Lean, but not light. In a small business, good training means everyone learns the basics in an afternoon, at least two people are certified to approve the agent's held actions so there is always cover, the owner learns to decide depth, and most of the practice happens in the real review queue rather than in courses.
Why is a small business different?
One person wears several hats. The owner may also be the finance lead, the person who sets up the agent, and the one who approves its payment drafts. That is efficient until it is dangerous: the person who designed an agent is the worst-placed to spot its blind spots, and a single trained person is a single point of failure the week they are ill or on holiday.
Small businesses also have an advantage. They can train everyone at once, see everyone's gaps in one view, and change practice the next morning. Good training uses the advantage and guards against the weakness.
What does the lean version include?
| Element | Who | What good looks like |
|---|---|---|
| Shared vocabulary | Everyone | Depth, the four verdicts, provenance, lapse, the record, in a few hours |
| Reviewer certification | At least two people | Assessed on real held decisions, not just a course |
| Depth decisions | The owner | Can explain why each capability sits at its depth, and what evidence would raise it |
| Agent design | One person, or nobody if you use pre-trained agents | Builds and reads certification results; is not the only reviewer of that agent |
| Calibration | The two reviewers | Decide the same few held cases monthly and compare |
| Renewal | The certified people | Planned, with dates in a shared calendar |
The second row is the one small businesses most often skip. "Two" is not arbitrary: with one reviewer, either the queue stops when they are away, or someone untrained approves in their place.
How do you keep it lean without cutting corners?
Start with pre-trained agents where you can. If the agent was trained for a common job and certified on your cases, you need fewer design skills in-house. You still need reviewers.
Keep depth low for longer. A small business can run an agent at depth 2, drafting while a person sends, for as long as it likes. Raise a capability to depth 3 only when your two reviewers have practised on its drafts and the agent has evidence for it. Depth is earned, not configured; the plan you pay for sets a ceiling, but evidence from certification sets the actual depth.
Make the queue the classroom. Every held action is a case with a right answer. Discuss the hard ones at a weekly check-in; ten minutes is enough. The idea is developed in the review queue is a classroom, if you run it like one.
Separate building from approving. If the owner designs the agent, someone else should be the first reviewer of its drafts. A fresh pair of eyes catches what the designer assumed.
How do you work out what each person needs?
A short training needs analysis: list each person, the decisions they make with the agent, and the skills those decisions need. In a business of ten, this fits on one page. What a training needs analysis for AI agents looks like shows the format. Treat the agent's arrival the way you would a new hire's, with a defined first month and someone responsible for it; how to onboard an AI agent like a new hire covers that side.
An example
Take a hypothetical online shop with 12 staff. The owner, the operations manager and two customer service staff are the people who will touch the agent, which drafts refund decisions and supplier payment proposals. All 12 take the vocabulary course in one afternoon. The operations manager and one customer service lead take reviewer certification while the agent drafts in shadow for three weeks. The owner, who set up the agent with a supplier's help, approves nothing in the first month and instead reads the record each Friday. After a month, refunds under a small value on returned items move to depth 3. Supplier payments stay at depth 2, and both reviewers must agree on any change of supplier bank details.
A fuller first-deployment plan is in what a good AI agent training plan for a first deployment is. Pryme Intelligence Academy's free readiness assessment covers teams of up to 25, which includes most small businesses whole.
Questions people ask
Do small businesses need training more than large ones?
In some ways, yes: each person covers more ground and there is less redundancy. The argument is made in small teams need AI agent training more than large ones.
Can the owner be the only reviewer?
Only for a short time and at depth 2. A second trained reviewer should be in place before anything moves to depth 3.
How much time does it take each month?
After certification, a calibration session and a weekly look at hard cases. The guide to training teams for AI agents explains why that habit matters more than the course.