Graham McNicoll

Every experiment you run produces a piece of knowledge about your users. Run a few hundred, and you have built something genuinely valuable: a record of what works and what does not for your specific product. Here is the problem... A single person cannot hold the findings of a thousand experiments in their head. So teams often relearn things they already knew. Someone proposes an idea that was tested and killed eighteen months ago. A pattern that wins every time gets rediscovered by a new hire who had no way of knowing it was settled. The knowledge exists, sitting somewhere nobody looks. This is one of the places AI changes the math in a way that actually helps. A model can read across every experiment you have ever run and surface the patterns. Pulled into context through an MCP server or API, those learnings can inform the next thing you build before you waste a cycle testing it again. The real asset an experimentation program builds is the compounding record of everything you have learned. The teams that can leverage these learnings will outperform teams that don't. If you have run hundreds or thousands of experiments, do you document them well? Is the document actually usable for future projects?