The number is roughly right. The reading is wrong
The MIT statistic has already reached your board, compressed into one question: should we pause the AI program? That question cannot be answered at the altitude it is asked. What the study measured is not AI failure — it is the distance between what companies believed they bought and what the technology bills for before it pays. The authors themselves call their findings preliminary and directional; the viral retelling dropped every qualifier. This deep dive is NTT DATA’s methodologically honest reading of the number: what it can and cannot tell you about your own portfolio, the two opposite failure types behind every zero-ROI line, and the five-gate test to classify each initiative before your next portfolio review.
What the MIT NANDA study actually measured — its methodology, its self-declared limits, and the unanswered challenges to its data — and the 90-second board answer to “why does MIT say 95% fail?”
How to tell a comprehension failure (cured by information) from a calculation failure (cured only by changed incentives or authority) — and why the cheap remedy applied to the wrong one manufactures next year’s zero
The three patterns that fake success in the 5% — and the three questions that expose them: what was the baseline, what is the portfolio performance, what else changed in the same period
CIOs, VPs of Operations and Mining Directors in LATAM heavy industry who have read the MIT 95% statistic and are deciding whether to pause their AI programs or press forward
