Five weeks is realistic, among the shortest of any role. Support agents need about 18 hours of learning, then two weeks of supervised review on the live ticket queue and a final exam. High ticket volume means feedback arrives fast. The work that takes longest is learning when a polite, accurate draft is still the wrong reply.
Why support learns quickly
A support agent reviewing an AI agent's drafts may see dozens of held replies, refunds and account changes in a single shift. Each decision gets feedback quickly, from a team lead or from the customer's next message. Patterns repeat, so judgement forms in days rather than weeks.
That speed is also the trap. A queue that moves fast rewards clearing held items, and a support agent who has learned to approve drafts that read well will approve the one that refunds the wrong order. Training has to slow people down at exactly the right moments.
How the five weeks split
| Week | Stage | What the support agent practises |
|---|---|---|
| 1 | Foundations, about 4 hours | Depth, the decision chain, held, refused and withheld |
| 1 to 3 | Operator learning, about 14 hours | Human review, teaching the agent, when something goes wrong |
| 3 to 5 | Supervised live review | Real held replies, refunds and account changes, each checked |
| 5 | Exam and sign-off | A 40-question, 60-minute written exam and review sign-off |
New starters who join support after the agent is live need the same path, plus product knowledge. A full onboarding sequence is in onboarding a new customer support agent to an AI agent workspace.
What the agent does and what stays with the support agent
The AI agent reads the ticket, order history and help articles at depth 1. It drafts replies, refunds and replacement orders at depth 2. Where it has earned depth 3, it sends routine replies and issues small refunds within the business's limits.
Stays with the support agent:
- Refunds or credits above the business's limit for the agent.
- Account changes such as a new email address or delivery address on a recent order.
- Complaints, threats of legal action, and anyone who seems vulnerable.
- Anything the agent refused, if the customer pushes back.
The key reading skill is provenance. "I never received it" is declared. The courier's delivery scan is sealed. A photo emailed by the customer is external. A refund drafted on a declared claim against a sealed delivery scan should be held, and the support agent should know why before deciding.
A realistic example
Take an online retailer with a support team of 25 and an agent drafting email replies. All 25 complete Foundations in week one, in two short sittings. Operator modules run for two weeks at about seven hours a week, with a rota keeping chat covered.
From week three, every held item a support agent decides is checked the same day. Calibration on the same twenty tickets, run as in calibration sessions for customer support agents, shows that several people approved address changes requested in a chat message from an unverified contact. The agent had held them, correctly. That pattern becomes a coaching point for week four.
Team leads need longer, because they also run the rota and coach the reviewers; see how long it takes to train support team leads.
Controls that keep a fast queue safe
- Do not measure support agents on held items cleared per hour.
- Require a reason for every decline, and spot-check reasons for approvals.
- Keep address and payment-detail changes held for a person, whatever the agent's depth.
- Brief the team before any depth change, since some tickets will stop reaching them.
IT administrators take a different path, with fewer decisions but higher technical stakes; compare the IT administrator timeline. The overall programme is in training teams for AI agents.
Where the Academy fits
Support agents who review held tickets every day fit the Operator track of Pryme Intelligence Academy. Lessons can be read or listened to as audio, which helps on a shift pattern. Certified Pryme Operator needs 32 of 40 on the exam and a two-week practicum of real review decisions, at least 95 in every 100 correct, with no missed red line.
Questions people ask
Can support agents train while the AI agent is in a pilot?
Yes, and they should. A pilot at depth 2 is the best time to learn review, because every action reaches a person.
What if support agents resist reviewing the AI agent's work?
Show them their own decline reasons improving the agent's drafts. People engage when they see their judgement change the system.
Does every support agent need certification?
Every support agent who approves held actions alone should be signed off. Those who only handle escalated contacts need Foundations at least.