AI Coaching Fails Without Human Intelligence
Originally published on Forbes Business Council - read the original here
Enthusiasm for AI coaching tools surges and crashes by the week. One moment they are the future of leadership development; the next, just another overhyped platform.
The truth is more useful than either position: They are a genuine supplement to coaching, not a replacement for it.
The adoption of AI coaching has nearly doubled in three years. In the latest Training magazine and Wilson Learning survey, around 40% of organizations report a high impact, and it's easy to see why. Managers like the feedback because it is confidential and safe.
But the honeymoon ends fast. AI coaching sessions sometimes only last a few minutes. Those minutes are often information, not coaching—and that is what you can expect when content is all there is.
The 10/20/70 Truth AI Cannot Escape
Of all the frameworks I have used in thirty years working on organizational effectiveness, I've seen the 10/20/70 model (based on Lombardo and Eichinger's 1996 book The Career Architect Development Planner) have the most impact on behavior and therefore on results. With this model, 10% of learning comes from content—models, frameworks, language. Twenty percent comes from socialization: the learner’s manager and their peers. The other 70% comes from doing the job—under pressure, on an ordinary Tuesday when the stakes are real.
AI coaching excels at the 10%. It is available at midnight, impossibly knowledgeable, endlessly patient and cheap. But that's still only 10%. Decades of research show that between 45% and 80% of what we teach never reaches behavior—a problem researchers named the Great Training Robbery 10 years ago. It has not touched the 20% or the 70%, where managers actually change.
The Boss, And The Peers Who Matter More
The best predictor of whether a manager changes is whether their own boss is in with them on the journey. The Training magazine survey also found that in high-performing organizations, managers are 77% more likely to actively support the people developing under them. An AI coach can tell you what good looks like, but you won't sweat its nudge the way you sweat a call from the boss. It can remind you; it cannot hold you accountable.
Peers matter even more. I have watched a manager sit unmoved through a workshop full of brilliant insights from the presenter, then change their behavior the next week because someone in their peer group tried the same uncomfortable thing and survived. "You too?" does work no chatbot can do. It makes the struggle normal and the change safe.
The machine cannot do that. Its opinion is only knowledge, so it carries little weight. The boss’s opinion carries weight but with implicit threat. A room of peers carries the weight of real experience without the threat. That combined weight is what behavior change needs.
Why The AI Coach Stays Generic
Human coaching works because the coach earns the right to push you. An AI coach never earns that, and it is not built to.
Even when it knows your name, team and goals, an AI coach stays generic in character. Oxford Brookes professor Tatiana Bachkirova puts it in stark terms: An AI coach paid for by an organization has no stake in challenging that organization. It exists to make you more aligned, more resilient, more compliant. She calls it identity regulation. I call it a coach with a conflict of interest. Coach Kim Witten puts it differently: It is like a horoscope, specific enough to sound personal and bland enough to fit anyone.
A real coach carries trust the chatbot has to fake. As my favorite coach told me 30 years ago, trust is the currency of change. Knowing your calendar is not the same as earning the right to push you to change.
Where AI Earns Its Keep
None of this means AI has no place; it just has to be aimed at the right thing.
AI does its best work supporting the 1:1, not replacing it. The weekly 1:1 between a manager and their report remains the central mechanism for oversight, guidance and development—where the standard gets set, drift gets caught and people get challenged and supported.
So, use AI to make that conversation better. A manager records a tricky exchange; the AI reviews it and reports what landed and what did not, then asks what they would do differently. The feedback still comes from the person whose opinion matters most to their career.
Put the 1:1 at the center, a peer group around it and an always-on AI layer that handles the 10% and preps managers for the 20%, while the leader stays the coach. The AI feeds the conversation but doesn't become the conversation.
The Only Number You Need
What matters is not how much people learn but how much they use when they go back to work.
Changing behavior takes three things: upskilling, coaching and measurement. The first two get the attention, but the third is where most programs fade.
But what should we be measuring? The answer is usage, not learning. The mechanism is a daily habit of plan, do, review. The learner plans the day, records what they actually did and lets AI do the heavy lifting: turning small daily entries into a clear read on what is changing and what is not. That summary feeds the next 1:1.
Do what you have always done, get what you have always gotten, but AI can now help you see, week by week, whether anyone is doing anything differently.
AI is most useful when it supports and enables better HI (human intelligence). The 1:1 has long been a critical point in management systems. AI allows managers to take this to the next level: better prepared, better supported and more human where it counts.