Train for peak before peak arrives. An online retailer's programme should cover the customer service, fraud, trading and finance teams separately, because each approves different agent decisions. It should include a shared grounding, reviewer tracks with practice on the retailer's own orders and refunds, rehearsal of high-volume days, and sign-off that lapses and is renewed.
Which teams approve what?
Online retail agents spread fast because the work is repetitive and the volumes are high. That speed is also the risk: a mistake repeated across thousands of orders in an afternoon.
| Team | Agent work | What reviewers must learn |
|---|---|---|
| Customer service | Answers order queries, issues routine refunds within limits | When a held refund is safe, and how to explain a refusal to a customer |
| Fraud and payments | Scores orders, holds suspicious ones, drafts chargeback responses | Which signals the agent cannot see, and when to release a held order |
| Trading and merchandising | Drafts product copy, sets up promotions, proposes price changes | How promotion rules stack, and what a held price change means |
| Finance | Matches supplier invoices, prepares payments | That an external claim cannot authorise a payment |
Leadership needs the common grounding: depth, held actions, the four verdicts and the record. Enough to set limits per team.
What may the agent do, and what stays with staff?
- Depth 2: draft product descriptions, customer replies and chargeback evidence packs; a person publishes or sends.
- Depth 3: issue refunds and set up promotions within the retailer's own limits and rules.
- Depth 4: any refund above the limit, any supplier payment, any bulk price change; held for a named person unless the agent has earned that depth for that capability.
- Stays with a person: account closures for suspected abuse, goodwill gestures outside policy, and any reply to a complaint that mentions legal action.
An example: a promotion that stacked
Take a mid-sized fashion retailer whose trading agent sets up discount codes at depth 3 within set rules. A planner asks for a sitewide code of 20 off orders above 100. The agent drafts it, and because the retailer's limit says sitewide codes need a person's approval, it holds the code. The held action shows that the code would stack with a clearance discount already live on a third of the range.
A trained trading assistant reads the verdict, sees the stacking note and adds an exclusion before approving. An untrained one sees a request from a manager, a correctly formed code and an approve button. Practice built from the retailer's own past promotions teaches people to read the whole held action, not just the request.
How should practice and peak rehearsal work?
Use real orders, refunds and promotions from the last few months, anonymised where needed. Each reviewer decides a set of held cases from their own team and records a reason. Include refusals the agent made correctly, so reviewers stop overriding them out of habit.
Then rehearse a peak day. Run a practice queue at several times normal volume and watch what happens to review quality. If reviewers start approving without opening the detail, you need more certified reviewers per shift, not faster ones. Freeze any increase in agent depth during peak trading.
How long should sign-off last?
A year is a sensible maximum, with a short update whenever an agent gains depth or the returns policy changes. Agent certificates lapse after 90 days, and a run that clears no depth drops an agent to depth 1 the same day, so service teams also need to practise handling refunds by hand.
For costs, see how much AI agent training costs for online retailers. For sequencing, read how online retailers should train staff before deploying AI agents and start with a readiness assessment in an online retailer. What online retailers get wrong about AI agent training and the AI agent training plan template for online retailers complete the set, and the full method is in training teams for AI agents.
Where the Academy fits
Pryme Intelligence Academy's Operator track suits service and fraud team leads: about 14 hours of study, then two weeks of real review decisions, 95 in every 100 correct with no missed red line, signed off by a Pryme assessor. An admin view shows each person's progress and certificate expiry, so you can see before peak who is current.
Questions people ask
Do outsourced service teams need the same training?
Yes, if they approve agent work for your customers. Ask the provider to show current certificates for every reviewer on your account, checked at source.
Should the trading team be certified?
The people who approve promotions and price changes should be. A pricing error published by an agent reaches every customer at once.
What is the minimum before a first refund agent goes live?
Every service team lead certified as a reviewer, everyone else through the common grounding, and one rehearsal of what to do if the agent drops to depth 1.