Contact centre managers decide what an AI agent actually does, because they set the limits it works inside and they staff the queue where its held work waits for a person. An agent can draft a goodwill credit in seconds; whether that credit is approved well, badly or not at all depends on rotas, targets and the manager's own judgement.
Who really decides what a contact centre agent does?
The certificate on an agent says what it has proved it can do. The minimum chain says what it may do on a given action: role scope, certificate, sources, the customer's own limits and the platform guardrails, with the lowest bound winning. In a contact centre, the customer's own limits are usually written by the contact centre manager. The goodwill ceiling, the contact reasons the agent may handle, and the ones that go straight to a person are operational settings, and operations owns them.
That makes the manager the author of the bound that decides most verdicts. When an agent holds a credit of 80 because the limit is 50, the verdict names the customer's limit as the deciding bound. Someone chose 50. If that person cannot explain why, the control is an accident.
Why does staffing the review queue matter so much?
A held action is only a control if a named person decides it in time. In a contact centre the review queue competes with live calls, chats and callbacks for the same people. If the intraday plan does not reserve reviewer time, held replies age, customers chase, and a supervisor approves a stack of them between calls to clear the backlog.
This is where many contact centre deployments quietly fail. The agent did its job: it drafted, it held, it named the reason. The queue was staffed by whoever was free. The manager who builds the rota is therefore building the control. Reviewer hours belong in the workforce plan alongside handle time and service level, not in the gaps between them.
What goes wrong when handle-time targets meet review work?
Contact centres run on speed measures, and they are useful ones. But a reviewer measured on handle time learns to approve. Reading the draft, checking its sources and comparing it with the account takes longer than clicking approve, and the dashboard rewards the click.
A trained manager separates the two. Review decisions are judged on whether they were right, sampled later, not on how quickly they left the queue. A reviewer who refuses a draft because the agent quoted a superseded tariff has done more for the operation than one who cleared forty items in an hour. If the manager does not set this measure, the floor will set its own, and it will be speed.
Which workspace signals should a contact centre manager read every day?
A contact centre manager already reads a wallboard. The workspace adds a different kind of signal, and it takes practice to read it well:
- Held items by age. Anything waiting longer than the customer's patience is a staffing problem, not an agent problem.
- Refusals by reason. A rise in refusals for one contact reason often means a policy changed and the agent's sources have not caught up.
- Overrides by reviewer. One person approving what everyone else refuses is a calibration gap, and it shows up here first.
- Certificate expiry. Certificates lapse after 90 days, and a run that clears no depth drops the agent to depth 1 the same day. A manager who does not know this will see a sudden flood of drafts and call it a fault.
- Withheld actions. These are platform decisions, the same for every agent at every depth. No setting changes them, so the right response is a process for a person to do the work, not a ticket asking for the limit to be raised.
What does a contact centre manager need to learn?
Most of it is not technical. The manager needs to understand depth well enough to know what each contact reason is allowed to do: depth 1 reads the account, depth 2 drafts a reply a person sends, depth 3 executes routine changes within bounds, and depth 4 covers refunds and other actions that are hard to undo. Depth is earned, not configured, so the manager cannot simply decide an agent is ready for more.
They also need to set a limit with a reason attached, read a verdict and see which bound decided it, and run a calibration session so reviewers on different shifts approve and refuse the same things. And they need to know what is recorded. Every verdict and every human decision is filed in an append-only record, so a complaint months later can be reconstructed, including who approved the reply and what they saw at the time.
Peers face the same shift from different seats. Compare how operations managers become the real control for back-office queues, and how risk managers own the limits behind the agent.
Does this change the contact centre manager's job?
It changes the weight of it. Less time goes on chasing individual contacts and more on designing the system that decides them: the limits, the rota, the measures, the escalation path. Whether the role survives is covered in will AI agents replace contact centre managers; the short answer is that the role moves closer to the control, not away from it.
Sector matters too. An insurer's contact centre handles claims notifications and customers in distress, which is different ground from a utility handling billing disputes and supply interruptions. The briefs on training contact centre managers in insurance in Australia and in utilities in Singapore cover those markets, and the wider map of who needs what is in AI agent skills by role.
Where the Academy fits
Pryme Intelligence Academy's Operator track is written for the people who run a live workspace every day, including team leads and operations managers. Its modules cover human review, reading the workspace, when something goes wrong, and running the rhythm. Certified Pryme Operator requires a written exam and a two-week practicum of real review decisions with no missed red line, signed off by a Pryme assessor, and it is valid for 12 months.
Questions people ask
Should the contact centre manager also be a reviewer?
Sometimes, for the highest-value held items. But a manager who reviews everything becomes the bottleneck. It is better to train a small group of reviewers, calibrate them, and keep the manager on sampling and exceptions.
Can a contact centre manager raise an agent's goodwill limit alone?
They can change the customer's own limit within the plan, but the agent's depth still comes from its certificate. Raising a limit does not raise depth. If the certificate covers drafting only, a higher limit produces more drafts, not more executed credits.
How many reviewers does one contact centre agent need?
It depends on the volume held, the time each decision takes and the hours the line is open. Work it out from held volume rather than total contact volume, and plan cover for every shift the agent runs.