Your AI Feature Might Lose Money Every Time Someone Uses It

Your AI feature might lose money every time a customer uses it, and most founders have no way to know.

Regular software has one thing going for it: once it’s built, every extra use is basically free. AI features broke that. Every time your product calls OpenAI or Anthropic, you pay. A customer who loves your AI feature and uses it fifty times a day might be your most expensive user, not your best one.

On a recent AI product build, I put metering in on day one. Every generation records its token count, its cost, how long it took, and whether it succeeded, using OpenTelemetry metrics on the server. It took maybe a day of work. Now when we consider a different model or a longer prompt, I can tell you what that change does to the cost of a single lesson before the bill arrives. When usage grows, I know whether margin grows with it.

Most teams skip this. The API bill shows up as one number at the end of the month, nobody can connect it to anything, and the pricing page was set by gut feel.

Ask your team one question: “What does one use of our AI feature cost us?” A team that’s paying attention gives you a number within a day. A shrug means your margin is a guess.

If you’ve priced an AI feature, what did you base the number on?

#FractionalCTO #StartupTech #TechLeadership