Why do people know what to do but not do it?

Ask a manager why the training did not work and you rarely hear "they didn't learn it." You hear something more specific and more frustrating: "they know it, but they don't do it." The learner passed the assessment. They can explain the framework in a meeting. Put them in front of the actual customer, patient, or difficult direct report, and the framework never appears. This guide is about that person: not the program-level question of how much training decays across an organization, but the individual psychology of why someone who demonstrably knows the material still does not act on it, and what kind of learning builds the judgment and confidence that finally closes the distance.

What is the knowing-doing gap?

The phrase comes from Jeffrey Pfeffer and Robert Sutton, whose book The Knowing-Doing Gap (Harvard Business School Press, 2000) documented a pattern they found across companies: organizations routinely know what to do, can articulate it in strategy documents and training programs, and fail to do it anyway. Their subject was organizational behavior, but the pattern they named operates just as reliably one level down, inside the individual learner.

At the individual level the gap looks like this. A salesperson completes negotiation training, scores well, and then discounts on the first pushback exactly as before. A new manager can recite the feedback model and still avoids the difficult conversation for three weeks. A clinician knows the updated guideline and reverts to the familiar regimen when the case in front of them gets complicated. In every case the knowledge verifiably exists. The behavior does not follow.

It is worth separating this from the related program-level problem. Our guide on the training transfer gap covers how application decays across a whole trained population, including Saks and Belcourt's (2006) widely cited finding that training professionals estimate only 34% of employees are still applying training material a year after a program. That is the aggregate view, and it is a program design problem. This guide asks the question underneath it: what is happening inside the one person who knows and does not do? Because until that question has an answer, no amount of program redesign knows what it is aiming at.

Why doesn't knowing produce doing?

Knowledge fails to produce action because acting in a real situation demands at least four abilities that content consumption never exercises.

Recognition. In a course, the situation arrives labeled. The module is called "Handling Objections," so the learner knows every example is an objection. Real situations do not announce themselves. The objection arrives disguised as a casual question, mid-conversation, while the learner is thinking about something else. Recognizing that this live, messy moment is an instance of the thing you studied is a distinct skill, and it is built by encountering situations that do not come with labels.

Choosing under ambiguity. A quiz presents one right answer and several distractors written to be wrong. A real decision presents several plausible options, incomplete information, and tradeoffs that the textbook version never mentioned. It is not black and white; it is gray. Knowing the principle is not the same as knowing which of three reasonable-looking moves the principle favors right now. That judgment is exactly what most assessments are structurally unable to measure.

Comfort with the act itself. The first time anyone does anything new, it feels risky. If training never provides a first time, the first time happens on the job, in front of a real person, with real consequences for getting it wrong. Hesitating there is not a motivation problem; it is a rational response to being asked to debut a behavior in the highest-stakes venue available. People do what they are comfortable doing, and comfort has a precondition that no slide deck can meet: having done it before.

Competing habits. Years of repetition have made the old response automatic: it fires without deliberation, a procedure that executes on its own the moment its trigger appears. The new approach exists in a different format entirely, as a description that must be consciously retrieved and assembled into action. When the live moment arrives, the two compete for retrieval, and the contest is not close. Pressure tilts it further, because deliberate recall is exactly what cognitive load degrades first, while automatic routines run untouched. The learner has not rejected the new approach; it loses the race to a response that no longer needs their attention.

There is a tempting fallback here: give it time, experience will teach them. The evidence says otherwise. Ericsson (2008), reviewing decades of research on expert performance, notes that length of experience shows only a weak relationship with actual observed performance; more years of doing the job does not reliably produce better performance. Experience without designed feedback mostly rehearses whatever the person already does, which is precisely the competing habit the training was meant to replace.

Why do people believe they're ready when they're not?

If learners could feel the gap, they would flag it themselves. The uncomfortable finding is that they cannot, and polished content experiences worsen the miscalibration.

Deslauriers et al. (2019) demonstrated this in a randomized experiment at Harvard. Students experienced identical course materials taught two ways: polished passive lectures or active instruction that made them work through the problems themselves. Students in the passive lectures reported a significantly stronger feeling of learning. On independent tests, they had learned significantly less. Learners take the smoothness of an expert presentation, its clarity and organization, as a proxy for how well they themselves understand it. Meanwhile the strain of actually working through problems registers as something going wrong, even though that strain is what durable learning feels like from the inside. The study's authors also point to a compounding factor from earlier metacognition research: novices are poor judges of their own competence, because the knowledge needed to evaluate a judgment is the same knowledge needed to make it.

For an instructional designer this is the trap hiding inside every satisfaction score. A program built on clear, well-produced content generates confident graduates and strong course ratings, and both signals are measuring fluency, not readiness. The learner is not lying on the survey. Their internal readiness gauge is simply connected to the wrong input. Which means the moment of discovering they cannot do it gets outsourced to the job itself, the one venue with real stakes, an audience, and no guidance, and the rational response to failing there once is to quietly retire the new behavior and go back to what works.

What actually builds the doing side?

The research on how real performance develops keeps arriving at the same set of conditions. Ericsson's work on deliberate practice, the research program behind much of what we know about expert development, identified them decades ago: well-defined goals focused on a specific aspect of performance, immediate informative feedback on each attempt, and repeated opportunities to refine after reflection, with the activities designed by someone who knows the path to competence (Ericsson, 2008; Ericsson & Harwell, 2019). Ericsson and Harwell are equally clear about what does not qualify: simply engaging in the activity, doing the job as demands arise, lacks the goals and the feedback, and improvement under those conditions is unreliable (Ericsson, 2008; Ericsson & Harwell, 2019).

Translated into instructional design, the conditions become concrete:

• Decisions with consequences in realistic situations. The learner faces situations that behave like work: unlabeled, ambiguous, populated with plausible options. They commit to a choice and watch the outcome unfold. This is what exercises recognition and judgment, the two abilities a quiz cannot touch.
• Expert guidance at the moment of choice. Feedback tied to the specific decision, delivered when the decision is made, is what converts a wrong turn into learning. It is the instructional equivalent of the teacher Ericsson describes, diagnosing the error while the reasoning behind it is still warm.
• Safe failure. The learner's first wrong choices need to happen somewhere that failure costs nothing real. This is not a comfort feature; it is the mechanism. The error diagnosis Ericsson puts at the heart of practice requires an error made in earnest, which means the learner must act on their genuine default rather than a guarded guess, and they will only put that default on the table where a wrong choice costs them nothing.
• Repetition across situations. Comfort is built by the second and third encounter, not the first. Facing a similar situation again, choosing better, and seeing the improvement is where the learner's internal evidence changes from "I understand this" to "I have done this." That track record is much of what confidence is, and it is also what recalibrates the miscalibrated readiness gauge: the learner has now felt the difference between fluent understanding and successful action.

Notice what this list does to the four failure mechanisms. Unlabeled situations train recognition. Plausible options train choosing under ambiguity. Safe first attempts create the comfort that makes acting rational. And each corrected repetition is a rehearsal deposited against the competing habit's daily head start.

How can you spot the knowing-doing gap in your own programs?

The gap is detectable before the performance data arrives, if you know which signals to read. Five worth checking:

1. Your best evidence of learning is recall. If the strongest thing you can say about a graduate is their assessment score, you have verified knowing and measured nothing about doing.
2. Graduates ask the hallway question. Learners who passed the course still approach the facilitator afterward with "but what do I actually do when...?" That question is the learner accurately reporting that recognition and judgment were never built. Treat it as data, not as an individual's gap.
3. The behavior appears only under observation. Managers report that graduates use the new approach when reminded, coached, or watched, and revert under pressure. Reversion under load is the competing-habit mechanism, visible in the wild.
4. Confidence has no track record behind it. Learners rate themselves ready to apply the skill, yet the program never once required them to act on it anywhere. Learners who report high readiness but have never once acted on the skill show the same miscalibration Deslauriers documented in the classroom, and if it carries over to workplace learners, that combination should lower your confidence, not raise it.
5. You hear the phrase itself. When a manager says the team knows the material but nothing has changed on the floor, take the wording seriously: it is a field report on the psychology this guide describes, an observation that recall verifiably exists and action does not. Notice also what the complaint is not about. Nobody is saying the course was unclear or the material was wrong. Responding with a refresher or a better-produced module throws more knowledge at a problem that was never about knowledge, and the gap survives the fix.

Any two of these together indicate the same structural issue: the program builds knowing and assumes doing.

How does the application step close the gap?

Everything above points at one design conclusion: between acquiring knowledge and being expected to perform, learners need a stage where they learn to apply what they know, under the conditions that actually build judgment and comfort. Realistic situations that must be recognized rather than announced. Plausible options that force real choices. Guidance at the moment of choice. Failure that teaches instead of costing. Enough repetition that acting on the knowledge stops feeling like a debut.

AliveSim's Guided Scenarios are built around exactly this mechanism. Learners commit to a decision inside a realistic, unlabeled situation, receive expert guidance the moment the choice is made, and meet a related situation again soon after, so judgment and comfort accumulate from lived decisions rather than from a definition someone read to them. Each additional repetition banks another rehearsal against the old habit, narrowing the head start it has by the time a real moment arrives. AliveSim

The larger point stands independent of any platform. The knowing-doing gap is not a character flaw in learners, a failure of motivation, or a reason to rewrite the content. It is the predictable output of programs that build recall and assume the rest. People do what they have learned to do, and most training has only ever taught them to know. Design the doing, and the phrase managers use about those graduates of your program starts to change.

References

• Deslauriers, L., McCarty, L. S., Miller, K., Callaghan, K., & Kestin, G. (2019). Measuring actual learning versus feeling of learning in response to being actively engaged in the classroom. Proceedings of the National Academy of Sciences, 116(39), 19251–19257.
• Ericsson, K. A. (2008). Deliberate practice and acquisition of expert performance: A general overview. Academic Emergency Medicine, 15(11), 988–994.
• Ericsson, K. A., & Harwell, K. W. (2019). Deliberate practice and proposed limits on the effects of practice on the acquisition of expert performance: Why the original definition matters and recommendations for future research. Frontiers in Psychology, 10, 2396.
• Pfeffer, J., & Sutton, R. I. (2000). The knowing-doing gap: How smart companies turn knowledge into action. Harvard Business School Press.
• Saks, A. M., & Belcourt, M. (2006). An investigation of training activities and transfer of training in organizations. Human Resource Management, 45(4), 629–648.

Published April 8, 2025 · Updated July 14, 2026 · 11 min read