90% startup lift claim

The viral “90% revenue increase” claim traces to a real March 2026 working paper, but the underlying result is that treated startups generated 1.9 times the revenue of a control group after learning how to redesign workflows around artificial intelligence. (insead.edu) The paper, “Mapping AI into Production,” was written by Hyunjin Kim, Dahyeon Kim, and Rembrand Koning and is listed by both INSEAD and Harvard Business School as a March 2026 working paper. It studied 515 high-growth startups in a field experiment. (hbs.edu) (insead.edu) Researchers call the bottleneck the “mapping problem,” meaning firms may have access to AI tools but still not know where in their workflow AI creates value. In the experiment, all firms got model access, application programming interface credits, and technical training, while the treatment group also got case studies showing how other companies had reorganized production around AI. (insead.edu) (hbs.edu) The treated firms found 44% more AI use cases, completed 12% more tasks, and were 18% more likely to acquire paying customers, according to the paper summary. The revenue gains were largest at the 90th percentile and above, which the authors say is consistent with AI expanding the upper end of performance more than lifting every firm equally. (insead.edu) The same paper says those faster-growing firms did not scale labor and capital at the same rate. Demand for external capital fell 39.5% relative to the control group, while labor demand was unchanged. (insead.edu) That is narrower than many social posts made it sound. The study did not say every startup that “uses AI” gets a 90% lift; it tested whether showing founders examples of AI-centered workflow redesign changed firm behavior and outcomes under experimental conditions. (insead.edu) (hbs.edu) The result also lands in a business climate where AI adoption is broad but firm-wide payoff is still uneven. McKinsey’s 2025 global survey said 88% of respondents reported regular AI use in at least one business function, while only 39% reported earnings-before-interest-and-taxes impact at the enterprise level, and it identified workflow redesign as a key trait of high performers. (mckinsey.com) Other Harvard Business School research points in the same direction on uneven outcomes. In a field experiment with Kenyan entrepreneurs using a GPT-4-powered business assistant, researchers found no average treatment effect they could distinguish from zero, while higher-performing entrepreneurs benefited by just over 15% and lower-performing ones did about 8% worse. (hbs.edu) So the clean version of the claim is this: a March 2026 startup experiment found that firms taught to rethink workflows around AI, not just bolt on tools, produced much stronger top-line results than a control group. The paper is real, the number comes from a working paper, and the effect is about 1.9x revenue in that experiment, not a universal rule for startups. (hbs.edu) (insead.edu)