> ## Content Index
> Fetch the complete content index at: https://www.howardgray.net/llms.txt
> Use this file to discover other available public pages before exploring further.

# 515 startups and the case for case studies
- URL: https://www.howardgray.net/515-startups-and-the-case-for-case-studies/
- Published: 2026-09-03T11:59:22.000Z
- Updated: 2026-09-03T11:59:22.000Z
- Description: Training gives you capability. A case changes what you think you are allowed to ask for.
- Author: Howard Gray
- Tags: A.I

515 startups were handed identical AI tools, identical credits and identical technical training. One group did one thing differently. And they made 1.9 times the revenue.

Their workshops were case studies - how AI-native companies had rearranged their work around the technology.

Everything else was the same, including the hours in the room. The control group's workshops covered general entrepreneurship instead.

The case-study firms found 44% more uses for AI than the control group, mostly in product and strategy rather than the places everyone looks first. They finished more work. They were 18% more likely to win a paying customer.

This is from a preregistered randomised trial out of INSEAD and Harvard, which is not a sentence I expected to write about workshops. \[1\]

But why did it work?

Training tells you what the tool does. A case shows you what somebody like you decided to do with it, including the bits they got wrong. One gives you capability. The other changes what you think you are allowed to ask for.

I've been arguing this for years - but mainly from lived experience and a decent but small set of metrics.

Now, someone has run the experiment properly.

These were startups, of course, and a startup can rearrange itself over a weekend in a way a global bank cannot. Take the size of the number with salt. But the overall direction stands up.

Most corporate AI programmes are still built as training. Access, a prompt library, a lunch and learn, a dashboard counting who logged in (somebody will screenshot it for a board deck).

Both groups had the tools and knew how to use them. The 1.9x went to the group that had seen what they were for.

P.S. Playable case studies and scenarios are what Wavetable builds - AI, customer experience, storytelling - so of course I would say all this. Happily, the experiment is someone else's.

\[1\] [Kim, Kim & Koning (2026), "Mapping AI into Production: A Field Experiment on Firm Performance](https://papers.ssrn.com/sol3/papers.cfm?abstract%5Fid=6513481&ref=howardgray.net)", INSEAD / HBS working paper, March 2026\. Verified 2026-08-24.