A training room can produce excellent evidence that learning happened and no evidence that work changed. Participants pass the end-of-course quiz, explain the model to a colleague, and leave with good intentions. A month later, the old decisions, shortcuts, and conversations are still in place.

That gap is a training transfer problem. Training transfer means applying and maintaining a skill in the job after instruction. Learning transfer in the workplace depends on more than recall. It requires a recognizable cue, a real opportunity, a consequence that supports the new behavior, and enough practice to perform under ordinary pressure.

There is no credible universal percentage for how much training gets “lost.” Results vary by skill, job, manager, follow-up, and measurement. A person who remembers the model but never gets a chance to use it has a transfer problem, not necessarily a memory problem.

Here are four field diagnoses for the ways training effectiveness goes wrong after the workshop.

1. The course objective describes knowledge, not work

If the objective says “understand,” transfer has nowhere to go.

Symptom

Participants can define the concept, pass a knowledge check, and describe why it matters. When a supervisor asks what will look different next week, the answer becomes vague: “I’ll communicate more strategically” or “I’ll use the framework with my team.”

There is no shared agreement about the situation that should trigger the behavior, the action the learner should take, or the evidence that would show it happened.

Root cause

The design stopped at learning rather than behavior. In Kirkpatrick’s four-level model, a quiz or demonstration during training sits around Level 2: learning. Level 3 asks whether behavior changed on the job. Those are separate questions.

Baldwin and Ford’s 1988 transfer model makes a similar distinction. Training inputs and learning matter, but transfer also requires generalization to the work setting and maintenance over time. A competency label such as “strategic thinking” is not a job instruction. No one can transfer a noun.

Fix

Write a one-page transfer specification before revising the course. It should name:

Trigger: the situation in which the skill should appear.
Action: what the person will do or say.
Quality bar: what competent performance looks like.
Evidence: an artifact, decision, observation, or record that can confirm it.
First opportunities: the next two or three occasions when the learner can try it.

For example, replace “understand vendor conflict-of-interest policy” with something observable: “Before approving a supplier, identify any personal or financial connection, pause the approval when required, and use the designated escalation route.” The exact policy will vary, but the design logic holds.

Keep the specification to one or two critical behaviors. Five new behaviors create a curriculum. One behavior tied to three real opportunities creates a transfer test. That is a better place to start this week than another content review.

2. Practice is polished; the job is not

Good role-play can be a trap.

Symptom: everyone performs well in the room, then struggles when the real counterpart interrupts, withholds information, outranks them, or brings a competing priority. The learner remembers the script but misses the cue that should have prompted it.

Root cause: practice has encoded the training conditions instead of the work conditions. The trainer gives helpful prompts. The case contains all the relevant facts. The other person follows the exercise rules. There is no time pressure and no cost for choosing the wrong response.

This is why “make it realistic” is incomplete advice. Maximum realism adds complexity, but complexity can hide the skill you are trying to build. The first practice round should usually be easier than the job. The important move is to increase difficulty in stages.

Use a three-pass rehearsal:

Supported pass: isolate one decision or behavior. Give the learner a clear cue, a short case, and immediate feedback.
Variable pass: change the facts, counterpart, and wording while keeping the underlying decision the same. The learner must recognize different routes to the same behavior.
Pressure pass: add a time limit, incomplete information, status differences, or a competing metric. Remove the script and let the learner recover from a poor first move.

The debrief should focus on transfer, not performance theater. Ask: What signal did you notice? What did you do next? What made the new behavior difficult? Which part of the situation at work will remind you to use it?

For a procurement analyst, practice should not merely recreate a polite classroom conversation about gifts from suppliers. It should vary the cue: a senior vendor offers an invitation, a familiar supplier sends a personal favor, or a deadline makes the questionable approval convenient. Match the decision cues and consequences, not every surface detail.

This fails when learners lack the basic skill in the first place. Pressure rounds are useful for testing transfer, but they cannot substitute for clear instruction and supported rehearsal. A difficult simulation can reveal a gap; it does not automatically close one.

3. The manager quietly cancels the training

The person who did not attend may determine whether transfer occurs.

A learner returns ready to use a new escalation protocol. The manager says the team is too busy this quarter. A colleague says the old process is faster. The local performance metric rewards closing tickets quickly, even though the new protocol adds a careful verification step. Nobody objects to the training. The work simply makes the trained behavior inconvenient.

That is a transfer-climate problem. In their review of transfer research, Grossman and Salas identified factors such as supervisor support, peer support, opportunity to use the skill, and consequences in the work environment. Support is not limited to approving the course or asking whether people enjoyed it. It includes protecting the first attempt and responding constructively when it is imperfect.

A useful manager pact can take 15 minutes. Before or immediately after the course, the manager and learner should agree on one real situation, one visible sign of good performance, and one way the manager will create an opportunity. The agreement can be as plain as:

“Use the new escalation step in the next two cases, even if it takes longer. I’ll protect the time and review one example with you on Friday.”

The manager does not need to become a full-time coach. One observation, one piece of specific feedback, and one removed obstacle are often more useful than a request to reinforce twelve concepts.

The frank limit matters here: training cannot overcome a contradictory incentive by itself. If a process metric penalizes the behavior the course promotes, another reminder will not fix the problem. Change the metric, policy, staffing assumption, or approval workflow—or describe the program as awareness training rather than a behavior-change intervention. That distinction protects both the learner and the credibility of L&D.

4. Evaluation stops at attendance and reaction

A survey can tell you that people liked the facilitator. It can sometimes tell you whether they remember the material immediately afterward. It cannot, by itself, tell you whether the skill traveled into the job.

The missing question

Many post-course evaluations ask whether the content was relevant and whether the learner feels confident. Those questions have value as diagnostic signals. They are not evidence of generalization or maintenance. Confidence can rise because the learner now has useful vocabulary, even when the surrounding work still prevents action.

Kirkpatrick’s model is helpful here because it separates reaction, learning, behavior, and results. Training effectiveness should be assessed at the level of the promise made. If the promise was a change in how managers conduct risk reviews, a smile-sheet score is not the main outcome. If the promise was a measurable reduction in processing errors, behavior evidence should come before a claim about business results.

Build a small transfer scorecard

Record the following for each observed opportunity:

The situation that created the opportunity.
The expected behavior.
What the person actually did.
The evidence available, such as a decision record, completed check, customer outcome, or manager observation.
The barrier: knowledge, skill, permission, workflow, time, or competing incentive.
The next support action.

Count opportunities, not merely calendar days. A skill used several times a week can be checked after the first three real instances. A skill used during a quarterly incident needs a different follow-up plan; a 30-day survey may arrive before the behavior has ever been needed.

Mark “no opportunity” separately from “did not perform.” That small distinction prevents L&D from blaming learners for a work condition. It also points to the right intervention. A knowledge gap may call for instruction. A skill gap may call for more rehearsal. A permission or workflow gap belongs with the manager or process owner.

Business results need even more care. Sales, retention, quality, and safety outcomes are influenced by staffing, seasonality, tools, and market conditions. Do not claim that training caused a result simply because the result followed the course. Use behavior evidence first, then test whether the intended business signal moves under comparable conditions.

A one-week transfer audit

Pick one recently completed course with a clear business reason. Do not begin by rebuilding its slides. Run a 45-minute audit instead:

Rewrite one vague objective as a trigger, behavior, quality bar, and evidence source.
Ask the learner’s manager to name the next real opportunity and the barrier most likely to interfere.
Review three recent work samples, or observe the next available opportunity.
Classify the gap as knowledge, skill, permission, workflow, or measurement.
Set one follow-up observation date and one support action.

The result may show that the course needs better practice. It may show that the manager needs to protect time. It may show that the behavior was never appropriate for that job. All three findings are more useful than another completion report.

By Friday, aim to have one behavior statement and one real observation date on the calendar. That is where training transfer becomes visible—and where the next design decision should begin.

Training Transfer Fails at Work for 4 Predictable Reasons