Can online retailers use AI agents before staff are trained?

Yes, if the agent drafts and a person sends. An online retailer can start an agent answering order questions, drafting replies and preparing returns notes while the service team trains. Issuing refunds, changing orders or cancelling accounts should wait until trained reviewers are in place, because peak trading turns one bad rule into thousands of bad outcomes in a day.

Why does peak trading matter so much?

Most online retailers see demand spike around sales events and holidays. Contact volume rises with it, and many retailers hire temporary staff to cope. That is exactly when an agent acting at depth 3 is most useful and most dangerous. An untrained seasonal reviewer facing a queue of hundreds of held refunds will approve them to clear the queue.

Peak is also when fraudsters test refund rules. A trained reviewer recognises repeat "item not received" claims to the same address. An untrained one does not.

What can the agent do from the start?

1. Order status replies drafted from the order and delivery systems (depth 2, rising to 3 once certified).
2. Returns notes, explaining the returns policy and generating the returns label request for a person to approve (depth 2).
3. Product questions answered from the current product catalogue with citations (depth 2).
4. Complaint summaries of the order history and contacts, for the person handling the complaint (depth 1).
5. Delivery exception drafts to customers when a carrier reports a delay (depth 2).

What should wait?

• Refunds and credits (depth 3), until reviewers are trained and the agent is certified on past refunds.
• Order cancellations and changes to delivery address.
• Account closures and fraud-related restrictions.
• Goodwill gestures, which are judgement calls about value and fairness.

What does a sensible start look like?

Take an online homeware retailer with a service team of 30 that doubles over the winter peak. In early autumn, it starts an agent drafting replies to order and returns questions. Permanent staff review every draft.

In week three, a team lead notices that the agent offers a full refund for items returned outside the return window, citing a clause from last season's policy that is still in the knowledge base. Staff had approved several. The old article is removed from the agent's sources, the lead runs a short session on checking the cited policy version, and the replies are corrected. Every permanent staff member completes foundational training before peak. Six senior advisers work towards operator certification. The retailer decides to keep refunds held throughout peak, with only certified advisers approving, and to revisit depth 3 afterwards. A full programme is in what an AI agent training programme for online retailers should include, and the planning detail in how online retailers should train staff before deploying AI agents.

Which controls suit a retailer?

• The retailer's own limits cap the agent at depth 2 for refunds until certified reviewers can sample its work.
• Only certified staff approve held refunds; seasonal staff handle drafts that do not move money.
• Repeat claims to one address within a set period always hold.
• Sources limited to the current policy version.
• Every verdict filed, so a chargeback or complaint can be traced.

Retailers often train for launch and then not for peak, among the patterns in what online retailers get wrong about AI agent training. The AI agent training plan template for online retailers gives a ready structure.

Where the Academy fits

Pryme Intelligence Academy's free Foundations track takes about four hours and suits temporary staff as well as permanent ones. Senior advisers who approve refunds take the Operator track, whose certificate is valid for 12 months and can be checked by anyone by number or QR code. The topic guide on training teams for AI agents covers who to train first.