New customer support agents joining a team where AI agents already answer routine tickets need two weeks that teach review before reply. Week one: what the AI agent resolves alone, what it drafts, what it hands over, and how to read its sources. Week two: live held drafts and hand-overs, each checked by a coach. Refund rights come last.
What does the AI agent handle, and what is left?
A note on words: here the support agent is the new person, and the AI agent is the software. In a support team with AI agents, the AI agent typically resolves routine requests alone within the business's limits, such as order status, address changes and password help (depth 3), drafts replies to more complex tickets for a person to send (depth 2), issues small refunds where it has been certified to, and hands over complaints, vulnerable customers and anything outside its bounds.
That leaves the new support agent with three kinds of work: reviewing and sending held drafts, taking over handed-over tickets with the context the AI agent gathered, and correcting the AI agent when it gets something wrong. It is more judgement and less typing than support work used to be. The support team lead owns onboarding; a certified senior adviser coaches. The team lead's own onboarding is set out in onboarding a new support team lead.
What happens each week?
- Day 1. Person (team lead): shows the AI agent's bounds: what it resolves, what it drafts, refund limits, and hand-over triggers.
- Day 1 to 2. Person (new support agent): learns to read sources. The order system is sealed. A help article is from the business's controlled knowledge base. What the customer says about their order is declared. A screenshot from the customer is external.
- Day 2 to 3. Agent, depth 2: drafts replies, which the coach reviews aloud while the new person watches.
- Day 4 to 5. Person (new support agent): reviews drafts on past tickets with the actual reply hidden, then compares.
- Week 2. Person (new support agent): reviews and sends live held drafts, each checked by the coach before sending.
- Week 2. Person (new support agent): takes hand-overs, reading the context the AI agent passed and not asking the customer to repeat it.
- End of week 2. Person (team lead): signs off solo review of drafts. Refund approval above the AI agent's limit comes after a further period.
| Step | The AI agent does | Depth | Stays with a person |
|---|---|---|---|
| Routine requests | Resolves within limits | 3 | Anything outside them |
| Complex tickets | Drafts reply with sources | 2 | Sending |
| Small refunds | Issues within certified limit | 3 | Larger refunds |
| Complaints, vulnerability | Hands over with context | none | The whole conversation |
| Corrections | Nothing; reason filed | 1 | Reason code |
What should a new support agent check in every draft?
Is the policy current? The AI agent quotes help articles. If an article is out of date, the reply will be too. A new person should know where the current policy lives.
Does it answer the question asked? Drafts can be accurate and still miss the point. Read the customer's message again before sending.
Is the tone right for this customer? A cheerful reply to someone whose order was lost for a third time reads badly. Adjust, and record why.
Should this have been handed over? If the customer mentions a complaint, distress or a reason they cannot pay, and the AI agent drafted a routine reply, take it over and record the miss.
How should corrections be recorded?
With a reason picked from a short list: wrong policy, outdated article, tone, missed hand-over, wrong order detail, or adviser judgement. Reasons let the team lead spot patterns and send them to the agent owner, who changes the AI agent's brief, sources or bounds. The new support agent should learn this on day one; a correction without a reason fixes one ticket and nothing else.
Which controls and evidence should onboarding leave?
- Checked sends in week two. Evidence: both names on each reply.
- Practice comparison. Evidence: the past tickets reviewed and differences found.
- Staged refund authority. Evidence: dated authority changes.
- Reasons on every correction. Evidence: reason codes by person.
What should the team lead measure?
Drafts the coach changed in week two, by reason. Missed hand-overs caught by the new person. Tickets reopened after the new person sent a reply. Customer satisfaction on their tickets once solo. And reason quality on corrections.
Where does it usually go wrong?
Sending without reading. Drafts look finished. Under queue pressure, a new person sends them unchanged. The checked week exists to stop this becoming a habit.
Asking customers to repeat themselves. The AI agent passes context on hand-over; new people who do not read it frustrate customers.
Refund rights too soon. Refund limits are a control; grant them after accuracy is proven.
Collections teams onboard with similar staged rights, as the plan for a new collections agent shows. For context on what support work adjoins, see whether an AI agent can onboard a business customer, and the AI agent training plan template for customer support agents sets the learning out. The wider approach is in training teams for AI agents.
Where the Academy fits
Pryme Intelligence Academy's Foundations track gives a new support agent the basics in about four hours, and its badge is free with no expiry. People who will review drafts daily continue to the Operator track, whose tests are set in their own industry and role. Lessons can be listened to as audio, which helps between shifts.
Questions people ask
Can a new support agent change the AI agent's replies before sending?
Yes, that is the job. Record a reason for each change so the pattern can be fixed.
What if the customer wants a person from the start?
Offer one. Hand-over rules should honour a request to speak to a person.
How soon can a new support agent approve refunds?
After sign-off on reviewing drafts and a further period of accurate decisions, set by the team lead.