Last updated September 28, 2026.
Only about 1 in 3 employees say they have received enough AI training, according to Boston Consulting Group's 2025 AI at Work survey of 10,635 workers in 11 countries. Among employees who received more than five hours of training, about 4 in 5 used AI several times a week or more, compared with about 2 in 3 of those who received less.
AI adoption training changes behavior when employees repeatedly use AI on real work, learn how to catch its errors, get support matched to their role and starting attitude, and return to managers and policies that reinforce use.
At The Decision Lab, most of our AI adoption work starts after the training budget has been spent, with organizations whose employees have been trained and still do not use the tools. Across that work, we find that training limited to explaining what AI tools can do changes what people intend to do far more than what they actually do.
We saw this pattern at a Fortune 500 HR technology company with a 20,000-person workforce. The company had deployed AI tools with training and internal communications behind them, and usage stayed low. We designed 20 candidate programs, each aimed at a specific barrier for a specific group of employees, and piloted eight of them for one month with more than 100 employees against a control group. Confidence in exploring AI tools, measured before and after the pilot, rose 41% among participants and fell 26% among control-group colleagues over the same period. The programs are now scaling to the client's full workforce, and we measure the conditions behind results like these with our SPROUT framework, validated with data from more than 20,000 employees.
Why most AI training does not change behavior
Most corporate AI training is built to transfer knowledge, usually what the tools can do and how to write a good prompt. Employees need that knowledge, but knowing how to use a tool and choosing to use it in the middle of a busy week depend on different conditions.
A large survey experiment in Denmark shows the gap. Economists Anders Humlum and Emilie Vestergaard surveyed 100,000 workers across 11 occupations exposed to ChatGPT, and gave some of them expert assessments of what the tool could do. The information shifted what workers believed and what they said they planned to do, but it had limited effects on whether they started using ChatGPT. The same study found that workers were often held back by employer restrictions and by the training they would need to use the tool.
Research on training transfer, the degree to which what people learn in training shows up in their daily work, points the same way. A 2010 review of 89 studies by Brian Blume, J. Kevin Ford, Timothy Baldwin and Jason Huang found that transfer was linked to learners' cognitive ability, conscientiousness and motivation, and to a supportive work environment. The work environment belongs to managers and the wider organization, and a course cannot change it from inside the classroom. The review also found that motivation and the work environment mattered more when training taught open skills, where there is no single correct way to do the task. Using generative AI at work fits that description, since the right request and the right amount of checking change from one task to the next.
Our Fortune 500 diagnostic, built on a 48-source review of the adoption literature combined with workforce interviews and surveys, showed the same thing inside one company. Employees knew the tools existed and had been shown how to use them. What held them back was the time it takes to change a workflow before any payoff arrives, doubt about whether AI output could be trusted, unclear guidance on what was allowed, and difficulty seeing where AI fit into their actual job.
How we define AI adoption training
At The Decision Lab, we define AI adoption training as a structured program intended to change how often and how well employees use AI in their real work, judged by what they do in the weeks after it ends. Under this definition, completion rates and quiz scores are inputs, and the outcome is repeat use of AI in a specific task. That outcome can also move for reasons outside the course, such as a manager who starts asking about AI in weekly team meetings, and we treat those reasons as part of the program design.
Five features of AI training that changes behavior
Across our AI adoption engagements, the programs that change behavior tend to share five features.
1. Practice happens on the employee's own work
Generic exercises teach the tool, while practice on a real task also produces the first instance of the new habit. The eight programs we piloted at the Fortune 500 client ranged from hands-on tasks built into employees' existing workflows to a mission-based challenge series run in Slack, and each was designed to clear one specific barrier. BCG's survey points the same way, with regular AI use sharply higher among employees who had access to in-person training and coaching.
2. Mistakes are part of the design
Nina Keith and Michael Frese pooled 24 studies with 2,183 participants on error management training, a method in which learners explore on their own and are told to expect errors and learn from them. Compared with training designed to prevent errors, it produced better performance on average. The advantage grew after training ended and was largest on tasks unlike the ones people had practiced. AI tools produce wrong answers unpredictably, and employees who have only seen polished demonstrations are likely to lose trust at the first bad output. We build deliberate failure cases into our programs so that employees have already caught and fixed an AI error before they meet one on a deadline.
3. Programs match where people start
In the Fortune 500 diagnostic, we mapped how employees move from first exposure to everyday AI use and found two groups with opposite failure modes. Enthusiasts were already experimenting and kept hitting friction, while skeptics distrusted AI's value and stalled at first contact. Enthusiasts need obstacles removed from tools they already want to use, and skeptics need a low-stakes first use along with evidence from their own role. A single course for both groups tends to bore the first and lose the second.
4. Managers and guidelines carry training into the job
BCG found that the share of frontline employees who feel positive about generative AI rises from about 1 in 7 to more than half when they report strong support from leaders, and only about 1 in 4 frontline employees report that level of support. For this reason, each program we piloted at the Fortune 500 client came with guides and support structures that let the client's own teams run it after we stepped back. Social risk also shapes what happens after training. Research by Jessica Reif, Richard Larrick and Jack Soll, published in the Proceedings of the National Academy of Sciences in 2025, found that people who use AI at work are judged by others as less competent and less motivated. Managers who use AI openly and treat responsible AI-assisted work as normal may reduce that social risk, a barrier that training alone cannot fully address. Clear usage guidelines settle the question employees raise most often after training: what they are actually allowed to do.
5. Success is measured on behavior, against a comparison group
When a vehicle manufacturer asked us to evaluate a new sales model that had changed how its sales staff were trained, we used statistical methods to separate the model's real impact from seasonal trends and the COVID-19 pandemic. The same logic shaped our Fortune 500 pilot. Had we measured only the pilot group, a 41% rise in confidence would have looked like the full effect, when colleagues without the programs were losing confidence over the same period.
A practical AI adoption training sequence
The five features translate into a sequence that leaders can run with one team before scaling to the rest of the organization.
- Pick one repeatable, high-value task for each role, such as a weekly client summary or a first-draft proposal.
- Write down which AI uses are allowed and which are off-limits for that task before anyone starts training.
- Run guided practice on real work that contains no sensitive data.
- Include at least one scenario where the AI gives a wrong answer, and have employees find and correct it.
- Give managers a place to get help running the program, such as a dedicated chat channel, and ask them to spend a few minutes of each weekly team meeting on how people used AI that week.
- Track repeat use of AI in the task for four to eight weeks after the program, alongside the time and quality of the finished work.
- Compare results with a similar team that receives the same program afterwards.
How to measure whether AI training changed behavior
We track four levels, from easiest to measure to most meaningful:
- Completion: who attended or finished the program. It shows exposure and nothing more.
- Confidence: how sure employees feel about exploring and using AI, measured before and after.
- Repeat use in a target task: whether employees keep using AI for the specific work the program was built around, weeks after it ends.
- Work outcomes: time taken and quality in that task, compared with a baseline.
In our Fortune 500 pilot, we measured confidence in exploring AI tools and the time employees had to use AI, before and after, for both the pilot and control groups. Most organizations we meet report only completion, and a program judged on completion alone can report success while usage stays flat.
Where training fits in the SPROUT framework
Our SPROUT framework measures five conditions for AI adoption: Social Readiness, Psychological Readiness, Role Fit, Organizational Enablement, and Understanding and Technical Capability.
Training acts directly on Understanding and Technical Capability, and well-designed practice can also lift Psychological Readiness by giving skeptics a safe first success. Social Readiness and Role Fit depend on team norms and job design, and Organizational Enablement depends on policy and leadership. A program that ignores those conditions tends to produce trained employees who still do not use AI.
Before designing training, we usually run our AI Adoption Diagnostic, 17 anonymous questions that need at least five responses, to see which conditions are actually weak. When the diagnostic shows that Understanding and Technical Capability is already strong, more training is likely to be the wrong investment, and the budget does more good spent on manager routines or clearer guidelines.
Frequently asked questions
Why is AI adoption stalling even after training?
AI adoption usually stalls after training because training addresses knowledge, while the barriers that remain are behavioral: no protected time to change a workflow, low trust in AI output, unclear rules on what is allowed, and managers who do not reinforce use. At one Fortune 500 client, trained employees rarely used the tools until those barriers were addressed.
What makes AI training effective for employees?
Effective AI training builds practice around employees' own tasks, with room to make and fix mistakes, and it continues after the session through manager follow-up. We pay most attention to the first few real uses after a program, since those early attempts decide whether the new habit or the old workflow wins.
How many hours of AI training do employees need?
We know of no validated minimum. Survey data from BCG link more than five hours of training with more regular use, but that comparison cannot separate the effect of hours from the kind of employer that offers them. In our experience, the share of those hours spent practicing on real tasks matters more than the total.
How do you measure the effectiveness of AI training?
Measure repeat use of AI in a specific target task in the weeks after training, and compare trained employees with a similar group that has not been trained yet. Without that comparison group, a company-wide trend such as falling confidence or a busy season can hide a real effect or create a false one.
What is the difference between AI literacy training and AI adoption training?
AI literacy training teaches what AI is and how to use it safely, and success is measured by knowledge. AI adoption training aims at a change in daily work, so success is measured by how often and how well employees use AI on real tasks afterwards. Most organizations need both, and literacy usually comes first.
What role do managers play in AI adoption after training?
Managers decide whether AI use feels expected or risky once training ends. BCG found that the share of frontline employees who feel positive about generative AI rises from about 1 in 7 to more than half with strong leadership support. Managers also set the practical rules, including which tasks should change and how AI-assisted mistakes will be handled.
