AI And Learning Transfer
In leadership programs, I often see managers leave a course with good intentions and a page full of notes. They have practised delegation, feedback, or coaching and can explain the principles clearly.
Two months later, their daily behavior may look much the same. They still take over difficult tasks, postpone uncomfortable conversations, and answer questions before team members have had time to think. The course delivered useful knowledge, while the workplace continued to reward familiar habits.
AI is making the content side of learning faster and cheaper. L&D teams can now create outlines, quizzes, videos, translations, role-play scenarios, and job aids in far less time than before. Synthesia’s 2026 survey of L&D professionals found that 88% were already saving time through AI, while 84% named faster production as its clearest benefit.
That matters because demand for reskilling is rising. The World Economic Forum’s Future of Jobs Report 2025 estimates that 59% of workers will need training by 2030 and that 39% of current skills will change or become outdated.
AI can help L&D produce more learning. It does much less to ensure that people use what they learn.
Learning Transfer Is Still The Main Constraint
Learning transfer describes whether people apply and sustain new knowledge and skills once they return to work. It has been studied for decades, yet discussion of the subject is still shaped by a widely repeated claim that only 10% of training transfers to the job.
That figure came from a personal estimate in a 1982 article rather than a measured study. Ford, Yelon, and Billington later described it as “the 10% delusion.” Their review offered a more useful picture: non-transfer was estimated at about 38% immediately after training, 56% after 6 months, and 66% after 12 months. The practical lesson is that transfer weakens over time unless something in the work environment supports the new behavior.
A meta-analysis by Blume, Ford, Baldwin, and Huang, covering 89 studies, found that motivation, ability, and a supportive work environment all influence transfer. The effect was especially important for open skills such as leadership, communication, and interpersonal behavior, where there is no single correct response.
Salas and colleagues reached a similar conclusion. Effective training depends on what happens before, during, and after the learning event. Course quality matters, but so do preparation, practice, feedback, manager support, and opportunities to use the skill.
Completion rates and knowledge checks capture only part of the picture. A manager can explain a delegation model and still review every detail personally. A salesperson can remember every discovery question and still move to a solution too early. In both cases, the person may understand the material while their working environment keeps pulling them back toward familiar behavior.
AI Improves Practice, But It Cannot Redesign The Workplace
AI already helps with several parts of learning design. It can create role-specific scenarios, provide immediate feedback during simulations, generate practice questions, and offer support at the point of need. It also makes it easier to tailor examples to different roles, markets, and experience levels. The limitation appears when the learner returns to work.
AI does not decide whether a manager will let an employee try a new approach. It cannot repair trust after a difficult first attempt, remove a conflicting incentive, reduce an unrealistic workload, or stop a senior leader from rewarding the old behavior because it seems faster.
Consider a manager learning to coach rather than solve every problem personally. An AI assistant can suggest questions such as, “What options have you considered?” It can simulate a coaching conversation and provide feedback. Then a real customer issue arrives on Monday morning, and the manager takes over because giving the answer still feels safer.
That is where transfer often breaks down. The manager knows what to do, but the surrounding system makes the old response easier.
The Five-Point Application Contract
I use a practical tool called the Five-Point Application Contract to move learning design closer to workplace use. It asks L&D teams, learners, and managers to agree on five conditions before the course ends.
1. Define The Behavior People Should Be Able To Observe
“Understand effective feedback” is too vague to guide action. A useful objective should describe what someone will do. For example: “Within two weeks, the manager will hold a feedback conversation in which they describe the observed behavior, explain its impact, invite the employee’s perspective, and agree on a next step.” That sentence gives the learner, their manager, and L&D the same definition of application.
2. Choose The First Real Use Case
Each learner should leave knowing where the new behavior will first be used. In a delegation program, this might be one recurring decision that will move to a team member. In feedback training, it should be a real conversation that already needs to happen. The first attempt should take place within days. A long delay makes it easier for routine work to take over.
3. Practice With The Resistance People Actually Face
Many exercises are cleaner than real working situations. Employees in role-plays listen carefully, respond rationally, and accept feedback. Real conversations include defensiveness, incomplete information, hierarchy, time pressure, and emotion.
Practice should include some of those conditions. AI simulations can help by changing the response, introducing disagreement, and allowing repeated attempts. Human judgment is still needed to make sure the scenario resembles the learner’s actual work.
4. Name The Person Who Will Follow Up
Someone needs to support application after the course. Too often, L&D assumes the manager will do it, while the manager assumes the course has already completed the development work. The person responsible for follow-up could be the learner’s manager, a peer, an internal coach, or a project lead. The task does not need to be demanding. Three questions can be enough:
- What did you try?
- What happened?
- What will you adjust next time?
Gallup has found that managers account for around 70% of the variance in team engagement. Engagement and learning transfer are different, but the finding illustrates how strongly managers shape the conditions in which new behavior is used.
5. Agree On Evidence And A Review Date
The final step is to define what evidence will be reviewed and when. That evidence might include a completed conversation, a delegated decision, fewer escalations, faster response times, or feedback from a colleague. A review after 30 days usually tells the organization more than a satisfaction survey completed immediately after training.
What This Looks Like In Practice
Consider an organization introducing a course on difficult performance conversations. The conventional version includes a 90-minute module, several examples, a conversation model, and a quiz. The LMS records completion and assessment scores.
A transfer-focused version begins before the course. Each participant identifies a real conversation they need to hold. During the session, they practice that conversation against realistic resistance. Before leaving, they schedule the discussion and share their planned approach with a manager or peer.
Within 7 days, the conversation takes place. The person responsible for follow-up asks what the participant tried, how the other person reacted, and what they would adjust next time. After 30 days, L&D reviews the application data. How many participants held the conversation? How many used the agreed structure? What prevented others from applying it? What should be changed for the next group?
AI can support much of this work. It can generate practice variations, help learners prepare, offer reflection prompts, and summarize recurring barriers across a cohort. The process still needs clear ownership and follow-through.
Personalization Should Continue After The Course
AI has renewed interest in personalized learning, but personalization is often treated mainly as a content issue. Learners receive different examples, recommendations, or pathways based on their role or previous activity. Application also needs to be tailored. Some people try a new behavior quickly and need help reflecting afterwards. Others need more preparation before they are willing to act. Some respond well to direct feedback, while others need more context before they can use it.
These differences should not be confused with fixed learning styles, for which the evidence is weak. A more useful question is how confidence, motivation, prior experience, and behavioral preferences affect practice and follow-up.
Behavioral tools can help with this. They give learners and managers language for discussing differences in pace, communication, decision-making, and responses to pressure without placing people into rigid categories. The insight becomes valuable when it changes how practice, feedback, and reinforcement are designed for a particular person.
Completion Data Should Be Joined By Application Data
AI will allow organizations to create more learning than many employees can reasonably absorb. That makes measurement more important. L&D teams should still track participation and completion, but they also need measures closer to performance:
- Time to first application
The number of days between training and the first real use. - Application rate
The share of participants who use the behavior within the agreed period. - Reinforcement rate
The share who receive the planned manager or peer follow-up. - Repeat application
Whether the behavior is used again after the first attempt. - Barrier data
The reasons people give for not applying the learning. - Performance evidence
The operational result expected to improve if behavior changes.
These measures are harder to collect than completion percentages. They are also more relevant to the reason the organization funded the program.
A course on delegation, for example, could measure how many participants transferred a real decision within two weeks, how often the decision was later escalated back to the manager, and whether team members reported greater clarity about their authority. A program on feedback could measure whether the planned conversations took place, whether managers used the agreed structure, and whether the issue was resolved or required further action. These are imperfect measures, but they tell L&D far more about workplace use than a quiz score.
Where L&D Should Focus Next
As AI reduces the time required to produce content, content volume will become a weaker measure of L&D capability. Strong learning teams will be distinguished by how quickly new skills reach the workplace and how reliably they remain there. That change will place more attention on transfer design, manager involvement, realistic practice, performance support, and application data. Instructional Designers will spend less time producing first drafts and more time improving the conditions in which behavior changes.
Within the next few years, leading L&D teams are likely to report time to first application and application rate for their most important programs, much as marketing teams report conversion rates. Completion figures will remain useful, but they will no longer be enough to show business value.
AI belongs in this system. It can improve access, relevance, practice, and support. Responsibility for the working environment still sits with managers and organizations. The most useful question for L&D is how much of the learning people use when the real work begins.
References:
- Baldwin, T. T., and J. K. Ford. 1988. “Transfer of Training: A Review and Directions for Future Research.” Personnel Psychology 41 (1): 63–105.
- Blume, B. D., J. K. Ford, T. T. Baldwin, and J. L. Huang. 2010. “Transfer of Training: A Meta-Analytic Review.” Journal of Management 36 (4): 1065–1105. https://doi.org/10.1177/0149206309352880
- Colquitt, J. A., J. A. LePine, and R. A. Noe. 2000. “Toward an Integrative Theory of Training Motivation: A Meta-Analytic Path Analysis of 20 Years of Research.” Journal of Applied Psychology 85 (5): 678–707. https://doi.org/10.1037/0021-9010.85.5.678
- eduMe. 2026. The State of AI in L&D: Trends, Stats, and Real-World Examples from 2026.
- Ford, J. K., T. T. Baldwin,and J. Prasad. 2018. “Transfer of Training: The Known and the Unknown.” Annual Review of Organizational Psychology and Organizational Behavior, 5: 201–25. https://doi.org/10.1146/annurev-orgpsych-032117-104443
- Ford, J. K., S. L. Yelon,and A. Q. Billington. 2011. “How Much Is Transferred from Training to the Job? The 10% Delusion as a Catalyst for Thinking About Transfer.” Performance Improvement Quarterly 24 (2): 7–24. https://doi.org/10.1002/piq.20108
- Gallup. 2015. State of the American Manager: Analytics and Advice for Leaders.
- Salas, E., S. I. Tannenbaum, K. Kraiger, and K. A. Smith-Jentsch. 2012. “The Science of Training and Development in Organizations: What Matters in Practice.” Psychological Science in the Public Interest 13 (2): 74–101. https://doi.org/10.1177/1529100612436661
- The Complete Guide to DiSC® Personality Styles. 2026.
- DiSC and AI: A Style-Based Guide to Embracing Technology. 2025.
- Synthesia. 2026. AI in Learning & Development Report.
- World Economic Forum. 2025. The Future of Jobs Report.
