Same Advice, Opposite Results. Guess What the Variable Was.
On two field experiments in East Africa: one with an AI business adviser, one with human coaches.
Picture two small-business owners in Kenya. Both get the same adviser. Both ask it questions, and both get answers. Five months later, one business is doing better by around 15 percent. The other is doing nearly 10 percent worse.
That is roughly what happened, on average, to two groups of entrepreneurs in a recent experiment. The adviser was an AI. Next door in Uganda, a different experiment gave entrepreneurs human coaches, and the results ran the opposite way. Put the two side by side and they say something useful about how to be a mentee.
What happened with the AI adviser
Nicholas Otis, Rowan Clarke, Solène Delecourt, David Holtz and Rembrand Koning gave 640 Kenyan small-business owners an AI business adviser, built on GPT-4 and delivered through WhatsApp, which they could consult as often as they liked. A comparison group received business training guides. The study appeared in Management Science in 2026.
Across the whole group, the AI made no detectable difference to revenues or profits. The average hid a split. Owners who had been performing above the median before the experiment may have gained more than 15 percent. Those below it did nearly 10 percent worse. The researchers found that the gap did not come from the questions people asked or the quality of the answers. It came from which pieces of advice the owners chose to put into practice.
What happened with the human coaches
Stephen Anderson and his colleagues ran a randomized experiment with 930 entrepreneurs in Uganda. Half were matched with volunteer coaches, experienced managers based mostly in advanced economies, who worked with them remotely by video call, email and phone. Coached entrepreneurs increased their monthly sales by about 28 percent on average and were far more likely to change direction in their marketing.
The detail that matters most is who gained. The effect was larger for entrepreneurs who began with the least strategic focus. In a companion study, coaches with marketing backgrounds pushed owners to research competing products, rethink what they offered and ask customers what they wanted. The owners did that work themselves. A maker of rice sauce added a spice blend that customers preferred. A mechanic stopped trying to fix everything and concentrated on quick oil changes and tire rotations. Both businesses grew.
Why the results point in opposite directions
The studies were not designed to be compared, so treat what follows as a plausible reading rather than a proven one.
An AI adviser answers the question you bring. If you already know which question matters, that is a fast and capable service, and the strong performers in Kenya seem to have used it well. If you are not sure what your problem is, you get a confident answer to the wrong question, and acting on it can leave you worse off.
A human coach works from the other end. A good one listens, notices what you have not mentioned and decides which question deserves the time. The Ugandan entrepreneurs with the least focus had the most to gain from someone doing that for them, and they gained the most.
What this means for you as a mentee
The lesson for a mentee is to match the problem to the adviser.
When you know exactly what you need, a chatbot is a sensible first stop. It will draft the email, check the numbers and explain the jargon at any hour. Save your mentor’s time for something better.
When you are stuck and cannot say why, take that to your mentor, untidied. The temptation is to arrive with a polished question so that you look in control. Resist it. The Ugandan results suggest the vague, unfocused problem is where a human adviser earns their keep, and a mentor cannot diagnose a problem you have already disguised.
Then do the work yourself. The coached entrepreneurs in Uganda researched their own competitors and changed their own products. The Kenyan owners who did worse were the ones who acted on advice that did not fit their situation. In both cases the outcome turned on what happened after the conversation ended.
Finally, check which kind of mentee you are this month. Clear on your problem, use the tools and move quickly. Lost, find a person and say so.
The honest conclusion
Ask a chatbot and it will help you run faster in whichever direction you are already facing. Ask a mentor and they may mention that you are facing a wall.
Robert Rosenfeld is the author of The Accidental Mentee: Conversations Worth Showing Up For and From Manager to Mentor: Conversations You Cannot Rehearse. [Publication date and where to buy to be confirmed.]
Source: Nicholas Otis, Rowan Clarke, Solène Delecourt, David Holtz and Rembrand Koning, “The Uneven Impact of Generative Artificial Intelligence on Entrepreneurial Performance: Evidence from a Field Experiment in Kenya,” Management Science (published online July 10, 2026). Read the original here.
Source: Stephen Anderson and colleagues, field experiment on virtual coaching for entrepreneurs in Uganda, Marketing Science [year and full citation to be confirmed]. Read the original here.
