An MIT report cited by Harvard Business Review on 1 July 2026 found that 95% of generative AI projects fail. A separate survey of more than 6,000 senior executives across four countries by the National Bureau of Economic Research found that roughly 90% reported no measurable productivity improvement from AI over the past three years. Two large, credible studies, the same uncomfortable conclusion.
The Actual Argument
Northeastern University professor David De Cremer, writing in HBR, makes the case that the failure is not the AI itself. It is what he calls the urgency trap: leaders reaching for AI as a quick fix for an immediate bottleneck instead of working out where it actually fits their business first. A tool bought under pressure to solve this week's problem rarely gets the planning it needs to solve next year's.
Why This Matters More for a Small Business Than a Big One
A large company can absorb a failed AI pilot as a rounding error in its budget. A South African SME spending R10,000 or R50,000 on an AI tool or integration cannot. The pattern behind most of the 95% failure figure is not bad technology, it is buying or building an AI feature because a competitor has one, or because the sales pitch was convincing, without first mapping which specific task it needs to fix and what success actually looks like.
What to Do Instead
Before adding any AI tool to your business, name the exact task it replaces and the time or cost it currently takes. If you cannot answer that in one sentence, you are not ready to buy or build it yet. This is not an argument against AI, plenty of the automation work we have covered here pays for itself in weeks. It is an argument against buying it the way most of that 95% did: fast, reactive, and without knowing what winning looks like.