What's an active customer? If you don't define it, AI will

Third meeting to define what an "active customer" is.

Bought in the last 30 days. No, 90. Opened the last email. Has the app installed (even if they haven't opened it since 2023).

Until recently this was an inside joke on any data team. A drag, but harmless: every department showed up to the committee with its own number, people argued for a while and life went on.

Buuut now there's a new guest at the table: the AI agent.

Every company wants to "talk to its data." You ask the copilot "how many active customers do we have?" and it answers in two seconds, with a gorgeous chart and total confidence. Using which of the fourteen definitions? Whichever one it found first. Or one it came up with on its own, which is worse.

And that's where the joke stops being funny. A person, when in doubt, asks. The agent quietly picks a definition, uses it to segment, to send a campaign, to decide who gets the discount… and nobody finds out the criteria changed.

I recently wrote on LATAMIA about companies that forget why they make their decisions. With agents deciding on their own, that forgetting goes from taking years to taking weeks.

That's why that incredibly boring thing called the semantic layer (the official dictionary of what each number means) has finally had its moment. It used to be a luxury for well-organized teams. Today it's what separates a useful agent from one that makes things up with a straight face.

If humans can't agree on what an active customer is, AI is going to decide for them.

And heads up: the definition can be any of the fourteen. What matters is that there's just one, written down, and that the agent reads it before answering.

Have you asked your copilot how many active customers you have yet? Try it and compare it with the committee's number.