Training needs analysis: how to tell whether training is the right instrument

A training needs analysis is the decision about whether training is the right instrument, not a survey of what people would like to learn. It starts from a result that is not happening, works down to the behaviour that would produce it, and stops at the constraint blocking that behaviour. Only one kind of constraint is a training need.

October and November are budget months. The procedure is nearly identical in every organization we work with: HR sends managers a form, managers ask their people, people tick what sounds useful, and the plan is the sum of the ticks.

What that produces is not a training needs analysis. It is an inventory of preferences.

The difference shows up a year later. Attendance was good, room scores were high, and the result the budget was approved for is unchanged.

TL;DR

• A training needs analysis answers one question: is training the right instrument for this result? A form asking people what they would like to learn answers a different one.
• The evidence that needs analysis increases the effect of training is thin. In Winfred Arthur and colleagues' meta-analysis, of 397 effect sizes, only 22 reported any needs assessment, and the authors found no clear pattern.
• The need chain, four links: Result, Behaviour, Constraint, Instrument. Read it downward, from the result that is not happening, not upward from the topic that sounds good.
• Eduardo Salas and colleagues write that systematic needs analysis is often skipped and replaced with rudimentary questions such as “what training do you want to take?”, and that employees often cannot articulate what they really need.
• Four reasons a behaviour does not happen: the person does not know how, is not clear what is expected, has never been told, or it makes no difference. Only the first is a training need.
• A good analysis also produces the opposite verdict: Salas and colleagues state that it can show a nontraining solution to be the better alternative.

What is a training needs analysis, and what is it not?

It is the process that establishes what needs to be developed, for whom, and inside which organizational system. Salas and colleagues describe three levels: organizational analysis, job and task analysis, person analysis. What it is not is a collection of preferences. A person can say what interests them, but cannot see the organization's constraint from where they sit.

Eduardo Salas, Scott Tannenbaum, Kurt Kraiger and Kimberly Smith-Jentsch describe needs analysis as the first step in any training effort, across three components: organizational, job and task, and person analysis (Salas, Tannenbaum, Kraiger and Smith-Jentsch, Psychological Science in the Public Interest, 2012). It is a synthesis of the international literature, predominantly North American, not a study of organizations in this region.

Put simply, a training need is the distance between a result the organization requires and a behaviour that is not happening, where the reason is that the person does not yet know how to produce it. If the reason is anything else, you have a real need, but not a training need. That is the architectural error we described in why training programs fail: the program is sound, and it was bought in place of a different problem.

Why does a training needs analysis fail when it runs on a form?

Because it asks people what they would like, not what is blocking them. Salas and colleagues write that systematic analysis is often skipped and replaced with rudimentary questions such as “what training do you want to take?”, and that employees are often not able to articulate what training they really need. A form collects requests, not diagnoses.

The form does not fail because of its design. It fails because it puts the wrong question to the wrong person, in two ways that compound.

It asks about topics rather than behaviours. A topic is the name of a program, and between a topic and a result sits a behaviour. A manager who ticks assertive communication has described no behaviour, so nobody can check whether anything changed.

It asks for a diagnosis from someone without the data. A person sees their own work, not where the handover between two teams stalls, or why a priority gets rewritten three times a quarter. That is a limit of position, not of the person.

The result is a plan nobody can defend. We wrote separately about the vendor who accepts the topic and the partner who insists on the result, in capability partner or training vendor.

What does the research actually say about training needs analysis?

Less than you would expect. Winfred Arthur and colleagues meta-analysed 397 effect sizes from 162 sources, covering 33,325 participants, and found a mean training effect of 0.62 in d units. Of those 397 data points, only 22 reported that a needs assessment had been conducted, and the authors identified no clear pattern.

Worth saying plainly, because it is not what we usually hear in budget conversations. Winfred Arthur Jr., Winston Bennett Jr., Pamela Edens and Suzanne Bell published a 2003 meta-analysis of training in organizations: 397 independent effect sizes from 162 sources, 33,325 participants, and a mean effect of 0.62 in d units against control conditions (Arthur, Bennett, Edens and Bell, Journal of Applied Psychology, 2003).

The part that matters for next year's plan sits elsewhere in the same paper: the authors also coded whether each study reported the components of a needs assessment. Only 22 of the 397 data points, roughly 6%, reported any. Those subgroups held four data points or fewer, the authors asked for caution, and they identified no clear pattern. They had expected more comprehensive needs assessments to produce larger effects. The data did not.

This is easy to get wrong, so we will be exact. The figure describes what published studies reported, not how often organizations run a needs analysis, and it is not evidence that needs analysis does not work. Absence of reporting is not absence of practice.

We state our position as a position, not a research finding. We do not run a needs analysis because a meta-analysis proved it enlarges the effect of training. We run it because it is the only step that can produce the conclusion “do not buy this”.

How does the need chain work, link by link?

As a chain read downward: Result, Behaviour, Constraint, Instrument. Start from a business result that is not happening, name the observable behaviour that would produce it, establish why that behaviour is absent, and only then choose the instrument. Training is the right instrument for exactly one category of constraint.

Link 1, the Result. Something an operator already tracks: deadlines met, retention of new clients, escalations that reach the director. It has a named owner and a measure that predates the training conversation.

Link 2, the Behaviour. What would a person do differently on a Monday morning if the result were happening? An observable action, not an attitude. “Be more assertive” cannot be observed. “Decline a request that does not fit the week, on the day it arrives” can.

Link 3, the Constraint. Why is the behaviour not happening now? This is the link the form skips, and the only one that decides what you buy. Four questions, in this order:

• Do they not know how? The person does not have the skill yet. This is a training need, and the only one.
• Are they unclear what is expected? The priority, the standard or the right to decide is ambiguous. This is a Clarity need, answered with a written sentence rather than a room.
• Has nobody told them? The behaviour has been missing for months and the conversation has never happened. This is a Dialogue need.
• Does it make no difference whether they do it? The correct behaviour costs something short term and nothing in the system carries it. This is an Accountability need.

Robert Mager and Peter Pipe formulated the question that separates the first category from the other three in 1970: could the person do this if it mattered enormously? If yes, you do not have a skill problem. It is a practitioner heuristic rather than an empirical finding, and we use it as a filter.

Link 4, the Instrument. Only now is it chosen, and the choice is close to trivial, because link 3 removed most options. A skill need calls for repeated practice where the behaviour can be rehearsed. The other three call for architectural decisions, free in money and expensive in nerve. Where the constraint is individual, it gets written into the person's development plan.

How do you run a training needs analysis in a real organization?

With three short conversations and a single result on the table. One with the owner of the result, about what is not happening and how it is measured. One with two or three people doing the work, about what stops them. One with the direct manager, about what happens today when the behaviour is missing. Then write the constraint.

You do not need a whole-organization questionnaire. Pick one result, the most expensive of those not happening, and hold three thirty-minute conversations, an interval we choose by design, not for a demonstrated effect.

With the owner of the result. What is not happening, how it is measured, and since when. The question we open with: what would look different in your own reporting if this were solved?

With two or three people doing the work. Not what training they need. What stops them, on a Wednesday at three. If the answer is another department, a system or an approval, you have found the constraint and it is not a training one.

With the direct manager. What happens today when the behaviour is missing. If the answer is that nothing happens, link 3 is in your hands and the answer is not a program.

The failure pattern repeats often, and what follows is a composite of it rather than any single client. A sales team does not follow the opportunity qualification process, and the form returns a request for sales training. The people doing the work say something else: everyone knows how to qualify, but qualifying properly reduces the number reported, and that number is the only thing discussed in the weekly meeting. The constraint was not skill, it was consequence.

When the verdict is “not training”, the analysis has not failed. It produced the one output a form cannot.

What you can do before next year's budget

Do not rebuild your needs analysis process. Take one line out of the plan you are assembling now, preferably the most expensive.

Write two things beside it: the business result it should move, with an owner, and the observable behaviour that would show it had moved. If you cannot fill both boxes, you already have your answer: that line entered the plan as a topic rather than a need.

Then put Mager and Pipe's question to the people who would attend. If they could do it when it mattered enormously, change the conditions instead of buying a program.

If you want to know which of the four capabilities is the real constraint before the budget is committed, a capability diagnostic establishes that from data rather than ticks. And if you recognise the form from the opening, what we ask of HR and L&D teams describes the alternative.

So we leave you with one question. If you had to cut half of next year's training plan tomorrow, on what basis would you decide what stays?

FAQ

What is a training needs analysis?

It is the process by which an organization establishes what needs to be developed, for whom, and in which context, before it buys a program. Salas and colleagues describe three levels: organization, job and task, person. It is not a survey of learning preferences, although in practice it often is one.

How do you conduct a training needs analysis, step by step?

Pick a business result that is not happening. Name the observable behaviour that would produce it. Then establish why that behaviour is absent: they do not know how, they are unclear what is expected, nobody has told them, or it makes no difference. Only the first is a training need.

What questions do you ask in a training needs analysis?

Not “what training would be useful to you”. Ask the owner of the result what would look different in their reporting if the problem were solved. Ask the people doing the work what stops them on an ordinary day. Ask the direct manager what happens when the behaviour is missing.

How long should a training needs analysis take?

For a single result we work with a week: three thirty-minute conversations and one written page. We prefer that to a process covering the whole organization over months, because what we need at the end is a decision rather than a document.

Sources

• Arthur, W. Jr., Bennett, W. Jr., Edens, P. S. and Bell, S. T. (2003). Effectiveness of Training in Organizations: A Meta-Analysis of Design and Evaluation Features. Journal of Applied Psychology, 88(2), 234-245. 397 independent effect sizes from 162 sources, 33,325 participants. Mean effect of 0.62 in d units. Only 22 of the 397 data points (roughly 6%) reported needs assessment components; those subgroups held four data points or fewer, the authors asked for caution and identified no clear pattern. Published literature, predominantly North American.
• Salas, E., Tannenbaum, S. I., Kraiger, K. and Smith-Jentsch, K. A. (2012). The Science of Training and Development in Organizations: What Matters in Practice. Psychological Science in the Public Interest, 13(2), 74-101. Training-literature synthesis, international and predominantly North American. Needs analysis is the first step, across three components: organization, job and task, person. The authors write that systematic analysis is often skipped and replaced with rudimentary questions such as “what training do you want to take?”, that employees often cannot articulate what training they really need, and that a well-conducted analysis can show a nontraining solution to be the better alternative. The source reports no figure for how often the practice occurs.
• Mager, R. F. and Pipe, P. (1997, third edition; first edition 1970). Analyzing Performance Problems: Or, You Really Oughta Wanna. Practitioner book, no DOI. The source of the question separating skill problems from the rest. A heuristic proposed by the authors, not an empirical finding.

By Szilárd Kacsó