PrymeIntelligence

Training teams for AI agents·Article

What contact centres get wrong about AI agent training

What contact centres get wrong about AI agent training: script drills, handle-time targets on review, cohorts lost to turnover, coaxed refusals.

Contact centres usually train advisers to say the right words about the agent, then measure them on how fast they clear its drafts. Turnover thins the trained cohort within months, and advisers learn to coax refusals to keep callers happy. What works instead is assessed review skill, built into onboarding and protected from handle-time targets.

Why does script training miss the point?

Contact centres are good at scripts. When an agent arrives, the instinct is to write one: what to say when a customer asks if they are talking to a machine, how to introduce the agent's reply, what to say when something is held. Advisers rehearse it and pass a call assessment.

The script covers the conversation. It does not cover the decision. When the agent drafts a goodwill credit, a refund or an account change at depth 2, the adviser who sends it is approving it. They need to see which bound applied, whether the facts came from the account system or from what the caller just said, and whether the draft is right. A caller's statement that they were overcharged is declared until checked against the billing record; the agent may draft on it, but the adviser must confirm before sending. Script training never teaches that.

What happens when reviewers are measured on handle time?

They approve. A contact centre that moves review work onto advisers and leaves average handle time as their main measure has told them, in effect, that checking the agent's draft is a cost to be cut.

The record shows the result: approval rates near total, edit rates near zero and reasons left blank. It looks like a high-performing agent. It is actually an untested one, and when the agent's certificate comes up for renewal after 90 days, the evidence from those approvals says little about whether its drafts were sound. Measure review quality separately: sample decisions, check them, and reward reviewers who catch errors. If speed targets stay, set them for the conversation, not for the approval.

How does staff turnover undo launch training?

Contact centres train in cohorts, often at launch, in a classroom block. Months later a large share of that cohort has moved on, and new starters learn the agent from whoever sits next to them.

That breaks the chain of competence quickly. Agent review skill must be part of standard onboarding, assessed before a new starter approves anything, with approval rights granted only once they pass. A contact centre can adopt the practice of tying approval rights to a current certificate, so an unassessed new joiner can handle calls and draft notes while their held approvals still go to a certified colleague. Plan the cost of that rolling training from the start; how much AI agent training costs for contact centres sets out what drives it.

Why should advisers never coax a refusal?

An agent refuses when an action falls outside its bounds: a credit above the customer's limit, a change the role does not allow, a request backed only by an external source. An adviser under pressure from an unhappy caller is tempted to find a way round. They rephrase the request, split one large credit into three small ones, or process it manually without the hold.

Each workaround defeats a control the business chose. Training must make the reason for each bound plain and the right response routine: explain to the caller that the request needs a named person, escalate it through the proper route, and record why. Team leads should review refusals with their teams, because a refusal is often the best lesson in the queue about where the business draws its lines.

What should customers be told?

Advisers need to explain held and refused actions in plain language without blaming the system or promising an outcome. "Your refund is with a colleague who approves larger amounts, and you will hear by the end of tomorrow" is better than "the computer won't let me".

There is also a legal angle in some markets. In the EU, Article 50 of the AI Act requires that people be told when they are interacting with an AI system, so where an agent replies directly to customers the disclosure must be built into the conversation, and advisers must know when it applies. This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

Where do team leaders fit?

Team leaders in contact centres are trained to coach calls. With agents, they also need to supervise the queue: spotting the adviser whose approvals never vary, running calibration where advisers decide the same held cases and compare, and staffing for the surge when an agent's run clears no depth and drops to depth 1 the same day, sending its routine work back to people. Utilities, which run large contact centres of their own, make related mistakes; what utilities get wrong about AI agent training covers them.

What should a contact centre do instead?

  1. Separate conversation skills from review skills, and assess the second on real held cases.
  2. Remove handle-time pressure from approvals; measure review quality on sampled decisions.
  3. Build agent review into onboarding and grant approval rights only after assessment.
  4. Teach the reason behind every bound so advisers stop coaxing refusals.
  5. Train team leaders to supervise the queue, not just the calls.

The detail of content is in what an AI agent training programme for contact centres should include, and the general order in the guide to training teams for AI agents. Rolling training costs money every month, so frame it as a control when you ask for it; how to get budget approval for AI agent training shows how.

Where the Academy fits

Pryme Intelligence Academy's lessons can be read or listened to as audio, which fits advisers who learn between shifts. Lessons, labs and tests are free, so onboarding each new starter adds no course fee. Reviewers and team leads take the Operator track, certified through a written exam and two weeks of real review decisions signed off by a Pryme assessor.

Questions people ask

Should every adviser be trained as a reviewer?

Only those who approve held actions. Others need the vocabulary so they can explain what is happening to callers. Separating the two keeps reviewer training focused and lets new starters take calls sooner.

How do outsourced contact centre sites fit in?

Hold them to the same standard: the same assessment, the same approval rules and the same register of who is certified. A partner site's approvals sit in your agent's record just like your own staff's.

What should a contact centre measure after training?

Sample reviewer decisions for correctness, track edit and refusal rates by adviser, and watch for anyone who never refuses. Measure the time held actions wait as well, so customers are not left waiting for a named 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.