What Is Learning Transfer?

By Brian Tr · Updated Oct 7, 2026

A training program can score well on every end-of-course survey and still change nothing about how people work. Learning transfer is the name for that gap. It asks a simple question: after the course, did people use what they learned, and did they keep using it?

Researchers have studied the question for more than a century, and the corporate version, usually called transfer of training, has its own literature. This article covers what the term means, the models researchers use to explain it, what the largest reviews found about what helps, and how transfer connects to evaluation. It also looks at the most repeated number in the field, the claim that only about 10% of training transfers, and where that number really comes from.

Key Facts

• Baldwin and Ford's 1988 review defines transfer as learned behavior that is generalized to the job context and maintained over a period of time on the job (Baldwin and Ford, 1988).
• Their model groups the causes into three inputs: trainee characteristics, training design and work environment (Baldwin and Ford, 1988).
• A meta-analysis of 89 empirical studies found positive links between transfer and cognitive ability, conscientiousness, motivation and a supportive work environment (Blume et al., 2010).
• Most predictors in that meta-analysis were stronger for open skills such as leadership than for closed skills such as computer software (Blume et al., 2010).
• The Learning Transfer System Inventory was validated on 1,616 training participants and measures 16 transfer system constructs (Holton, Bates and Ruona, 2000).
• The oft-quoted "10% of training transfers" figure was cited from Georgenson (1982) in Baldwin and Ford's review, and a later paper found it lacks empirical, behavioral evidence (Ford, Yelon and Billington, 2011).

Defining learning transfer

Baldwin and Ford, in the 1988 Personnel Psychology review that most later work builds on, define positive transfer as the degree to which trainees effectively apply the knowledge, skills and attitudes gained in training to the job. For transfer to have occurred, they write, learned behavior must be generalized to the job context and maintained over a period of time on the job (Baldwin and Ford, 1988).

Two words in that definition do a lot of work.

Generalized. The job is rarely a copy of the classroom. Someone who practiced a coaching conversation with a colleague in a workshop has to run a different one with a real, irritated employee on a Tuesday afternoon. Transfer means the skill survives the change of setting.

Maintained. Using a new skill once the week after training is not enough. The definition asks for use over time. That makes transfer a close cousin of retention, which is why it pairs naturally with the research in spaced repetition and the forgetting curve.

It also helps to separate transfer from learning. Learning is what happens in the program: the person can do or explain something at the end. Transfer is what happens after: the person does it at work. Baldwin and Ford treat learning and retention as outputs of training that feed into transfer, but they also hypothesize that trainee characteristics and the work environment affect transfer directly, whether or not the person learned much in the program (Baldwin and Ford, 1988). A person can learn a lot and still not use it if the job gives them no chance to.

Near transfer and far transfer

Researchers often split transfer by distance. Near transfer is applying a skill in a context close to the one where it was learned: practicing a CRM data-entry routine in a training sandbox that looks like the live system, then using it in the live system. Far transfer is applying it in a context that differs more, such as using a negotiation framework from a workshop in a very different kind of deal.

The distinction matters for course design, but it is slippery. Barnett and Ceci, in a widely cited 2002 paper in Psychological Bulletin, argue that debates over whether far transfer happens made little progress because researchers failed to specify the dimensions along which transfer can vary. They propose a framework of nine dimensions and conclude that estimating a single effect size for far transfer is misguided, adding that the evidence for transfer under some conditions is substantial while critical conditions for many key questions remain untested (Barnett and Ceci, 2002). That paper comes from education and cognitive psychology, not workplace training, but its warning carries over. "Does training transfer?" has no single answer. It depends on what was taught, to whom, and to which situations.

The Baldwin and Ford model

The 1988 model is the standard starting point because it sorts the causes of transfer into three groups.

Source: Baldwin and Ford, 1988.

The model then links those inputs to two training outputs, learning and retention, and to the conditions of transfer, generalization and maintenance. Baldwin and Ford specify six linkages among them. The reading that matters for practitioners is the structure: transfer is not decided in the classroom alone. A good design is necessary, but a learner's motivation and the job's willingness to let them practice sit beside it with equal standing.

The authors also pointed to research gaps, among them the need to test different operationalizations of training design and work-environment factors, to build a framework for studying trainee characteristics, and to improve how transfer itself is conceptualized and measured (Baldwin and Ford, 1988). Measurement is still a weak point today, as the next section shows.

What later reviews found

Two reviews from the 2000s updated the picture, and a third summarized the field a decade later.

Burke and Hutchins (2007). In Human Resource Development Review, they integrated empirical research from management, HRD, training, adult learning, performance improvement and psychology. They organized the evidence around three factors, learner characteristics, intervention design and delivery, and work environment influences, and identified the variables with substantive support and the most pressing gaps (Burke and Hutchins, 2007). That is the same three-way split as Baldwin and Ford, now supported by far more studies.

Blume, Ford, Baldwin and Huang (2010). Their meta-analysis in the Journal of Management pooled 89 empirical studies. It opens by noting that the literature suffers from inconsistent measurement of transfer and significant variability in findings. Its results confirmed positive relationships between transfer and cognitive ability, conscientiousness, motivation and a supportive work environment (Blume et al., 2010).

Two moderator findings are worth remembering:

1. Open versus closed skills. Most predictors, including motivation and work environment, had stronger relationships to transfer when training focused on open skills, such as leadership development, than on closed skills, such as computer software.
2. Same-source, same-context measurement. When transfer outcomes came from the same source in the same measurement context, relationships were consistently inflated (Blume et al., 2010). A survey in which trainees rate both their motivation and their own use of the skill will look stronger than the truth.

Ford, Baldwin and Prasad (2018). Their review in the Annual Review of Organizational Psychology and Organizational Behavior synthesizes what is reliably known about the generalization and retention of learning, then argues for more precision in how variables and interventions are specified, more attention to transfer criteria and transfer trajectories, and research that examines learning from a problem-centered perspective (Ford, Baldwin and Prasad, 2018). The "trajectory" idea is useful: transfer is a path over weeks and months, not a single yes or no.

The Learning Transfer System Inventory

If the work environment matters this much, it should be measurable. Holton and colleagues built the Learning Transfer System Inventory (LTSI) for that job. In the 2000 validation study, it was administered to 1,616 training participants from a wide range of organizations. Exploratory factor analysis found sixteen transfer system constructs, and a second-order analysis suggested three higher-order factors: climate, job utility and rewards (Holton, Bates and Ruona, 2000). The authors discuss it as a diagnostic tool for organizations.

In practice, an LTSI-style survey gives a training team something better than a satisfaction score. It shows whether the barrier is that people were not motivated, that the training did not fit their jobs, or that the workplace gave them no support and no reward for trying. Those three problems need different fixes.

The Broad and Newstrom matrix

Baldwin and Ford describe what shapes transfer. Mary Broad and John Newstrom, in their book Transfer of Training (1992), turned the question into an action grid: who can do something, and when. The matrix crosses three roles, the manager, the participant and the trainer, with three times, before, during and after the program, which gives nine cells (Training magazine summary of the matrix).

The value of the grid is the coverage it forces. Training plans often name an owner only for the delivery itself, the cells the trainer controls. A plan that names who acts in all nine cells, especially the manager before and after the program, covers more of the transfer system the research describes. The matrix says what to cover, not how to do it, so each cell still needs its own tactics.

What helps: supervisor support, opportunity and practice

Across these sources a few levers keep appearing.

Supervisor and peer support. Baldwin and Ford list supervisory or peer support as a core work-environment factor, and Blume and colleagues confirm a supportive work environment as a positive predictor of transfer (Baldwin and Ford, 1988; Blume et al., 2010). In practice that puts the manager at the center, since the manager decides how much room the new method gets.

Opportunity to perform. Baldwin and Ford name constraints and opportunities to perform learned behaviors as part of the environment. A person trained in a new reporting process who is not allowed to use it for three months will lose it. For sellers of training, this is why a program tied to a live project, in the spirit of microlearning delivered close to the moment of need, fits the logic of the model better than a course taken in the abstract.

Practice and relevance. Training design in the original model includes job relevance of content and sequencing. Practice with feedback belongs to both the design and the follow-up. Retrieval and spacing from the earlier article apply here too.

Motivation and ability. Learners who want to apply the skill, and who have the underlying capability, show better transfer in the meta-analytic evidence (Blume et al., 2010). That makes selection and framing part of the design: telling people why the skill matters to their own work is not a soft extra.

Good design upstream. If the program trains the wrong behavior, no amount of manager support rescues it. Transfer is easier when the skill gap was real in the first place, which links back to training needs analysis and the structured design steps in instructional design basics.

Transfer and Kirkpatrick Level 3

Transfer is what Level 3 of the Kirkpatrick model is trying to observe. Kirkpatrick Partners describe Level 3 as Behavior, whether behavior actually changes on the job, and Level 4 as Results. In a short article on the subject, they point to line managers and supervisors as the people best placed to put "Required Drivers" in place: processes and systems that reinforce, monitor, encourage and reward performance of critical behaviors on the job (Kirkpatrick Partners, "Unlocking Kirkpatrick Levels 3 and 4").

Put the vocabularies side by side and they line up. Required Drivers are an applied version of the work-environment inputs in the Baldwin and Ford model. A Level 3 measure is a transfer measure. And Level 4 results, the business outcomes buyers care about, cannot appear if Level 3 never happens. The full model is covered in the Kirkpatrick model, and its role in buyer conversations is in proving training ROI.

The "only 10% transfers" claim

You will find some version of this in many training decks: only 10% of training transfers to the job, or 10 to 20% on a generous day. It is worth tracing, because it shows how a rough guess becomes a "fact."

Baldwin and Ford's 1988 review is one place the claim appears in print. It states that American industry spends up to $100 billion a year on training and development and that not more than 10% of this actually results in transfer to the job, and it attributes the estimate to Georgenson (1982) (Baldwin and Ford, 1988). That review is highly cited, which gave the number a respectable-looking trail.

In 2011, Ford, Yelon and Billington examined where the belief came from. Their paper in Performance Improvement Quarterly says it provides evidence of the source of the generalization and of its acceptance "despite the lack of empirical, behavioral evidence." They also present five questionable assumptions behind the idea that transfer is minimal, and offer four strategies for planning, assessing and reporting transfer, including stating realistic transfer goals, setting quantitative standards of successful transfer and reporting the complete transfer story (Ford, Yelon and Billington, 2011).

The fair reading: there is no measured study behind "10%." That does not mean transfer is high. Blume and colleagues' meta-analysis shows the factors that lift or lower it, and Ford, Baldwin and Prasad's later review stresses how much remains unknown. It means nobody should quote a single percentage as the transfer rate of training. If a vendor does, ask what behavior was measured, how, by whom and how long after the program.

What this means for EdTech companies and L&D buyers

For builders. Transfer is where a learning product proves its value, and also where most products stop short. A course platform can deliver content and track completion. It usually can't create the manager conversation, the chance to practice or the reward for using the skill. Products that build those in, such as manager nudges, on-the-job practice tasks and follow-up prompts, give themselves more of a case. Completion and activity data from learner engagement metrics show who took part, not who applied anything.

For buyers. Ask what happens in the "before" and "after" cells of the Broad and Newstrom matrix. Ask how the vendor measures use on the job, and whether the measure relies on the same people rating their own motivation and their own behavior, which the meta-analysis warns inflates results. Be careful with transfer claims built on the 10% figure in either direction.

Related reading

• Spaced Repetition and the Forgetting Curve
• The Kirkpatrick Model
• Proving Training ROI to Buyers
• Instructional Design Basics
• Training Needs Analysis