How to Measure Training ROI: The Kirkpatrick and Phillips Models Explained

Why is measuring training ROI so hard?

Training is one of the few business investments that is routinely approved without a forecast of return and rarely evaluated afterwards. Part of the reason is genuine difficulty. The effect of a programme shows up weeks or months later, in the behaviour of people working in a system full of other influences: new targets, new managers, a new product, a good or bad quarter. Separating what training caused from everything else takes deliberate effort.

The other part of the reason is habit. Many learning teams measure what is easy to count (attendance, completion rates, satisfaction scores) and report those as success. Finance and senior leaders are not persuaded, and when budgets tighten, training is among the first lines to be cut. A credible evaluation approach changes that conversation. It shows which programmes work, which do not, and gives the learning and development function a seat at the table where investment decisions are made.

Two models dominate the field. Donald Kirkpatrick’s four levels, first published in a series of articles in 1959, give the structure. Jack Phillips’ ROI Methodology, developed from the late 1970s onwards, adds a fifth level and a disciplined process for turning results into a financial return. Most serious evaluation work uses both.

What are the Kirkpatrick four levels?

Each level asks a different question, and each is harder to measure and more valuable than the one before.

The most common mistake is to stop at Level 1 or 2. A programme can score brilliantly on satisfaction and knowledge tests and still change nothing at work, because the environment does not support the new behaviour. Level 3 is where most learning fails, and it is the level that tells you whether the programme was worth running.

In later work, James and Wendy Kirkpatrick reframed the model as the “New World Kirkpatrick Model”. Its key shift is to plan in reverse: start by agreeing the Level 4 result with stakeholders, identify the critical behaviours at Level 3 that will drive it, and then design learning and reinforcement to produce those behaviours. They also emphasise “required drivers” at Level 3: the coaching, follow-up, job aids and accountability that make new behaviour stick after the session ends.

What does the Phillips ROI Methodology add?

Phillips kept Kirkpatrick’s four levels and added a fifth: return on investment. More importantly, he added a process and a set of conservative rules that make the number credible to finance. The core elements are:

• Isolating the effects of training. Before claiming a result, separate the improvement caused by the programme from improvement caused by anything else.
• Converting data to money. Turn the business improvement into a financial value using standard values (the cost of an error, the margin on a sale, the cost of replacing an employee who leaves) or expert estimates.
• Tabulating fully loaded costs. Count every cost of the programme, including participants’ time away from work.
• Calculating ROI. Compare net benefits with costs.
• Reporting intangible benefits. Improvements that cannot credibly be turned into money (engagement, teamwork, brand) are reported alongside the ROI, not forced into it.

The two formulas you need are simple:

Benefit-cost ratio (BCR) = programme benefits ÷ programme costs
ROI % = (programme benefits − programme costs) ÷ programme costs × 100

A BCR of 2.5 means every £1 or $1 invested returned 2.50 in benefits. The equivalent ROI is 150%, because ROI uses net benefits. Mixing the two up is a common source of confusion in board reports, so always state which one you are quoting.

How do you isolate the effect of training?

This is the step that separates a credible evaluation from wishful thinking. If sales rose 12% after a sales programme, but a price cut and a new marketing campaign launched the same month, the programme cannot take credit for the whole 12%. The main isolation techniques, from strongest to weakest, are:

1. Control groups. Compare a trained group with a similar group that has not yet been trained. The difference in improvement between the two is attributed to training. Staggered rollouts make this practical: the second wave acts as the control for the first.
2. Trend line analysis. Project where the metric was heading before the programme, and attribute only the improvement above that trend. Works when the metric was stable and nothing else major changed.
3. Forecasting models. Where the relationship between a metric and its drivers is well understood, use a model to estimate the expected value and compare with the actual.
4. Adjusted estimates. Ask participants and their managers what share of the improvement they attribute to the programme, and how confident they are in that estimate. Multiply the two. If a manager says 60% of the improvement came from training and is 80% confident, attribute 48% (0.6 × 0.8). This is the weakest method, but with the confidence adjustment it is far better than claiming everything.

What does a full ROI calculation look like?

Here is a worked example for a customer service programme. Forty agents attended a two-day course on first-contact resolution, followed by six weeks of coaching from team leaders.

Step 1: Fully loaded costs.

Step 2: Measure and isolate the result. Repeat calls fell in both the trained team and a comparison team that would be trained the following quarter. The trained team’s improvement exceeded the comparison team’s by an amount equal to roughly 4,300 fewer repeat calls over the following twelve months.

Step 3: Convert to money. Finance’s standard cost of handling a call, including staff time and systems, is $35. 4,300 × $35 = $150,500, which we round down to $150,000 to stay conservative. Following Phillips’ guidance for short programmes, only first-year benefits are counted.

Step 4: Calculate. BCR = $150,000 ÷ $60,000 = 2.5. ROI = ($150,000 − $60,000) ÷ $60,000 × 100 = 150%.

Step 5: Report intangibles. Customer satisfaction scores and agent confidence both rose. Because they could not be converted to money credibly, they are reported alongside the ROI rather than added to it.

Notice how much of the cost was participants’ time. Leaving it out would have nearly doubled the apparent ROI, and any finance director reviewing the numbers would have caught it.

When is a full ROI study worth doing?

Not every programme justifies the effort. A full Level 5 study takes time and money, and running one on a one-hour compliance module wastes both. A sensible rule of thumb is to evaluate every programme at Level 1, most at Level 2, important ones at Level 3, and only a small selection at Levels 4 and 5. Choose those for ROI analysis when the programme:

• is expensive or reaches a large number of people;
• is linked to a strategic objective that senior leaders care about;
• will run for several years, so the evaluation will inform future investment;
• is controversial, with leaders questioning whether it is worth the money;
• has a clear link to a business metric that is already measured.

Large leadership development programmes, sales capability programmes and major reskilling initiatives are the usual candidates.

How do you make Level 3 behaviour change actually happen?

Evaluation is not only about measuring; it should improve results. The biggest lever is what happens after the session. Research on learning transfer consistently points to the same factors: a manager who discusses the learning before and after the programme, an early opportunity to apply new skills on real work, reminders and practice spaced over time, and accountability for using the new approach.

Practical ways to build that in include pre-programme conversations between participants and their managers, action plans reviewed at 30 and 60 days, short microlearning reinforcement in the weeks after, peer practice groups and coaching from team leaders. These cost little compared with the programme itself, and they are what move the result at Levels 3 and 4.

How should you report training evaluation to leadership?

Executives want three things: did it work, what did it cost, and should we do more of it? A one-page summary that answers those questions is more effective than a detailed evaluation report. Lead with the business result and the ROI or BCR, state the isolation method in one sentence so readers trust the number, show costs in full, list the intangible benefits, and end with a recommendation: scale, adjust or stop.

Reporting a programme that did not deliver, and recommending it be stopped, does more for the learning function’s credibility than any number of positive reports. It shows the evaluation is honest. For the steps that come before evaluation, see the guide to training needs analysis, and for the full cycle visit the Learning & Development hub.

Frequently Asked Questions