breed — digital learning experts
breed — digital learning experts

E-learning agency based in Munich. Award-winning since 2014.

The Kirkpatrick Model in e-learning

Completing training is not proof of impact. The Kirkpatrick Model helps you evaluate the learning experience, learning outcomes and application at work.

Tobias BertenbreiterUpdated: November 17, 2026
Abstract light trails in blue and red

At a glance

The Kirkpatrick Model distinguishes four levels of training evaluation: reaction, learning, behaviour and results. For e-learning, completing a course is only a starting point. Whether people have learned, apply what they learned at work and contribute to the intended results requires different evidence at each level.

Four levels, four different questions

The levels help us ask the right questions. They do not form an automatic chain of impact: positive ratings do not demonstrate learning, and passing a test does not prove transfer. The current explanation from Kirkpatrick Partners describes an interconnected system and recommends starting with the intended results.

LevelKey questionAn e-learning example
1 · ReactionHow relevant and useful do participants find the experience?Brief feedback on practical relevance, usability and perceived value.
2 · LearningWhat do participants know or know how to do after training?A realistic task tied to a defined learning objective.
3 · BehaviourWhat actually changes in the workplace?An observation or work sample several weeks after training.
4 · ResultsWhich intended outcomes does the initiative support?An appropriate measure, such as the error rate in a trained process.

Keep the evidence distinct. An LMS can record course completion. That tells us a person worked through the course. Appropriate tasks and real-world application are needed to show what they can now do.

Evaluation starts before production

If success criteria are considered only after launch, there is often no baseline for comparison. We therefore plan evaluation alongside the learning objectives and audience. What should change in the business? Which behaviour contributes to that change? What do people need to be able to do?

For example, a service team needs to release a machine reliably after maintenance. A short final test can check understanding. Transfer depends on whether the required checks are performed during real releases. Records of those checks provide different evidence from course statistics. A lower error rate could be a relevant outcome, but new checklists, different machines or changed processes might also influence it.

  • Agree a concrete outcome and a suitable measure.
  • Describe the required behaviour and observable learning objectives.
  • Define baselines, data sources and evaluation dates.
  • Assign responsibility for observing application and analysing the evidence.
  • Decide what to change if transfer does not happen.

What digital data can and cannot tell us

Digital training makes many actions visible: completions, answers, repeated attempts and choices between different solutions. These data are useful when their meaning is clear. A long completion time might reflect careful study, interruptions or difficulty understanding. Many clicks alone do not demonstrate learning.

High test scores also need context. Does the task measure the agreed objective? Did learners simply recognise answers, or did they apply knowledge? A before-and-after comparison can reveal change, but does not establish that training alone caused it.

For levels 3 and 4, LMS data are often insufficient. We can supplement them with work samples, structured observations or relevant process measures. Data collection should be limited to what the agreed evaluation question needs.

Transfer needs support at work

Training can prepare people for a new task. Application also depends on opportunities, time, feedback and support. The introduction to the New World Kirkpatrick Model explicitly addresses these conditions.

In our service example, a checklist at the workstation, an initial supervised release and a later conversation with a manager could support application. If transfer does not happen, we look beyond the course: can the instructions be found? Do they match the actual machine? Are there conflicting requirements? The answers help us decide whether to change the training or the working environment.

Where AI can support evaluation

AI can help sort open feedback or suggest recurring themes across many responses. Interpreting the findings remains a specialist task. An automatically generated summary can miss important exceptions and needs to be checked against the original material.

Usage data alone cannot reliably tell us how well a person performs a task at work. Such predictions need a validated basis. A small, understandable dataset tied to a clear question can be more useful than a large collection of automatically calculated metrics.

Evaluating impact also means stating the limits

The model creates a shared language for objectives and evidence. It does not replace sound measurement or consideration of other influences. Not every project needs an extensive study at every level. Scope and methods should suit the importance of the learning objective and the available data.

The practical question is: what do we need to know to improve both the training and its application? Asking this early turns evaluation into more than an end-of-course survey.

Sources and editorial notes

Kirkpatrick Partners: The Kirkpatrick Model · Definition and interpretation of the four levels.

Kirkpatrick Partners: Introduction to the New World Kirkpatrick Model (PDF) · Planning, behaviour and supporting conditions.

Revised on 17 November 2026. The examples are illustrative, not measured client results. The original German publication remains at its existing address.

Tell us what you need. We’ll find the right solution.

Whether it’s a single course, a learning journey or a learning platform like notch.learn, our first call is about understanding what you need and how we can help.