Training is an investment, and like any investment it needs a return that someone can point to. Evaluation models exist to answer a simple but uncomfortable question: did the program actually work? Without a framework, feedback tends to collapse into a smile sheet at the end of the day and a vague sense that people seemed to enjoy themselves.
The three names that come up most often are Kirkpatrick, Phillips, and CIPP. Each one solves a different problem. Kirkpatrick gives you a ladder of outcomes. Phillips adds a financial rung at the top. CIPP treats evaluation as a tool for improving the program while it is still running. Knowing which one to reach for, and what each one cannot tell you, makes the difference between evidence and decoration.
Why Training Evaluation Models Matter
Evaluation models help analyze the efficacy of a training program. That sounds academic, but the practical version is blunt: they tell you where to look for proof and what kind of proof counts. A model forces a team to decide in advance what success looks like, which measures will demonstrate it, and when those measures will be collected.
Without that structure, evaluation defaults to what is easiest to collect. Attendance gets counted. Satisfaction gets scored. Nothing gets linked back to performance. A model does not guarantee good evidence, but it at least makes the gap visible when the evidence is thin.
The Kirkpatrick Model: Four Levels of Evaluation
Donald Kirkpatrick developed four levels of evaluation in 1959. The framework was later redefined and updated, and it is still a widely recognized way to assess the effectiveness of training programs. The four levels are reaction, learning, behaviour, and results, and they run from the moment a session ends to the point where organisational performance either moves or does not.
Level 1: Reaction
Level 1 gauges learner satisfaction. It captures how delegates felt about the training, their personal reactions to the experience, and whether they liked and enjoyed it. Reaction data is usually gathered at the close of a session, which makes it cheap and fast to collect.
The disdain many practitioners feel toward Level 1 comes from overuse rather than from the level itself. A satisfied learner is not automatically a changed learner. Satisfaction is a starting signal, useful for spotting a broken room, a confusing handout, or a facilitator who lost the group, but it is not a verdict on whether anything was learned.
Level 2: Learning
Level 2 asks whether participants actually gained the knowledge, skill, or confidence the program promised. This is where assessment enters: knowledge checks, practical exercises, observed demonstrations, and scored simulations all give evidence that learning moved from the slide to the person.
Level 2 is often skipped because it requires designing assessment alongside content. That is a mistake. If learning did not happen, there is little point investigating whether behaviour changed, and no point at all calculating a return.
Level 3: Behaviour
Level 3 examines whether behaviour on the job changed. It looks at transfer, meaning the degree to which what was learned in the room shows up in daily work. This is the level where most programs quietly fail, because the workplace rarely rewards the new behaviour immediately.
Evidence here needs follow-up. Observation, manager input, and work samples collected weeks after the session say far more than a post-course survey. The gap between knowing and doing is the central problem of workplace learning, and Level 3 is the only place it gets confronted directly.
Level 4: Results
Level 4 connects the program to outcomes the organisation cares about, such as output, quality, retention, or cost. It is the level executives ask about and the level learning teams find hardest to prove, because many forces act on a business metric at once.
Attribution is the recurring difficulty. Results may improve after training, but improvement alone does not demonstrate that training caused it. That limitation is exactly what the next model tries to address.
| Level | What it measures | Typical evidence |
|---|---|---|
| 1. Reaction | How learners felt about the training | End-of-session feedback |
| 2. Learning | Knowledge, skill, and confidence gained | Assessments and observed practice |
| 3. Behaviour | Whether on-the-job behaviour changed | Follow-up observation and manager input |
| 4. Results | Organisational outcomes affected | Business and performance metrics |
The Phillips Model: Adding a Fifth Level
The Phillips model keeps the four Kirkpatrick levels intact and adds a fifth: return on investment. Where Level 4 asks whether results improved, Level 5 asks whether the money those results represent outweighs what the program cost.
Two ideas carry most of the weight. The first is isolation, the process of separating the effect of training from everything else that might have moved the metric. The second is conversion, which translates an operational result into a monetary value so it can be compared against program costs. Both steps involve assumptions, and honest reporting means stating those assumptions rather than burying them in a spreadsheet.
The model also insists on a chain of evidence. An ROI figure built on a weak Level 2 and an unmeasured Level 3 is arithmetic without meaning. If learners did not learn and behaviour did not shift, a favourable ratio is measuring something other than the training. Teams planning a full Level 5 study should confirm the current formula, isolation techniques, and reporting conventions with Phillips’ published material before presenting results to a finance audience.
The CIPP Model: Context, Input, Process, and Product
The CIPP model consists of four steps: context, input, process, and product. It functions as an evaluation method to assess and improve programs rather than only to judge them after the fact.
- Context looks at the need being addressed and the environment the program operates in.
- Input examines the resources, design, and approach chosen to meet that need.
- Process tracks how delivery actually unfolds.
- Product reviews the outcomes and whether they were worth the effort.
The practical difference from Kirkpatrick is timing and intent. Kirkpatrick is largely backward looking, asking what a completed program achieved. CIPP can be used while a program is still being shaped, which makes it more useful when the question is how to fix something rather than how to prove it. Readers who need the full detail of each step and its associated tools should consult the original CIPP source material.
Formative and Summative Evaluation
Formative evaluation methods provide real-time feedback about training programs. The timely feedback makes it possible for mid-course modifications to be made, which is a very different proposition from discovering in a final report that a module confused everyone.
Summative evaluation arrives at the end and asks what the program achieved overall. The two are complementary rather than competing. A program evaluated only in summative terms can be proven good or bad but cannot easily be improved, while formative data keeps the option of correction open. Models such as CIPP lean toward formative use, while Kirkpatrick and Phillips are usually applied summatively.
Comparing the Models Side by Side
| Model | Structure | Best suited to | Main limitation |
|---|---|---|---|
| Kirkpatrick | Four levels: reaction, learning, behaviour, results | Structuring an evaluation plan from day one | Weak on attribution at Level 4 |
| Phillips | Kirkpatrick plus a fifth ROI level | Defending spend and reporting to finance | Depends on assumptions at isolation and conversion |
| CIPP | Context, input, process, product | Improving a program while it runs | Less focused on monetary return |
| Formative approaches | Continuous, multi-method feedback | Mid-course correction | Does not by itself prove final outcomes |
How to Choose a Training Evaluation Model
Start with the decision the evaluation must support. If the question is whether a new induction program is worth continuing, Kirkpatrick gives a clear structure and enough evidence to argue either way. If the question is whether a costly technical program justified its budget, the Phillips fifth level forces the cost conversation into the open.
If the question is how to make a program better before the next cohort arrives, CIPP and formative methods earn their place. Many mature learning functions combine them: CIPP or formative feedback during delivery, Kirkpatrick levels for outcome reporting, and an ROI calculation reserved for the programs large enough to warrant the extra work.
A useful discipline is to decide the evaluation plan before the program launches. Choose the levels you will measure, name the evidence for each, and set the collection points. Anything measured only because it was convenient will rarely persuade anyone who controls a budget, and a model chosen after the fact tends to justify a conclusion rather than test one.
Frequently Asked Questions
What are the four levels of Kirkpatrick’s training evaluation model?
The four levels are reaction, learning, behaviour, and results. Reaction captures learner satisfaction, learning measures knowledge and skill gained, behaviour looks at whether on-the-job performance changed, and results connect the program to organisational outcomes. The framework was developed by Donald Kirkpatrick in 1959 and has since been redefined and updated.
How does the Phillips model differ from Kirkpatrick?
Phillips keeps the four Kirkpatrick levels and adds a fifth level covering return on investment. That fifth level compares the monetary value of results against program costs. Getting there requires isolating the effect of training from other factors and converting operational results into money, both of which rest on assumptions that should be stated openly.
What is the CIPP model used for?
CIPP stands for context, input, process, and product. It is an evaluation method used to assess and improve programs, including training. Because it can be applied while a program is being designed or delivered, it suits teams that want to correct course rather than only judge a finished course. Verify detailed step definitions with the original source.
Is Level 1 reaction evaluation enough on its own?
No. Level 1 gauges learner satisfaction and personal reaction, which is useful for spotting delivery problems but says nothing about whether anyone learned anything. A program judged only on happy feedback can look successful while changing no behaviour and improving no results. Reaction works best as the first rung, not the whole ladder.
What is the difference between formative and summative evaluation?
Formative evaluation provides real-time feedback during a program, which makes mid-course modifications possible while participants are still in the room. Summative evaluation happens at the end and reports on overall achievement. The two are complementary: formative data helps improve a program, while summative data helps decide whether it delivered.
