Every function in a company has a list of things AI could supposedly do. Most of those lists are fiction until something gets built.
My rule is simple. Take in the use cases, build the proof of concept fast, and let the result argue for itself. At Spin Master we set up a rapid PoC engine with each function. Marketing brought a use case, we built something real in days or weeks, and then everyone could see what the value actually was instead of debating it in a meeting.
Gartner found that at least 50% of generative AI projects were abandoned after proof of concept by the end of 2025. (Computer Weekly) Most of them didn’t fail in the build. They failed in the handoff from demo to production, which is exactly the gap a disciplined PoC process closes.
This does three things. First, it kills bad ideas cheaply, which is a feature, not a failure. Second, it gives good ideas evidence, which is what executives need before they’ll fund them. Third, and most important, it gives skeptics something to touch. A demo converts people faster than any slide deck.
The pattern I trust: intake, rapid build, honest evaluation, then scale only what earned it. Big-bang AI transformations fail because they scale the unproven. Start small, move fast, and let results do the persuading.