agents.

Pryme Intelligence Academy: free courses on the real product, audio lessons, labs in a free workspace, and certificates anyone can check on VerifyDoc.ai. Teams get seats, an admin view and a free readiness assessment.