AI nearly doubles business output. And the main challenge is figuring out exactly where to apply it. I broke down a fresh large-scale study on how AI affects companies' real revenue.
Researchers from INSEAD and Harvard Business School ran a field experiment with 515 fast-growing startups. The authors (Hyunjin Kim, Dahyun Kim, and Rembrandt Koning) set out to test how solving the "mapping problem" — finding specific areas in the production process where algorithms create value — affects firm performance.
Half of the startups in the sample were shown successful case studies of how other companies had reorganized their work using AI.
After seeing this, the firms started identifying new AI use cases 44% more often (especially in product development and strategy).
Efficiency shot up: companies started closing 12% more tasks and attracting paying customers 18% more often.
In the end, the experimental group's revenue was 1.9x higher than the control group's. The need for external investment dropped by 39.5%, while labor demand remained unchanged (AI scales the output itself, not the cost of producing it).
Where exactly did AI drive this growth? The key insight from the study — AI should be embedded not in isolated micro-tasks, but in end-to-end systems. Here's what that looks like in practice:
🔸 Comprehensive development: algorithms don't just write chunks of code — they assist with rapid prototyping, from testing product hypotheses to collecting feedback and automatically adjusting features.
🔸 Complex ops: instead of basic auto-generated support responses, companies set up AI agents. These route emails on their own, scrape data from third-party sites, and update CRM statuses without any human involvement.
🔸 Strategy and analytics: AI was used to synthesize massive market datasets, identify non-obvious user behavior patterns, and stress-test business models.
The anomalous revenue spike shows up most clearly in the charts for top-performing startups (90th percentile and above). Meaning AI doesn't just marginally improve low-margin projects — it massively expands the ceiling of what's possible for market leaders.
Yet another study showing that AI genuinely improves productivity even at its current (far-from-perfect) capabilities. But the main bottleneck isn't weak technology — it's founders not knowing where to apply it.
Reminder that I cover exactly where to apply AI in my AI-Sprint! Link drop!
Just kidding (that was a parody vibe-coded project). But if you don't know how to start using AI effectively, write down your routine repetitive tasks (work calls/therapy calls, setting up Notion or build.in, gathering information/research, news digests, etc.).
And ask a good reasoning model (ChatGPT 5.4 Thinking is plenty) to help you build a plan for integrating AI into your routine.