How AI solves coaching's three critical gaps

Abstract

Coaching works. Decades of research and plenty of hard-won experience back that up. Inherent in coaching are also three limitations:

1.    It’s dependent almost entirely on what the coachee chooses to bring to the room

2.    There is almost no way to credibly prove that sustainable change actually happened

3.    The value of each session leaks away in the long gaps between sessions

This paper covers:

·      Why coaching has never closed these three gaps, and what closes them now

·      Why AI coaches fall short of human ones when they try to work alone

·      What evidence-based, continuous coaching looks like in practice

·      What we’re testing at BTS, from real-time meeting feedback to organization-wide behavior tracking

Coaching’s three biggest historical limitations

Most of the AI coaching market is chasing the wrong fix: building bots that sound like coaches. AI won’t replace your coach. The real game changer now is that we can leverage AI to make coaching evidence abundant, continuous, and usable, pulling real signal from meetings, feedback, and everyday work, and turning it into something a coachee and coach can work with, both in the room and in the weeks between sessions.

Coaching has proven benefits. But three problems still go largely unaddressed in most 1-1 coaching relationships: how dependent the process is on what the coachee brings, how little evidence exists that real change actually happened, and how much of that benefit drains away in the long silence between sessions.

1.  Coaching runs on self-reporting

The coaching conversation runs almost entirely on the coachee's own view of change, their narrative, what they choose to bring into the room. We've tried to widen that lens with 360s and line-manager three-ways early in the engagement, but once that initial assessment is done, we're back to one person's perspective on their own world for the rest of the relationship.

Freedman and Perry (2010, p.200) found that when coaching relies on the client's face-to-face narrative alone, the leader ends up giving "a partial and heavily biased picture of what [is] taking place," and probably wouldn't have recognized how their behavior was perceived by the people around them. That blindness is well documented: Leadership IQ's 2026 study of 1,204 employees found that 58% of C-suite leaders overestimate their own performance, compared with 44% at other levels (Leadership IQ, 2026). Rely entirely on an executive's own account, and the engagement ends up solving an imagined problem, while the behaviors their team experiences every day go unaddressed.

De Vries (2025) writes extensively on the dangers of purely dyadic coaching. Without multi-rater feedback, he notes, coaching can narrow in on a leader's comfort zone instead of their actual impact. Ongoing checks are rare, they're a heavy administrative lift, so coaches end up relying on instinct to spot blind spots or discern whether real change is happening. Even good coaching can drift into being too "fluffy," too dependent on whatever the narrator brings, deaf to the real-time impact on the people around them.

2.  Progress happens in moments too small to evaluate at scale

Most coaching happens every two to four weeks, an hour at a time. Leadership happens all day, every day. Coaching's actual output isn't a document or a decision, it's a change in how someone behaves across a hundred small daily interactions, like delegating differently, giving feedback differently, or listening differently. None of that happens in the coaching session itself, it happens afterward, scattered across the ordinary workday, in a Slack message where a task gets handed off, in the tone of an email sent under pressure. The evidence isn't missing. It's everywhere, just never gathered in one place.

To actually see whether coaching worked, you'd need to observe those moments, not once, but across dozens of them, over months, at a volume no human coach could ever realistically sit in on. Coaching has always been built around one person watching, listening, and remembering, and a single coach, however skilled, was never going to sit in on every delegation, every feedback conversation, every tense meeting their coachee had that week. That's not a failure of effort, it's a failure of reach, the proof exists in a hundred scattered places, and nobody's ever been positioned to gather it.

3.  Coaching’s impact fades between sessions

A coaching session can produce a real breakthrough, a genuine shift in how someone sees a problem, a real commitment to try something different. But that breakthrough has to survive contact with four weeks of an unmanaged calendar before anyone finds out whether it stuck. Jason Norman, founder of Executive AI Partners, calls that stretch the "Execution Gap in Leadership," the space between knowing what to do and actually doing it under pressure. Coaching, as he puts it, is still largely episodic, while leadership is continuous (Gambill, 2026). That's not a failure of coaching. It's a failure of continuity.

We've tried to close that gap before, mostly with text messages, email check-ins, and "homework" between sessions, and there's some evidence it helps. But a 2026 review found workplace coaching lagging far behind therapy on this front and flagged inconsistent follow-through as a recurring problem (Wang et al., 2026). The trouble is that a text or a homework sheet only runs one way: it asks something of the coachee, then goes quiet. If they don't follow through, no one notices, asks why, or adjusts. The gap doesn't get filled. It just gets a reminder.

That silence is expensive. A freshly formed intention decays fast once the pressure of the working day takes over (Murre and Dros, 2015). Add four weeks of silence on top, and the coachee often arrives at their next session having half-forgotten what they committed to, with no record of what they tried or what got in the way. The coach starts again instead of building on last time.

‍