Training feedback form template for AI agent courses

Most training feedback forms ask whether people enjoyed the course. This one asks whether they now decide differently. It has three parts: a short form at the end of the course with scenario questions, a 30-day follow-up for the learner, and a 30-day check for their team lead that compares what the record shows before and after.

What should a feedback form on agent training measure?

Satisfaction is easy to collect and tells you little. A reviewer can love a course and still approve a held payment on an emailed bank change. What matters is behaviour on the job: do people read the evidence behind a draft, do they recognise which bound held an action, do they refuse under pressure, do they know what to do when an agent's certificate lapses.

That is why the form leans on scenarios and on the record. A scenario question shows whether the learner can apply an idea. The record shows whether they did, in real held actions, weeks later. This turns feedback into a feedback loop: results go back into the course and into calibration, and the next group learns from the last.

It also exposes the gap between courses that teach people to use AI and courses that teach them to oversee it. The argument that prompting courses will not prepare your team for agents that act is easy to test with this form: ask the scenario questions after each kind of course and compare.

How do you use the form?

Give part A at the end of the course, on paper or online, in under ten minutes. Send part B to the learner 30 days later. Send part C to their team lead at the same time. Keep the scenario questions specific to the team's own agent; replace the examples below with cases from your own record. Collect feedback on the agent itself separately, using the approach in how to collect feedback on an AI agent's work; this form is about people.

The form template

Part A: end of course (learner)

1. Which agent and team does this course cover? [free text]
2. In one sentence, what is the difference between a held action and a refused one? [free text]
3. Scenario: the agent has drafted a payment to a supplier whose bank details changed by email yesterday. What do you do? [free text]
4. Scenario: a colleague asks you to approve a held action while their certificate has lapsed. What do you do? [free text]
5. How confident are you deciding [the team's most common held action]? [1 to 5]
6. Which part of the course was least clear? [free text]

Part B: 30 days later (learner)

7. Name one decision you made differently because of the training. [free text]
8. Name one situation the training did not prepare you for. [free text]
9. How confident are you now deciding [the same held action]? [1 to 5]
10. Have you used the incident procedure? If so, did it work as taught? [yes / no / not used, with comment]

Part C: 30 days later (team lead, from the record)

11. Held actions this person decided in the period. [number]
12. Of these, how many were later overturned? [number, before and after training]
13. Any missed red lines? [yes / no, with detail]
14. Did their decisions match the team in the last calibration session? [yes / partly / no]
15. One coaching point for this person. [free text]

Worked example

An illustrative hospital trains 30 patient services coordinators to review an agent that books appointments and drafts replies. Part A shows most coordinators answer scenario 3 well but many miss scenario 4: they would approve for a colleague with a lapsed certificate "to help out". The course is changed the following week to cover lapsed approvers explicitly.

At 30 days, part B shows the main gap was not in the course at all. Several coordinators had to switch to draft-only mode during a system incident and did not know how; that skill now has its own short session, built on how to train a team to use draft-only mode during an incident. Part C shows overturned approvals fell for 24 of the 30 coordinators; the six whose did not are paired with a buddy. The cost side of the hospital's programme is set out in how much AI agent training costs for hospitals.

Where the Academy fits

Pryme Intelligence Academy builds part A's approach into its module tests, which are set in the learner's own industry and role. For teams, the admin view shows progress, results and certificate expiry, which supplies some of part C. Parts B and C still belong to the business, because only it can read its own record. The guide to training teams for AI agents explains how these results fit with certification.