Most corporate training is evaluated with a satisfaction survey handed out at the end. "The trainer was knowledgeable, the content was useful" is nice to hear, but it doesn't answer the real question: did the training change how people do their work?
This question matters even more for AI training, because the goal isn't to transfer knowledge; it's to build a new way of working.
The Kirkpatrick model: measuring at four levels
The most widely used framework for measuring training impact is Donald Kirkpatrick's four-level model. Here's how it applies to AI training:
Level 1 – Reaction: how did participants find the training?
- Satisfaction and relevance of the content to their work
- Share of participants who agree that "I can use what I learned in my job"
This level is necessary but not sufficient.
Level 2 – Learning: what did participants learn?
- Short tests before and after: safe-use rules, prompting techniques, risks
- Practical task: each participant completes a real work task with AI and the output is assessed
The pre-test also lets you tailor the programme to participants' level.
Level 3 – Behaviour: are employees applying what they learned?
The most critical and most often skipped level. Indicators to check 30, 60 and 90 days after training:
- Share of active users of approved AI tools (from enterprise tool usage reports)
- Contributions to the prompt library and number of shared use cases
- Change in shadow AI use (surveys or network monitoring data)
- Managers' observations: in which tasks has the team started using AI?
Level 4 – Results: what was the business impact?
- Time saved on targeted tasks (report writing, emails, document summaries)
- Quality: error rates, number of revisions, customer satisfaction
- Risk: number of AI-related data or policy incidents
- New opportunities: use cases proposed and implemented after the training
Sample measurement table
| Level | Metric | How to measure | When |
|---|---|---|---|
| Reaction | Satisfaction and applicability | Survey | End of training |
| Learning | Knowledge gain | Pre- and post-test | Before and after |
| Behaviour | Active usage rate | Tool usage reports | 30, 60, 90 days |
| Behaviour | Shared prompts and examples | Prompt library | 60, 90 days |
| Results | Time per task | Sample task measurement, self-reporting | 90 days |
| Results | AI-related incidents | Information security records | 90 days and ongoing |
Tips that make measurement easier
- Record a baseline. Without a pre-training measurement, there's nothing to compare against.
- Pick a few meaningful metrics. Three to five metrics tied to the goal are enough.
- Involve managers. Team managers are best placed to observe behaviour change.
- Collect stories too. Concrete examples such as "this report now takes me 30 minutes instead of 2 hours" are as persuasive as numbers.
Conclusion
The way to demonstrate the value of AI training is to treat it not as an event but as a measurable transformation programme.
All our corporate training programmes start with a pre-training level assessment and finish with post-training measurement, a report and an adoption plan for lasting use. Contact us to design a measurable programme for your organisation.
Sources