Kirkpatrick Model: The 4 Levels, Real Examples and Templates
The four Kirkpatrick levels as the 2026 model defines them, real evaluations from five countries, what an LMS can and can't prove, and two free templates.
The Kirkpatrick Model is a four-level framework for evaluating training: reaction, learning, behaviour and results, which Donald Kirkpatrick introduced in 1959 and Kirkpatrick Partners revised in 2026.
- Kirkpatrick Partners says to plan backwards from the Level 4 business result, because that result decides what to measure at every other level.
- Supervisors should start checking behaviour within weeks of training, because Level 3 measurement should not wait until day 90.
- In ATD's 2019 survey of 779 talent development professionals, only 38% of organisations evaluated training at Level 4.
- A learning platform can record completion and Level 2 scores, but Levels 3 and 4 need managers' observations and business data.
Five questions, about a minute. Pick one training programme you run or are planning, and answer for that one. This is an unvalidated planning checklist. Inspect individual answers; the total cannot certify four-level coverage or causal impact.
Go from storyboard to an interactive course in minutes
Picture a manual handling course with a 100% completion rate and an average quiz score of 94%. Everyone who was asked said they liked it. Then the quarter's injury register lands on your desk, and the number of back and shoulder injuries in the warehouse has not moved. Every figure you collected was true, and none of them answered the question the business was asking. The Kirkpatrick Model is built to catch that gap. It works when you plan it backwards, starting from the business result.
I'm an instructional designer at Coursebox, where we make an AI LMS with built-in learner reporting, so I see what training teams can and cannot prove with the data their platform collects. Below I walk one warehouse course through all four levels, show real evaluations from five countries, test what our own reporting can evidence at each level, and give you two free templates.
What is the Kirkpatrick Model?
The Kirkpatrick Model is a four-level framework for evaluating training: reaction, learning, behaviour and results. Donald Kirkpatrick set it out in a four-part article series in the Journal of the ASTD in 1959 and 1960, and much of the training evaluation guidance since has been built on it.
The levels ask four different questions:
- Did people find the training relevant?
- Did they learn what it taught?
- Do they do things differently at work?
- Did the result the organisation cared about change?
Plan the evaluation backwards. Kirkpatrick Partners, the company run by the Kirkpatrick family, is clear that you plan them in reverse: "it is critical to start with Level 4: Results".
| Level | What it measures (2026 definition, shortened) | The question it answers | Typical evidence |
|---|---|---|---|
| 4. Results | Targeted organisational outcomes | Did the business result change? | Injury rates, sales, errors, customer scores, already in company systems |
| 3. Behaviour | Whether people perform the critical behaviours at work, with support | Do people do it on the job? | Manager observation, checklists, work samples, system data |
| 2. Learning | Knowledge, skills, attitude, confidence and commitment | Did they learn it? | Pre-test and post-test, practical demonstration, scenario tasks |
| 1. Reaction | Whether people find it favourable, engaging and relevant | Did they find it relevant and engaging? | Short survey, pulse checks during the course |
Case study: An intensive care unit in Egypt, where behaviour improved and results did not
A government fever hospital in Shebin El Kom trained 106 nurses and doctors in infection control and evaluated all four levels in 2024.
- Level 3: observed aseptic practice by nurses rose from 40% to 84%.
- Level 4: "no significant changes were observed" in hospital-acquired infections, mortality or costs.
- The lesson: the authors say two months of follow-up was "likely insufficient" for results to show (Antimicrobial Resistance and Infection Control, 2025). Behaviour can change well before the result does, so time your Level 4 check to the metric.
The 2026 update: the model has a new name and a new layer
If you learned the model as the "New World Kirkpatrick Model", that was the 2015 version, when James and Wendy Kirkpatrick added ideas like leading indicators and required drivers. In 2026 Kirkpatrick Partners replaced it with the "new Kirkpatrick Model", built on Vanessa Alzate's book Building a Culture of Evaluation (ATD Press, May 2026).
The main addition is the performance environment: the systems, managers, tools and incentives around the learner. Kirkpatrick Partners says it "can support or block progress at each level". In practice you evaluate the workplace as well as the training, because people cannot change how they work if the workplace stops them using the skill.
The 2026 page also says the model is "not linear. It is cyclical", so you check levels during the programme as well as after it.
Case study: Emergency care training across Nepal
Nepal's National Health Training Center evaluated its six-day Primary Emergency Care course using records from 295 graduates, with follow-up assessments of 48 graduates at 11 hospitals.
- Level 2: knowledge rose by 35.9% after training, then fell by 7.2% by follow-up.
- Level 3: follow-up included clinical skills assessments and simulations; interviews and workplace observations also showed application and peer-to-peer knowledge sharing.
- What shaped it: the hospital environment, from equipment to staffing, affected whether people could use the skills (BMC Emergency Medicine, 2026). That is the performance environment the 2026 update describes.
The four levels, worked through one course
I'll use a hypothetical manual-handling course for 120 pickers at a distribution centre, Site A. It prepares them to check loads, use handling aids and ask for help under the site's approved procedures, with the intended business result of fewer manual-handling injuries. A second site with similar work, Site B, is trained six months later, giving us a comparison for changes in the injury measure. Every number is illustrative. This example shows how a business outcome determines the behaviour, learning and experience we need to evaluate.
Following the model's own advice, I'll start at the top.
Level 4: Results
Level 4 measures whether the targeted organisational outcome changed. Kirkpatrick Partners defines it as "the degree to which targeted organizational outcomes occur as a result of an experience or initiative and performance support".
- What to measure: manual handling injuries per 100 pickers, the number the business asked about.
- Where the data lives: the injury register, owned by the work health and safety (WHS) coordinator. Most Level 4 data already exists somewhere.
- The worked numbers: in the six months before training, Site A had 4.0 injuries per 100 pickers and Site B had 3.8. In the six months after, Site A fell to 2.0, while untrained Site B fell to 3.6. Site A improved by 2.0 and Site B by 0.2: an extra reduction of 1.8 injuries per 100 pickers at the trained site. That is an observed difference, not proof that training caused it.
- What it cannot prove on its own: a new trolley, a quieter season or a new supervisor could also move the number. Check earlier trends at both sites and changes in staffing, workload, equipment and reporting. Use the same measurement period and comparable exposure; a comparison site strengthens the evidence only when it is a credible comparison.
This calculation is a difference-in-differences: compare the change at the trained site with the change at the untrained one. The arithmetic is straightforward. A causal interpretation assumes the sites would have followed similar trends without training, with no other change affecting one site differently. With infrequent injuries and only two sites, the result may also reflect chance. Use this example to plan your evidence, and get analytical help before making a consequential causal claim.
Case study: Compliance education at the US Federal Aviation Administration
The FAA designed five interventions to help around 7,200 regulated aviation companies follow drug and alcohol testing rules. It measured Level 4 by comparing the trend in its own compliance data before and after the interventions, a method called an interrupted time series.
- Result: sanction costs fell by 24.21% from before to after the interventions (Evaluation and Program Planning, 2019).
- The lesson: existing compliance data can support Level 4 evaluation. The researchers used interrupted time-series modelling; drawing a before-and-after trend line alone would not provide the same evidence.
Level 3: Behaviour
Level 3 measures whether people do the critical behaviours at work and are supported in doing them. Fewer organisations measure it than Levels 1 and 2, and its data helps explain whether Level 4 moved because of the training.
- Critical behaviours: checks the weight label, uses a trolley for loads over the site weight limit, asks for a team lift on awkward loads, keeps the load close without twisting, and reports pain the same shift.
- How to measure it: shift supervisors observe 10 lifts per picker against a checklist.
- When: day 14 and day 30, then monthly. Kirkpatrick Partners says behaviour measurement "cannot begin at 90 days after learning", so the old advice to wait three to six months is out of date.
- What to record alongside it: what got in the way. If pickers skip the trolley because there is only one per aisle, that is a performance environment problem, and retraining will not fix it.
A later check also helps. Five SAFE obstetric anaesthesia courses in the Republic of Congo and Madagascar were evaluated with interviews at 4 months and again at 12 to 18 months. Participants reported the same changes in teamwork and preparation at both points, and named senior support as a catalyst and its absence as a barrier (Anesthesia and Analgesia, 2019). The check-ins in the kit below ask about obstacles as well as behaviour for that reason.
A behaviour observation checklist with the warehouse example filled in, a check-in script from week two to day 90, and a summary sheet for the training team. Editable Word file, no email required.
Level 2: Learning
Level 2 measures whether people gained the knowledge, skills, attitude, confidence and commitment the job needs. Confidence and commitment are part of the current definition, so ask about them here.
- How to measure it: a short pre-test and post-test on choosing the safest way to move a load, such as a trolley, a team lift or a solo lift, plus a practical lifting demonstration scored on a rubric.
- Why the pre-test: an end-of-course quiz alone shows who passed. A pre-test shows how much each person learned.
- Confidence and commitment: two survey items at the end, such as "I am confident I can use the team-lift rule on my next shift."
A short end-of-course check tells you whether people can answer it, not how much they learned. If many first attempts score full marks, the check may have little room to show improvement. Add a pre-training assessment and a realistic applied task if you want to distinguish learning gain from an end-of-course pass.
Level 1: Reaction
Level 1 measures whether people find the training favourable, engaging and relevant. Kirkpatrick Partners singles out relevance as the item that best predicts whether behaviour will follow.
- What to ask: three items at the end. Was this relevant to your shifts? Did it hold your attention? What might stop you using it at work?
- When: at the end, plus a quick pulse check halfway through so you can fix problems while the course is running.
- What not to do: ask only whether people enjoyed it. A 1997 meta-analysis found that reactions about usefulness relate more strongly to learning and on-the-job performance than reactions about enjoyment.
How to use the Kirkpatrick Model, step by step
- Agree the Level 4 result with the person who asked for the training. One metric, its baseline, a target and who owns the data. If nobody can name a metric, the request is not ready.
- Choose a comparison. A site, team or intake trained later, or a time series of the metric going back at least a year.
- Pick three to five critical behaviours that most affect that result, and decide who will observe them and when.
- Design the learning and its assessment so it builds those behaviours, with a pre-test or baseline.
- Write the reaction questions around relevance and engagement.
- Check the performance environment before launch: will managers, tools and time allow people to use the skill?
- Collect the data, then read the levels together.
| Level | When to measure | Who collects it |
|---|---|---|
| 1. Reaction | End of the course, plus a pulse check during it | Trainer |
| 2. Learning | Before the course and at the end | Trainer or the platform |
| 3. Behaviour | From week 2, then day 30 and monthly | Managers or supervisors |
| 4. Results | On the metric's normal cycle, for long enough to see change | Data owner, such as WHS, sales or quality |
Agree the result, plan all four levels backwards, and read the results together, with the warehouse example filled in. Editable Word file, no email required.
If injuries fall but you collected no evidence of changed lifting habits, you cannot tell whether training contributed to the drop. Missing observations do not prove that behaviour stayed the same. Read Level 4 beside Level 3, record other workplace changes, and report the limits when the evidence does not line up.
What training teams ask us, and what the research says
In the support threads I reviewed, the evaluation question we get most is a version of "has this learner finished, and what did they score?" For training providers and Australian registered training organisations (RTOs), that report is evidence they hand to an auditor or a client. I screened roughly 807 threads using targeted searches; completion and progress reporting appeared across about 12 customers, while I found no explicit request to measure workplace behaviour or training ROI in that searched set. Those searches are a view of support demand, not a representative survey of our users. One college asked for a built-in comparison of pre-test and post-test scores to show learning gain.
Taken together, the training teams we hear from focus on completion and scores and rarely raise Levels 3 and 4. ATD's 2019 evaluation research, which surveyed 779 talent development professionals, shows a similar pattern: about eight in ten organisations evaluated at Levels 1 and 2.
The top barrier in that survey, named by 41%, was the difficulty of isolating the training's influence on results. The comparison-site method in the Level 4 section is the simplest answer to that I know.
I tested what Coursebox can evidence at each level
I wanted to know exactly which levels our own platform can evidence, so on 6 October 2026 I checked our reporting and built the evaluation pieces in Coursebox. Coursebox is our product, so weigh my read with that in mind.
The reports. Coursebox's reports show who is enrolled, how far each learner has got, their assessment scores and who has earned a certificate, with a CSV download for each. That covers completion and Level 2 scores. For Level 1, our help centre recommends embedding a survey tool such as SurveyMonkey in the course and reading the results there.
The Coursebox reports screen: enrolments, completions and per-course statistics.
The supervisor's piece. Level 3 depends on supervisors, so I built a short course for them rather than adding evaluation lessons to the pickers' course. I asked for a 15-minute course that prepares warehouse supervisors to run the 30-day follow-up after manual handling training: the four lifting behaviours to watch, a week 1 and week 4 floor checklist, a two-minute check-in with each picker, and a log of near misses and injuries to compare with the 30 days before training.
Coursebox built "Manual Handling Follow-Up": three sections, four pages and about 19 minutes. It covers the four behaviours to watch, running the floor check with a two-minute coaching conversation, an assignment where the supervisor practises correcting an unsafe lift, and building the follow-up record. The pickers' course stays focused on lifting, and the people who collect Level 3 evidence get their own short preparation.
| Level | What Coursebox records | What it cannot show on its own | What to add |
|---|---|---|---|
| 1. Reaction | Responses to a survey tool embedded in the course; you read the results in that tool | Whether the course felt relevant, unless you ask | A three-item survey on relevance and engagement |
| 2. Learning | Quiz and assignment scores, rubric grades, completion | Learning gain, without a baseline | A pre-test or a harder applied question |
| 3. Behaviour | Completion of a supervisor follow-up course, if you build one; the observations happen on the floor | Whether people use the skill on the job | Supervisor observations and check-ins |
| 4. Results | Not tracked in the platform; the metric lives in business systems | Whether the training moved the metric | The metric, a baseline and a comparison group |
My conclusion: use the platform for completion and Level 2, an embedded survey for Level 1, and a short supervisor course to prepare the people who observe at Level 3. An LMS may store or connect workplace evidence, but it cannot substitute for suitable observations and business measures at Levels 3 and 4.
Criticisms of the Kirkpatrick Model
The model has shaped training evaluation for more than 60 years, and it has well-documented weaknesses.
- The levels are not causally linked. In 1989, Alliger and Janak named three "problematic assumptions": that the levels rise in value, that each causes the next, and that they correlate. Good reactions do not guarantee learning, learning does not guarantee behaviour, and better behaviour does not guarantee results, as the Egypt example shows.
- It describes outcomes rather than telling you how to evaluate. Holton argued in 1996 that it "is really a taxonomy of outcomes and is flawed as an evaluation model".
- Attribution at Level 4 is hard. Kirkpatrick Partners now argues for contributive ROI, which treats training as one contributor to the result. They call it a "more credible approach than trying to isolate training as the sole cause". The Phillips ROI Methodology takes the opposite view and adds a step to isolate the training's effect.
- Level 2 lumps different kinds of learning together. Will Thalheimer's Learning-Transfer Evaluation Model (LTEM) splits learning into knowledge, decision-making and task performance across eight tiers.
These limits are reasons to read the levels together, use a comparison where you can, and say clearly what your evidence can and cannot show.
Kirkpatrick vs other evaluation models
| Model | What it adds | Best for |
|---|---|---|
| Kirkpatrick (2026) | Four levels plus the performance environment | Most programmes: a shared language with stakeholders |
| Phillips ROI | A fifth level for return on investment, with an isolation step | High-cost programmes where finance wants a percentage |
| LTEM | Eight tiers that separate knowledge, decisions and task performance | Designing better Level 2 and 3 measures |
| Brinkerhoff Success Case Method | Stories from the most and least successful learners | Finding out why training worked for some people and not others |
If you want a broader view of how to evaluate the effectiveness of a training program, or a model built around programme planning such as the CIPP model, those are worth comparing too.
In practice I use Kirkpatrick as the spine, borrow LTEM's tiers when designing assessments, and use the Brinkerhoff Success Case Method when a programme's results are mixed and I need to know why. If finance asks for a percentage return, the training ROI calculation sits on top of Level 4. And if you are designing the course itself, Fink's taxonomy of significant learning helps you set goals beyond knowledge that Levels 2 and 3 can then measure.
Build the evaluation around the result you need
The Kirkpatrick Model gives you a useful set of questions, but the four levels are not a chain of guarantees. Positive reactions do not establish learning, and a passed assessment does not tell you whether people can apply the skill in a workplace that supports it. Plan backwards from the intended result, then gather evidence at the levels that can explain what happened.
For your next programme, agree the business result and the critical behaviour before choosing the assessment or survey. Record a baseline, identify who will support the behaviour at work and plan when you can observe it. If the result changes, check exposure, other changes and the strength of the comparison before claiming an effect. The aim is an evaluation that helps you improve the programme and make a credible decision about its future.
Frequently Asked Questions
Reaction, learning, behaviour and results. Level 1 checks whether people found the training relevant, Level 2 whether they learned it, Level 3 whether they do it at work, and Level 4 whether the organisational result changed. Kirkpatrick Partners recommends planning them in reverse, from Level 4.
Yes. Kirkpatrick Partners revised it in 2026 to add the performance environment, the systems and managers around the learner, and it remains a widely used shared language for training evaluation. Its known limits, such as weak links between levels, are reasons to read the levels together rather than to drop it.
Start within the first few weeks after training. Kirkpatrick Partners says behaviour measurement cannot begin at 90 days after learning. By then, the early weeks when people most need support to build the habit have passed. Observe again at around 30 days and then monthly.
The levels are not causally linked, so good scores at one level do not guarantee the next. It describes outcomes rather than how to evaluate, Level 4 attribution is hard, and Level 2 groups different kinds of learning together.
Phillips adds a fifth level that converts results into a return-on-investment percentage and requires isolating the training's effect. Kirkpatrick Partners prefers contributive ROI, which treats training as one contributor to the result rather than the sole cause.
Carolina Martin
Carolina is a highly experienced learning designer with tech start-up experience. She is based in Sydney and previously designed courses alongside Coursebox founder Travis in a previous job. She is passionate about delivering engaging learning experiences and brings the same perspective to customer experience too. She manages our customer support documentation and product testing processes.
