The Kirkpatrick model is an evaluation frame that separates participant reaction, learning, behaviour change at work and organisational result. Its practical value lies not in the hierarchy of levels but in the fact that each level demands a different kind of evidence — and that evidence from a lower level never substitutes for a higher one.
Quick Overview
What you’ll learn:
- What a post-course feedback form actually captures, and what it never will
- How to evidence behaviour change at work without relying on self-report
- Why a claim about business results needs more than a before-and-after comparison
- Who has to own the measurement for the measurement to happen at all
Who this article is for:
- People accountable for training evaluation and reporting on effects
- Internal trainers who design sessions together with their checks
- Line managers expected to observe the change inside a team
Reading time: 8 minutes
Where the Model Comes From, and What It Is
The frame set out at kirkpatrickpartners.com by the organisation that stewards the model today arranges evaluation around levels of reaction, learning, behaviour and results. In training practice it is used mostly at the first level, because that is where evidence is cheapest to collect — not because that is where it says the most.
The review of programme evaluation models published in Medical Teacher and indexed on pubmed.ncbi.nlm.nih.gov places the Kirkpatrick model alongside other approaches — goal-oriented, decision-oriented and context-oriented among them — and reads it as a possible frame to be chosen against the question the evaluation has to answer. That caveat is useful: the model is not a measurement procedure, only a taxonomy of evidence. The instruments have to be selected separately.
A study published in Evaluation and Program Planning and indexed on pubmed.ncbi.nlm.nih.gov examines the hierarchical assumption itself, testing how the levels relate to each other among private commercial banks rather than taking the ladder for granted. Its authors report that the hierarchical relationships held in that setting — reactions relating to learning, learning to behaviour change at work — which is a finding about one sector, not a general warrant to assume the chain elsewhere.
The Levels and What Each One Can Evidence
Reaction. A feedback form captures reception: relevance of the topic, quality of facilitation, logistics. It is useful for correcting delivery and useless as evidence of effect — satisfaction and learning diverge routinely.
Learning. The evidence is a check of knowledge or skill: a practical task, a knowledge test, a reproduced procedure. Credibility rises when the check also runs before the session, because then the gain is visible rather than the entry level.
Behaviour. The evidence is observation at work or a trace in a system: a new step in a procedure, a changed way of describing tickets, a different review cadence. Asking participants whether they apply what they learned measures memory of the course, not behaviour.
Results. The evidence is a process indicator that was named in advance and whose owner agreed to track it. This is where a causal claim is hardest, because a process indicator moves under many influences that have nothing to do with the course.
Why the Results Level Usually Stays a Declaration
The reason is not laziness in the learning team but the construction of the evidence. To attribute a movement in an indicator to a course you need a reference point — a group that did not attend, or at minimum a recorded prior state alongside a list of other events in that period. Without it, a before-and-after comparison describes the passage of time rather than the training.
The second difficulty is organisational. A process indicator belongs to the process owner, not to the learning function, so measuring it requires their agreement and participation. When that conversation happens after the course, the answer is usually that the data is not collected in that breakdown.
This is where work-environment conditions return. The meta-analysis in Human Factors, indexed on pubmed.ncbi.nlm.nih.gov, shows that sustaining a trained skill depends on what happens after the course — supervisor support, a genuine opportunity to apply the skill, a rhythm of reinforcement. The implication for measurement is direct: an absence of change at the behaviour level can be information about the work environment rather than about the quality of the session. Separating those two causes is a condition of an honest report.
The boundary of a defensible claim deserves naming too. “After the course, the number of errors in tickets fell” is a description, provided the prior state is recorded and other process changes are listed. “The course will improve ticket quality” is a promise — and attaching a figure to it does not turn it into evidence.
Designing the Measurement So That It Happens
Measurement is designed alongside the training, not after it. The minimum set: a named target behaviour, how it will be observed, who observes it, a review date, and a process indicator together with its owner. The roles distribute naturally: the trainer owns the learning-level check, the line manager owns behavioural observation, the process owner owns the indicator, and the training sponsor owns the fact that the review actually takes place.
A pre-course check is the cheapest improvement available to the whole arrangement — the same instrument is applied before and after the session, and the result is a gain rather than a bare level. At the behaviour level the equivalent move is recording the baseline: what the procedure looks like today, before anyone attends a session.
If you are planning measurement as part of the budget, it becomes a single conversation — the measures go into the annual plan alongside the cost lines, as set out in planning an annual training budget.
Read Also
- Neuroplasticity — how the adult brain shapes habits and learning — why behaviour change needs time and repetition
- Blended learning vs fully remote IT training: when to choose which — how format constrains what can later be measured
- Competency matrix (skill matrix) — a guide for managers — the baseline instrument for the learning level
Build Your Skills
This topic is developed in the Train the Trainer Academy and Motivational appraisal interview — feedback and employee development courses. The first covers designing sessions together with their checks; the second builds the conversation in which behavioural observation turns into an agreed step. Contact us to match the scope to your role.
Frequently Asked Questions (FAQ)
Which assessments belong to the Kirkpatrick model?
The model arranges assessment into kinds of evidence: how participants received the session, a check of knowledge or skill, observation of behaviour at work, and a process or organisational indicator. Each kind needs a different instrument, and a lower kind does not substitute for a higher one — a satisfaction form is not evidence of behaviour change.
How does the newer version of the model differ from the original?
The original arrangement described the levels as successive stages of evaluation after a course. The newer treatment described at kirkpatrickpartners.com starts from the results end: the expected outcome and the behaviours leading to it are agreed first, and only then are the session and its checks designed. The change is in the order of planning, not in the count of levels.
How do you measure training effectiveness without a full research project?
By choosing a single target behaviour and a single process indicator, with a recorded baseline and a review date. Such a measurement does not yield a causal claim, but it does yield an honest description — provided other changes from the same period are listed. That is the difference between a report that survives scrutiny and a number without context.
Can soft-skills training be measured at all?
Behaviour can be measured; the competence itself cannot. For a feedback course the observable trace is whether development conversations happen, whether they close on an agreed step, and whether that step returns at the following meeting. Such a measurement needs a recorded baseline and a manager who observes — and it yields no causal claim about company results, which the report has to say outright.