"90% of the content in a training course is useless." Really?!

1. But where does this 10% myth come from?

This figure comes from an article by David Georgenson published in 1982, in which the author asks a purely rhetorical introductory question: “How many times have you heard training directors say that they estimated that only 10% of the content presented in a classroom translated into a change in behavior at work?” A simple question. No data. No citation. No study. Yet the article was impactful and influential enough to be cited as a reference by generations of authors.

Little by little, after forty years of citations, this rhetorical question became a “research finding.” A snowball effect: everyone citing the previous author, who had cited Georgenson. Robert Fitzpatrick documented the mechanics in detail. A form of intellectual laziness.

Rather embarrassing... Just goes to show that you should always check your sources.

But is there actually some truth behind this figure?

2. Let's focus on management.

If we look at a meta-analysis conducted by Lacerenza, Reyes, Marlow, Joseph & Salas in 2017, which brought together 335 independent studies (26,573 participants), published between 1951 and 2014, focusing exclusively on management and leadership training: it seems that training works better than people say!

Indeed, the impact of management training on actual behavior is positive, measurable, and shows an average improvement in behavior of around 20 to 30%. Neither negligible nor miraculous.

But that's not the most interesting part of the study... Because it doesn't just tell us whether it works, but under what conditions. Indeed, a 20 to 30% improvement is an average; some training programs appear to be much more effective than others.

And when you look closer, you realize that what makes the difference between training that is forgotten and training that sticks comes down to three success factors. Just three. Each one multiplying the effect on behavior by a factor of 2 to 5.

Practice. Feedback. Iteration.

Management training that simply focuses on transmitting information is 4x less effective than training that includes demonstration and hands-on practice. That's it.

What you know evaporates. What you have done is remembered.

This leaves us with one question, and it is the question that drives the rest of this article: if these 3 ingredients are decisive, are they actually being used in training today?

3. What about repetition (practice/feedback/iteration) when you're a manager?

Management is an open, motor, conversational skill. Like flying an aircraft or performing surgery, it is acquired through repetition under constraints.

If we make a Fermi estimate for a standard 2-day management training course:

• 2 days of training: 14h
• Remove theory and slides (50%): - 7h
• Remove discussions and debates (25%): - 3h30
• Remove logistics, introductions, breaks, ice-breakers (10%): - 1h30

That leaves 2 hours for practice. If we account for rotating between the actor/partner/observer roles, we can hope to get 40 minutes of actual practice.

If we take into account the truly effective practice time (the rest being context-setting and awkward laughter), we end up with 20 minutes.

20 minutes. That's roughly the amount of truly effective training time delivered over those two days.

No airline pilot takes off after 20 minutes in a simulator. No surgeon operates after spending 20 minutes on a mannequin.

This type of training, although very common in management, simply isn't enough.

The problem, then, is not that “training doesn't work,” but that “current training programs are far from optimal”: we rely on exposure to content (which doesn't work particularly well).

The result: management is learned on the job, where we find practice, feedback (our very unhappy client), and repetition. But at what cost? Every mistake comes at a high price and ultimately affects clients, teams, and managers themselves.

4. Simulation is a solution, and it is extensively documented

Practice, feedback, and iteration are precisely what simulation-based learning is designed to provide.

The field in which the effect of simulation has been measured most rigorously is healthcare. The meta-analysis by Cook et al. (2011), based on 609 studies and 35,226 learners, explores the impact of simulation (I'll let you read the paper; it is particularly interesting) and shows three things:

• The immediate effect on skill improvement is massive: the average learner outperforms 88% of untrained learners and is faster than 87% of them.

• Long-term retention is significantly better: after training, the average learner still outperforms 80% of untrained learners.

• And this is important for intellectual honesty: the variability is enormous. These meta-analyses therefore tell us that simulation can produce very large effects, not that any simulation will do so. Realism and feedback quality are both very important to its educational impact.

In short, simulation works incredibly well—if it's done right!

5. What AI changes in practice

Nevertheless, although we know that simulation is a great solution, it has been applied very little to management until now because it has generally been complex and expensive.

Real-life, human-based practice creates all kinds of problems: organization, location, trainer availability, and significant costs if you want regular repetition.

Virtual reality addresses some of these problems, but creates others: the headset, filmed content, rigid scenarios that lack realism, and cost.

But the arrival of multimodal generative models removes the constraints that previously made hands-on practice impossible to industrialize:

Realism. Including in complex situations. A scenario no longer needs to be written in advance: the other person improvises, pushes back, gets defensive, comes back around. You can simulate a disagreement that escalates or an announcement that is poorly received, not just a textbook case.

The freedom to make mistakes. This is the most underestimated point. In a classroom role-play, the learner performs in front of their colleagues. So they aren't really practicing: they're putting on a performance. In front of a model, there is no one to impress. They can fail, start over, try an overly direct formulation to see what happens. This is the very foundation of deliberate practice.

High-quality automated feedback. Every interaction can be analyzed, commented on, and compared against assessment frameworks automatically by models specialized in the field.

An economy that flips. Traditional training costs the price of a session each time it is repeated: the cost scales linearly with volume. An AI simulation has a fixed design cost, followed by a marginal cost of just a few cents for each additional repetition. It's not “cheaper”; it's a different cost model that makes ten hours of practice as accessible as twenty minutes.

Going from 20 minutes of practice to 10 hours is no longer a question of budget.

Conclusion

Corporate training is not ineffective. It is simply not optimal and underdosed when it comes to repetition: Practice. Feedback. Iteration.

The body of evidence has been consistent for forty years: people learn, then fail to act. Open-ended skills, with management at the top of the list, are precisely the ones that suffer the most from this.

The only training modality that consistently produces large effects on actual behavior is repeated simulation with structured feedback.

What was missing wasn't the evidence. It was an affordable solution that could be deployed at scale. A role-play with a professional actor costs between €800 and €1,500 per day for a group. The same exercise, performed with an AI, costs a few cents per repetition and can be performed a thousand times.

The 10% myth survived for forty years because it was convenient: it provided an excuse. We can now do without it.

At YouMight, we are building a platform to de-risk management transitions and prepare managers to face any situation through AI-powered simulation. Not content. Repetition, with counterparts who don't simply agree with you and personalized feedback that pushes you to your limits.

Sources

• Fitzpatrick / Work-Learning Research — the origin of the 10% myth
• Lacerenza, Reyes, Marlow, Joseph & Salas (2017), Journal of Applied Psychology
• University of Nebraska — summary of Lacerenza et al. with percentage translations
• Taylor, Russ-Eft & Chan (2005) — meta-analysis of modeling-based learning
• Blume, Ford, Baldwin & Huang (2010), Journal of Management
• Cook et al. (2011), JAMA — simulation in healthcare professional training
• npj Digital Medicine (2025) — randomized trial on an AI-simulated patient
• Kestin et al. (Harvard, 2025) — randomized trial comparing an AI tutor with active learning
• Kluger & DeNisi (1996), Psychological Bulletin