The Kirkpatrick model, and why everyone stops at level 1
Kirkpatrick's four levels of training evaluation explained plainly, with what each level actually costs to measure: and why almost everyone stops at the first.
Donald Kirkpatrick's model, set out in the 1950s, is still the most used frame for evaluating training. It describes four levels, from easiest to hardest to measure. Its popularity comes from its clarity; in practice its use almost always stops at the first.
Level 1: reaction
What attendees thought of the course: the delivery, the pace, the perceived usefulness. This is the end-of-session sheet.
Cost to measure: near zero. Which is why it is done everywhere. Its limit is well known: it measures satisfaction, which correlates weakly with learning.
Level 2: learning
What attendees know or can do that they could not before. That means measuring before and after, on the same things.
Cost to measure: moderate. It is the first level that takes any organising: write a questionnaire, run it twice, compare. It is also where the ratio of effort to information is best, and the level most training providers could reach without changing anything else.
Level 2 is the only jump that genuinely changes the conversation with a client, and it is the cheapest of the three that remain.
Level 3: behaviour
What attendees do differently at work, weeks or months later. That means observing the job, or asking their managers.
Cost to measure: high. You have to leave the room and involve people who were not on the course. A well-built delayed evaluation gives a useful approximation: not an observation, but a retention signal.
Level 4: results
The effect on the organisation: quality, safety, revenue, turnover. The level everybody cites and almost nobody measures.
Cost to measure: very high, and often impossible to isolate. Too many variables act on a business metric at once to cleanly attribute a share to a two-day course. Claiming level 4 without a serious design is usually reconstruction after the fact.
Where to spend the effort
The practical reading is this: level 1 is already done, level 4 is out of reach in most settings, and level 3 needs resources. The largest marginal gain is at level 2, because it is the first to produce evidence rather than opinion, and it costs a questionnaire.
It is also the level you can extend: if questions are tagged to competencies, asking them again weeks later adds a retention signal to level 2. See measuring what was learned and immediate and delayed evaluation.
In practice
SkillLens is built around level 2: measure before, during and after, compared per competency. Level 1 stays useful and the product collects it too, through an end-of-session NPS.