AI adoption isn't an AI problem

I went into 15 interviews with UK B2B marketers expecting to measure something close to a single number: how much AI has this company adopted.

By the third or fourth conversation, that question had stopped making sense.

The same company could sound completely different depending on which room you asked in. In the language a company used publicly, AI was already central to how the business worked. In the actual interview, with someone doing the work, the picture was thinner — a pilot here, a policy nobody had actually used.

That wasn’t dishonesty. It was two different jobs being done by the word “AI.” One version was there to reassure a board or a client. The other was whatever was actually running in a workflow, checked against whether it saved anyone real time.

So the research question changed. Not “how much AI adoption,” but “which conversation is this — the one for the outside, or the one for the inside — and how far apart are they?”

That produced a spectrum instead of a score. A handful of companies had folded AI into routine, ROI-checked work so thoroughly that people had stopped calling it “AI.” It was just how the job got done. At the other end, some had an active public AI story with comparatively little happening underneath it. Most moved between the two, depending on what the market around them was doing that quarter.

The finding that stuck with me wasn’t about any single company. It was that “are you using AI” is close to a useless question, because it assumes one honest answer exists. The more useful question is which of the two conversations you’re being handed, and how much distance separates them.

I think about that distance a lot now, whenever I look at what a company, including my own projects, claims to be doing with AI versus what a person inside it would actually tell you if you asked properly.