Atul Gawande had been a surgeon for eight years and had done more than two thousand operations. For years he had compared his results with national data, and he had begun beating the averages. His complication rates moved steadily lower. Then, a couple of years before he wrote about it in 2011, they stopped falling.
So he asked a retired surgeon who had trained him to watch him work. The old teacher stood on a step stool in Gawande’s operating room, taking notes, while Gawande removed a thyroid with a cancerous nodule. He had done the operation around a thousand times. It took 86 minutes and went beautifully.
The coach had rarely done this operation himself, and Gawande wondered whether he would find anything useful to say. Then they sat down in the surgeons’ lounge.
Twenty minutes against five years
The notepad was dense with small print. For about half an hour, the operating light had drifted out of the wound, and Gawande had not noticed. At various points his right elbow rose to the level of his shoulder. “You cannot achieve precision with your elbow in the air,” the coach said. That one twenty-minute talk, Gawande wrote, gave him more to work on than he had had in the past five years. In a later TED talk he said a coach gives you “a more accurate picture of your reality.”
That picture is the R in a coaching structure called GROW. Before you tell, ask what they want, what’s happening, what could work, what they’ll do. Goal: what would you like to achieve? Reality: what is happening right now? Options: what else might work? Will, or Way Forward: what will you do first?
GROW took shape in the 1980s and spread through John Whitmore’s 1992 book, Coaching for Performance. Credit is disputed: Whitmore’s firm, Performance Consultants, says he devised it with colleagues, while InsideOut Development says its founder, Alan Fine, co-created it. Whitmore, his firm says, famously said that “any dictator can use GROW.” Four headings don’t make a conversation coaching.
The question that scared him
Gawande’s coach had always taught this way. At Gawande’s first splenectomy as an intern, he waited for Gawande to draw the incision, then asked, “Is that really where you want it?” The cut proved too small. Years later he told Gawande to let his own residents struggle thirty seconds longer: “Get them to think.”
The tennis coach Timothy Gallwey, author of The Inner Game of Tennis (1974), built a method on a similar hunch. By his own account, he once had a student whose racquet went back too high swing in front of a mirror, then told him not to fix it, just to notice it. Over ten shots the take-back dropped from neck height to below the waist. “Before you try to change something,” Gallwey told Tennis.com, “become more aware of the way it is.”
Ride in a friend’s car to the same office every day and a year later you may still not know the way. A boss’s answers can work like that; one you worked out yourself comes with some of the map.
When telling wins
Beginners are different. A 2006 review by Paul Kirschner, John Sweller and Richard Clark argued that decades of studies show minimal guidance teaches less well than strong guidance, until learners know enough to guide themselves. Asking a new hire “what could you try?” about a system they have never seen is a quiz. Nobody needs a Goal question when the fire alarm is ringing.
What the studies can and can’t say
The evidence is for coaching in general, not for GROW in particular. A 2014 meta-analysis led by Tim Theeboom found positive effects on every outcome it measured, though its 18 studies mixed in life coaching and student samples.
A 2016 meta-analysis by Rebecca Jones and colleagues kept only workplace coaching, 17 studies, and also found positive effects. But it left out managers coaching their own staff, exactly where a manager would use GROW, and its authors noted that studies often didn’t say which coaching approach or model they used. Neither can say whether GROW beats other ways of coaching. It is a practitioner framework, a sensible order for questions, and the two meta-analyses cited here can’t validate GROW on its own.
The Friday report
Imagine a team lead whose best analyst keeps missing the Friday report. The obvious fix is an order: start Wednesday. Instead the lead asks what a good Friday would look like, then what happens on Thursdays now. The analyst says Thursdays vanish into quick requests from sales. Asked for options, she suggests a shared request form, and says she’ll set it up Monday. The plan is hers. Had she joined last week, the instruction might have been the kinder choice.
That conversation belongs in a regular one-on-one.
A tutor that won’t hand over the answer
A chat assistant asked for an answer will usually give one. A 2025 PNAS study by Hamsa Bastani and colleagues measured what that does to learning. Nearly a thousand students at a high school in Turkey practised math with no AI or with one of two GPT-4 tutors. One behaved like plain ChatGPT. The other was given teachers’ hints and the correct solution, and told not to give the answer but to guide students step by step.
During practice, the plain version raised scores by 48% and the hint-giving tutor by 127%. Then came an exam without AI. Students who had used the plain version scored 17% lower than those who never had AI. They had often asked for solutions and copied them, the authors found. The hint-giving tutor largely avoided that drop.
It was one school, one subject, four sessions in 2023, and the tutor also held the right answers, which helped its hints. What it shows is that this tutor design, with teachers’ hints and the correct solutions, largely avoided the exam drop seen with the plain version. It doesn’t show that software can coach. Gawande’s coach kept watching. On a video of a later operation, the light drifted out of the wound again. This time Gawande caught it in four minutes.