Can contact centres use AI agents before staff are trained?

Yes, if the agent only reads and drafts. A contact centre can start an agent summarising calls, suggesting replies and preparing case notes while advisers train. Letting the agent answer customers directly, or act on accounts, should wait until the people who will review its work are trained, because a contact centre's volume turns small errors into large ones.

Why does volume change the calculation?

A contact centre may handle thousands of contacts a day across shifts. If an agent at depth 3 makes an error that a trained reviewer would catch in a sample, the same error repeats across every similar contact until someone notices. Untrained reviewers do not notice. They see a confident, well-written reply and approve it.

The other factor is turnover. Many centres hire continuously. A training plan that covers today's staff but not next month's joiners leaves the centre with reviewers who never learned how to read a held draft.

What can the agent do while people train?

• Call and chat summaries written into the case record for the adviser to check (depth 2).
• Suggested replies the adviser edits and sends (depth 2).
• Knowledge lookups from the current help articles, with citations (depth 1).
• After-contact work, such as drafting the follow-up email or the case note (depth 2).
• Routing suggestions that a supervisor confirms (depth 2).

These uses build skill as they go. An adviser who edits suggested replies all day learns where the agent is weak.

What waits for trained reviewers?

• The agent replying to customers without a person sending (depth 3).
• Refunds, credits or account changes (depth 3 and 4).
• Handling of complaints and vulnerable customers, which should be held for a person at every depth.
• Any action outside the help articles the agent was certified on.

In the EU, the AI Act's Article 50 requires that people are told when they interact with an AI system. Once the agent replies directly, scripts and disclosures must change, and advisers need to explain them.

What does a phased start look like?

Take an outsourced contact centre running customer service for an energy retailer, with 200 advisers on three shifts. Weeks one to four: the agent writes call summaries and suggests replies in chat. Team leads sample summaries daily and find the agent sometimes records a customer's promise to pay as a confirmed payment. They feed that back as a correction to the agent's brief.

During the same weeks, all advisers complete foundational training in short sessions fitted around shifts. Twenty experienced advisers and all team leads work towards operator certification. By week ten, those twenty are certified, the agent has been certified on the centre's own past chats, and the centre allows it to answer balance and payment-date questions directly at depth 3, with certified advisers reviewing a daily sample. The planning detail is in how contact centres should train staff before deploying AI agents, and the checklist for each new reviewer is the reviewer onboarding checklist for contact centre.

Which controls keep it safe early on?

• The centre's own limits cap the agent at depth 2 until certified reviewers are in place on every shift.
• Every suggested reply the adviser changes is recorded with a reason, which shows where the agent needs work.
• Team leads sample summaries and replies daily.
• New joiners do not review held drafts until trained.
• Every verdict is filed, so a customer complaint can be traced to who sent what.

Contact centres tend to repeat a few mistakes, such as training only the day shift, covered in what contact centres get wrong about AI agent training. Utilities face a similar sequence; see whether utilities can use agents before staff are trained.

Where the Academy fits

Advisers can take Pryme Intelligence Academy's free Foundations track in short lessons, read or listened to as audio, which suits shift work. Team leads and senior advisers who review held drafts take the Operator track. Its exam is 40 questions in 60 minutes, with a pass mark of 32 of 40, followed by a two-week practicum on real review decisions. The topic guide on training teams for AI agents covers sequencing across shifts.