The Kirkpatrick Model: common mistakes and best practice

The Kirkpatrick model is more than sixty years old, yet it remains the reference when it comes to evaluating training. Almost all training providers know it. But that’s not the issue. The issue is that many use it incorrectly or incompletely, given their objectives, available data, and sector constraints, resulting in evaluations that cannot be used to their full potential.

The model becomes truly useful when it is adapted to the training objectives, the data you can collect, and the real-world constraints. Without taking these into account, you end up just filling out questionnaires “for form’s sake", without ever knowing whether the training actually had an impact.

The Kirkpatrick model offers four levels of evaluation, each linked to the previous one.

The first level, reaction, measures participants’ satisfaction right after the training. It is the easiest level to implement, using an immediate feedback questionnaire.

The second level, learning, checks what participants retained or whether they are able to do what was covered during the session.

The third level, behaviour, looks at whether participants actually apply what they learnt once they are back in their role.

The fourth level, results, aims to determine whether the training produces a measurable impact on the company: fewer errors; higher productivity; fewer accidents, depending on the training topic.

Many training providers want to jump straight to level 4 because that’s what clients or funders ask for. They want numbers that prove the training has a real impact on the company. The problem is that an outcome rarely depends on a single cause. If a team’s productivity increases after a training course, it may be thanks to the training, but it could also be due to new software, an organisational change, or a period of higher activity. Identifying precisely the training’s share in that result is difficult, expensive, and, with a training provider’s resources, sometimes impossible.

The best practice is not to abandon this level, but to be realistic about what you can demonstrate. You can also track indirect indicators, for example, the number of incidents before and after a safety training or compare learners’ skill levels before and after the course. That information can already be useful as it stands.

Conversely, many training providers stop at the satisfaction questionnaire at the end of the session. It’s quick, effective, easy to automate, and it produces a nice average score to display. The problem is that a satisfied learner hasn’t necessarily learnt something useful. A training course can be enriching and well delivered, with a great trainer, without enabling participants to leave with real skills. Conversely, a demanding course may be poorly rated in the moment while being highly effective later on.

Stopping at level 1 gives a false sense of quality control. You think you are evaluating the training when you are actually evaluating the session atmosphere. Setting up a simple pre- and post-training test helps, at a minimum, to verify what was truly acquired, instead of limiting yourself to satisfaction surveys.

This is the most common mistake and the easiest to correct. Many evaluations are designed generically, with standard questions that refer to no specific objective defined when the training was designed. You ask whether the training was useful, whether the pace was right, whether the trainer was competent—but never whether the initial objective, set at the start of the training, was achieved.

Moreover, training evaluation is only useful if it measures what was announced at the start. If the objective of a training course is for participants to be able to carry out a specific procedure independently, the evaluation must verify that, and it must not be limited to a general impression of satisfaction.

This link between objectives and evaluation is at the heart of quality frameworks, notably the Qualiopi certification, which requires demonstrating that the provider measures the achievement of its stated objectives.

Once you accept that a result never depends on a single cause, the question becomes more practical.

The first step is to choose the right indicators before the training even starts. A provider that waits until the end of the session to decide what to measure is starting with a problem that is hard to fix. If the training is about safety, you look at the number of incidents or accidents in the months that follow. If it is about customer relations, you can track the complaint rate or customer satisfaction over the relevant period. The indicator must exist before the training; otherwise, there is nothing to compare.

The second step is to accept working with indirect data rather than waiting for perfect proof. Few organisations have the means to isolate the exact effect of a training course amid everything else happening in a company. However, a consistent set of signals, a drop in errors that coincides with the training and feedback from managers who observe a concrete change on the ground, is more than enough to demonstrate real value, especially in a Qualiopi context where the goal is to prove a serious approach, not scientific causality.

Finally, level 4 works better when it builds on what has already been measured at level 3. If you have verified that learners are applying the right actions on the job, it becomes easier to link that observation to a business result rather than starting from scratch several months after the training. The levels should not be treated as separate boxes; they build on one another.

The Kirkpatrick model is not meant to be applied in exactly the same way to every training programme. A two-hour session on internal software does not require the same depth of evaluation as a multi-month programme for learners. Systematically imposing the four levels, with the same level of rigour, often leads to evaluation feedback that no one actually takes the time to analyse.

For example, for mandatory safety training, it is relevant to go up to level 3 by checking on the job that the right actions are being applied, because the consequences of poorly delivered training can be serious. Conversely, for a lighter training course, well-designed levels 1 and 2 can be more than enough. The key is not to do all four levels at any cost but to choose the ones that are truly useful, with a way of collecting information that the provider can sustain over time without increasing the workload.

The Kirkpatrick model is a good starting point, but it is not a checklist to apply automatically. It is truly useful when the evaluation is tied to a clear objective, when the chosen level matches the stakes of the training, and when the results are analysed and used.