PrymeIntelligence

AI agent skills by role·Playbook

How to set an approver rule for the contact centre

Set an approver rule for the contact centre: who approves agent goodwill, refunds and account changes during live contacts, tied to current certificates.

Contact centre approvals happen while a customer waits, so the rule must be quick to apply. Advisers approve agent drafts and goodwill within a small limit. Team leaders approve refunds above it. Changes to contact details and payment details always need a second check outside the conversation. Every approver is named on a live rota and holds a current certificate.

What makes a contact centre approver rule different, and who owns it?

Most approver rules assume the held action can wait an hour. In a contact centre it often cannot: the customer is on the phone or in a chat, and the agent has just held a refund or an account change for approval. The rule must say who approves, within what limit, and how fast, without inventing new authority on the call. Our guide to AI agent skills by role covers what contact centre staff need to know.

The contact centre manager owns the rule, agreed with fraud and complaints leads. The agent owner configures it. How managers then check decisions is covered in how contact centre managers review an AI agent's decisions.

What are the steps?

  1. Contact centre manager: lists the agent's actions by depth. Depth 1: reads account, orders and contact history. Depth 2: drafts replies and complaint acknowledgements. Depth 3: answers routine questions, books callbacks, applies goodwill within a small limit. Depth 4: issues refunds above the limit, changes contact or payment details.
  2. Contact centre manager: assigns the small goodwill band to the agent at depth 3 if certified, and anything just above it to the adviser handling the contact.
  3. Contact centre manager: assigns refunds above the adviser band to the duty team leader, with a response-time target measured in minutes.
  4. Contact centre manager: holds every change to contact or payment details for a check through a channel the customer already has on file, approved by someone other than the adviser on the call.
  5. Contact centre manager: ties rota places to current certificates. Only certified people appear on the approver rota.
  6. Agent owner: configures the holds and files the rule.
Step The agent does Depth Stays with a person
Read account Pulls history and orders 1 Nothing
Draft reply Drafts for an adviser 2 Adviser sends
Small goodwill Applies within limit 3 Nothing within limit
Larger refund Holds for team leader 4 Duty team leader
Contact or payment details Holds for verified check 4 Second person, off the call
Complaint outcome Drafts acknowledgement only 2 Complaints handler

Why are contact detail changes held, even when the customer sounds genuine?

A change of email address or phone number is often the first step in an account takeover. Once the contact details belong to someone else, every later security check goes to them. On a live call, the caller may be persuasive and know a lot about the customer. That information is declared, not verified.

So the rule holds the change and verifies it through a channel already on file, such as a code sent to the existing number, and a second person releases it. The adviser on the call does not approve, because they are the one under pressure. The agent tells the customer plainly what happens next. This is one of the cases covered in what contact centre managers should never let an AI agent do.

Which controls make the rule work, and what evidence do they leave?

A live rota of named approvers. Evidence: the rota for each shift, with certificate status for each name.

Response-time targets per band. Evidence: time from hold to decision for each held action.

No approval of one's own call. For detail changes, the approver is not the adviser handling the contact. Evidence: approver and handler names on each change.

Goodwill within limits. The agent refuses goodwill above its bound and says which bound decided. Evidence: refused verdicts naming the limit.

Refunds to the original payment method. Evidence: payee details on each held refund, matched to the original payment.

What should you measure?

  • Customer wait time on held actions, by band and shift.
  • Detail changes approved by the adviser on the call. The target is none.
  • Goodwill paid per contact, before and after agents, to check limits are holding.
  • Refusals that led to complaints, which may show a limit set too tight.

The approver authority matrix for the contact centre puts bands and roles on one page.

Where does it usually go wrong?

The rota is out of date. Approvers on leave stay on it, held actions sit, customers hang up. Update the rota daily.

Goodwill becomes the default answer. Agents and advisers settle disputes with credits because it is fast. Track goodwill by reason.

Team leaders approve without context. The held action should show the conversation so far and the sources used. A refund approved blind is not an approval.

Credit and risk rules are copied in. Compare the approver rule for the credit team and the approver rule for the risk team: those decisions can wait for a full review; yours often cannot.

Where the Academy fits

Pryme Intelligence Academy's Operator track is designed for team leads and operations managers who run a live workspace every day, including human review and running the rhythm. A contact centre can put only people with a current Certified Pryme Operator certificate on the approver rota, and the admin view shows who is due for renewal.

Questions people ask

Can the agent apply goodwill without any approval?

Within a small limit at depth 3, if certified and if your house rules allow it. Above that limit, it holds for an adviser or team leader.

What should the customer hear while an action is held?

A plain statement that a colleague is checking and how long it usually takes. If the wait is long, offer a callback rather than keeping them on hold.

Should the agent tell customers it is an AI?

Yes. Customers should know when they are dealing with an AI system, and should be able to ask for 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.