Written by the AgileFever editorial team. Program details were checked against the live program pages in October 2026. We provide corporate AI training. The measurement ideas here apply to any provider, and we say where our programs fit.
Most training reports stop at completion: how many people attended, how many finished, how they rated the instructor. Those numbers matter, but a leader asking “did it work?” is asking about something else: did people change what they do, and did the work improve? This guide shows how to measure that without building a research project.
A measurement ladder
| Level | Question | Example for AI training | Where the data lives | When to look |
|---|---|---|---|---|
| 1. Reaction | Did people find it useful? | Post-session rating and one open comment | Survey | End of the program |
| 2. Learning | Did they learn it? | A practical task or capstone scored against a rubric | Assessment, capstone review | End of the program |
| 3. Behaviour | Are they doing it at work? | Share of pull requests or tickets that used the new practice; agents or automations shipped | Repos, ticket tool, manager check-ins | 30 to 90 days later |
| 4. Results | Did the work improve? | Lead time, cycle time, change failure rate, rework, contractor spend | Delivery tools, finance | 3 to 6 months later |
The four levels follow Donald Kirkpatrick’s model. Jack Phillips later added a fifth level that converts results into money, which is where ROI figures come from. Our companion post on how to measure corporate AI training ROI covers that step.
Delivery measures that engineering teams already have
| Measure | What it shows | Why it suits training evaluation |
|---|---|---|
| Lead time | Time from request to delivery | Falls if people remove waiting and rework |
| Cycle time | Time from start to finish of an item | Shows whether the team works faster once started |
| Throughput | Items finished per week | Shows output, though it can rise for other reasons |
| Deployment frequency | How often you release | A common proxy for delivery health in DORA research |
| Change failure rate | Share of releases that need a fix or rollback | Guards against speed that costs quality |
| Review time | Time pull requests wait for review | May change early when AI tools are adopted |
If you are new to flow measures, start with cycle time, lead time and Little’s Law. DORA is the DevOps Research and Assessment program, whose research popularised several of the delivery measures above.
Additional discounts are available. Contact us to know more.
A 90-day plan
| When | What to do | Output |
|---|---|---|
| Before training | Pick two or three delivery measures and record the last 8 to 12 weeks as a baseline | A baseline sheet |
| During | Choose a real project for each learner, so skills are used immediately | A project list with owners |
| Weeks 1 to 4 after | Check usage: who is applying what they learned | Behaviour snapshot |
| Weeks 8 to 12 after | Compare the delivery measures with the baseline, by team and work type | A short results note |
| Month 6 | Decide whether to expand, adjust or stop | A decision record |
Mistakes that hide the truth
- No baseline, so any later number has nothing to compare with.
- Measuring only attendance and calling it impact.
- Judging individuals by flow metrics. The measures describe the system, and using them to rank people makes them hide work.
- Expecting results in the first month. Behaviour changes before results do.
- Ignoring other causes. A new tool, a reorganisation or a seasonal peak can move the same numbers.
Where AgileFever fits
Our corporate programs follow four stages: assess, upskill, analyze and scale. The analyze stage is where measurement happens. If you are planning a program and want to agree the measures first, you can talk to us through the contact page. For the earlier steps, read the needs assessment framework and how to pitch an AI upskilling budget.
Questions people ask
How long before results show up?
Learning and early behaviour show within weeks. Delivery results usually take three to six months, and need a baseline to mean anything.
Can small teams measure this?
Yes. A team of five can track lead time and cycle time in a spreadsheet. Treat the numbers as a trend, not a precise score.
Should we survey learners?
Yes, but treat it as one signal. A rating tells you about the experience, not the outcome.
