The study defines ‘mapping problem’ as the difficulty firms face in identifying where AI can create value within their workflows and how to reorganise processes to translate task-level improvements into broader business outcomes.
“The goal was to push firms beyond obvious AI use cases — not just email or chatbots, but rethinking workflows, products, and even business models. Since best practices are still emerging, examples can spark experimentation to help firms discover new ways to organize around AI,” wrote Hyunjin Kim, the study’s lead researcher from INSEAD, in a post on X on Saturday.
The field experiment involved 515 high-growth startups that received standard training through INSEAD business school’s three-month AI Founder Sprint accelerator. Half of them were additionally shown case studies of similar companies that had reorganised their operations around AI.
The study showed that treated firms, which received peer examples, were more likely to experiment across a broader set of activities, suggesting that the challenge lies in discovery rather than access. The largest gains were concentrated among top-performing firms, indicating that AI may expand the upper bound of success rather than uniformly improving outcomes across all firms.
Treated firms were thus able to resolve the mapping problem, leading them to use AI 44% more, identifying a broader range of use cases, particularly in product development and strategy. These firms also completed 12% more tasks and were 18% more likely to acquire paying customers. The treated firms also required 39.5% less external capital, amounting to over $220,000 less on average, while maintaining similar employment levels. This indicates that AI adoption enabled startups to increase output without proportionally increasing inputs.
Commenting on the experiment, OpenAI co-founder and president Greg Brockman wrote in a post on X, “AI use is an emerging skill which improves businesses and unlocks entrepreneurship.”
The findings suggest that current approaches focussed primarily on access may be insufficient and that enabling firms to explore use cases and rethink workflows is critical to realising AI’s economic potential.
