A steering committee asks how the training will prove it worked, in numbers a sponsor will accept, and the honest answer is thin. Training evaluation models exist to make that answer solid, and the right one depends on the question your sponsor is asking, not on whichever framework the last vendor presentation used.
Kirkpatrick, Phillips, CIRO and Brinkerhoff each ask something different, and choosing the wrong one can cost a quarter spent proving a claim nobody asked for. You will see what each model asks, what deeper evaluation costs and how a team picks one. The section on how a team picks a model gives you three sponsor questions to start from.
TL;DR
Training evaluation models each answer a different question with a different kind of proof. Kirkpatrick checks reaction, learning, behavior and results, Phillips turns a result into a return on investment, CIRO checks whether a program should run before it starts, and Brinkerhoff studies the most and least successful participants instead of an average. Deeper evaluation costs more, though not in a straight line, so choose the model that answers your sponsor’s question. Name the model before the first session, budget for participants who drop out of follow-up, and build the reporting only after the model is settled.
Why Does a Model Beat a Feeling?
The numbers that are easiest to collect usually prove the least, which is why a model matters. In a published evaluation of a trauma care course, 78 of 80 participants, or 98%, said they were satisfied, a figure collected in an afternoon that says nothing about what changed.
Completion tells a similar story. A large online course on implementation research had 3,858 registrants, and about 30% earned a certificate.
The same course also gathered harder evidence, with 80.1% of surveyed respondents saying they had passed on what they learned to colleagues. A model exists to make a team collect that kind of evidence on purpose, before the survey results arrive.
| What gets collected | What it proves |
|---|---|
| Satisfaction score | People liked the session, nothing more |
| Completion rate | People finished, not whether it worked |
| Behavior evidence | Something changed after they left the room |
Collecting satisfaction data is not a failure, as long as it answers the question your program is asking. A wrong model costs the same hours as the right one, spent proving something the room already knew.
The Four Models and What They Ask
Four models cover most evaluation conversations in training and coaching, and each asks a different question at a different point in a program. The table sets them side by side before each is explained.
| Model | What it asks | When it is worth the effort |
|---|---|---|
| Kirkpatrick | Did people react well, learn, change behavior and produce results? | A program that runs more than once, where behavior change is the target |
| CIRO | Were the context and the inputs right before training started? | A costly or visible program, checked before the budget is committed |
| Phillips ROI | What did the result convert to in money, against the program’s cost? | A program a finance team will ask you to justify |
| Brinkerhoff | What did the most and least successful participants do differently? | A program with wide variation in outcomes, where an average would hide the story |
1. Kirkpatrick’s Four Levels
Kirkpatrick asks one question at four distances from the training room. Level one asks whether participants reacted well, level two whether they learned, level three whether their behavior changed and level four whether that change produced a result the organization cares about.
In the trauma care evaluation, participants’ scenario-based skill scores did not deteriorate when they were tested again six months later. Read the Kirkpatrick model in depth before you commit a whole program to all four levels.
2. CIRO Before Training Starts
CIRO works before Kirkpatrick would start, and its name gives the order of its checks: context, input, reaction and outcome. It asks whether the situation calls for training at all, and whether the design and budget fit it.
Reaction and outcome come only after those two checks. CIRO earns its cost on programs expensive or visible enough that a wrong decision outweighs a short delay, because no later evaluation can fix a program that started in the wrong place.
3. Phillips and Return on Investment
Phillips takes Kirkpatrick’s four levels and adds a fifth, return on investment, which a finance team recognizes on sight. The hard part is not the arithmetic but isolating the training’s own effect from everything else that happened in the same quarter.
Self-reports do not do that. In the online course, 66.3% of respondents felt their participation had benefited their workplace, and Phillips exists to separate that feeling from a return someone has measured.
4. Brinkerhoff and the Extremes
Brinkerhoff does not average anyone. It studies two small groups closely, the participants who did best and those who did worst, and asks what the successful group did differently.
An average can hide the real story, because ten participants with a modest average gain might be two standouts and eight who barely moved. Wide variation in outcomes is the signal to choose Brinkerhoff, while clustered outcomes suit Kirkpatrick.
When Is Going Deeper Worth It?
Every level past the first costs more effort to collect, but the cost does not rise in a straight line. A study of a road safety course run across twelve universities priced each Kirkpatrick level as a cost per point of measured effectiveness, and its figures were re-read on October 6, 2026.
Reaction cost $293.56 per point and behavior $436.79, the most expensive level. Results came in at $61.52 per point, the cheapest of the four.
Behavior is costly because it means following people back into their jobs, while results in that study came from interviewing instructors rather than retesting hundreds of students. One study cannot settle the question, but it shows that the deepest level is not always the most expensive.
The useful question is which level answers your sponsor, not how far the budget can stretch. Turning a result into a return still needs one of the models for measuring training outcomes.
How Does a Team Pick a Model?
Pick the question before the framework, and write it down before the program starts. A sponsor who cannot state the question has not decided what counts as proof.
- “Did people like it?” calls for Kirkpatrick level one.
- “Was this worth the budget?” calls for Phillips.
- “Should we run this at all?” calls for CIRO, asked before the program starts.
The decision usually involves the program owner, the L&D lead and whoever holds the budget, and it belongs in one conversation before launch. Skipping that conversation does not remove the decision, it only makes it by accident.
A team that already uses coaching tracking tools starts ahead, because a dated baseline and a record of each session already exist. A team starting from nothing pays for tracking and a model at the same time.
Budget for dropout as well. In the trauma care evaluation, only 57 of the 80 participants completed the six-month follow-up, so promising level four proof on a level one follow-up plan sets a program up to fail its own review.
Where a Model’s Job Ends
Choosing a model settles one question, what kind of proof a program needs, and it does not build the dashboard. Where the data lives, who sees it and how often is a separate decision made afterward.
Reporting cadence belongs to the L&D analytics tools a team assembles once it knows what it is measuring. A program that picks Phillips and then spends its first quarter building a dashboard has the order backward.
For coaching and mentoring programs, the evidence often lives in the software for coaches that runs the sessions. That makes the baseline, goals and session records available to whichever model the sponsor chose.
Conclusion
A committee rarely asks which model is best. It asks whether the money produced something real, and the four models answer that in different ways, from reaction and behavior to return on investment, readiness and the reasons some participants succeed.
Choose the model before the first session runs, plan a follow-up schedule your participants will realistically complete and agree with your sponsor what counts as success. For coaching and mentoring programs, Simply.Coach keeps that evidence with the engagement itself, so it is ready when the committee asks.
Key Takeaways
- Match the question. The four models measure different things, so pick by what your sponsor needs proven.
- Depth is not a straight cost line. In one study that priced all four levels, results was the cheapest to measure.
- CIRO comes first. It is the only model of the four that checks whether a program should run at all.
- Self-reported benefit is not a return. A feeling that training helped is not an ROI figure until the effect is isolated.
- Behavior outlasts a survey. A satisfaction score describes one day, while a later behavior check shows what changed.
- Follow-up loses people. Plan for participants missing later rounds and price that into level three or four.
- Tracking gives a head start. A team already recording sessions and baselines can apply any model faster.
When an evaluation shows that a program worked, the next question is where that evidence lives once the program ends. Simply.Coach keeps goals, sessions and reporting for coaching and mentoring programs in one place for program data, built for teams who answer to a sponsor.
Frequently Asked Questions (FAQs)
What are the four types of training evaluation?
Most sources use this phrase for Kirkpatrick's four levels: reaction, learning, behavior and results. Reaction covers satisfaction, learning covers what was retained, behavior covers change on the job and results covers the organizational outcome. CIRO, Phillips and Brinkerhoff sit alongside that model rather than inside it.
What are the 5 levels of training evaluation?
Phillips' model is the usual source: reaction, learning, application, impact, and return on investment. It takes Kirkpatrick's four levels and adds a fifth that converts the result into a monetary figure, and the fifth level earns its cost mainly on programs a finance function will name directly.
What are the different types of evaluation models?
Kirkpatrick, CIRO, Phillips, and Brinkerhoff are the four most widely used in training and coaching, and each asks a different question. Several others exist, including Kaufman's and the CIPP model, generally built as variations on the same core ideas of reaction, context, financial return, and individual variation.
What are the 5 basic types of evaluation?
This phrase usually refers to general program evaluation rather than training specifically: formative, process, outcome, impact, and summative evaluation. Training evaluation models like Kirkpatrick and Phillips are a specialized version built around one training program rather than a whole initiative.
What are training evaluation models?
Training evaluation models are structured ways of asking whether a training or coaching program worked, and what kind of proof counts as an answer. Kirkpatrick, CIRO, Phillips, and Brinkerhoff are the four most commonly used, and each is built to answer a different version of that question.
What is the Kirkpatrick Model of training evaluation?
Kirkpatrick asks the same question four times, at increasing distance from the training room: did people react well, learn anything, change behavior back on the job, and did that behavior produce a result the organization cares about. Most programs only need the first two or three levels.
What is the Phillips ROI Model?
Phillips takes Kirkpatrick's four levels and adds a fifth, converting the measured result into a return on investment a finance function will recognize. It earns its cost mainly on a program somebody senior will be asked to justify by name, because isolating the training's own effect is the hard part.
Which training evaluation model is best?
None of them is best in general. Kirkpatrick suits a program where behavior change is the real target, CIRO suits a costly program worth checking before it starts, Phillips suits a program that needs a financial number, and Brinkerhoff suits a program with wide variation between participants.