Are Your Free Users Profitable? How to Check AI Cost Against Stripe Revenue
Total AI spend is up again this month. Stripe says MRR is fine — maybe even pretty good. In standup, someone mentions a noisy customer on the free tier, everyone shrugs, and you move on. Then, usually late on a Thursday, someone drops the question in Slack: "Are we actually making money on free users, or are we just hoping paid covers it?" The channel goes quiet. People start pasting screenshots of the OpenAI usage page. Nobody has a single number that sits next to Stripe for the same account.
That silence is the whole problem. Your provider dashboard is honest about what you owe them. It is not built to tell you what you spent on workspace 8821 versus workspace 4410. Stripe is honest about who paid you. It does not know that one of those workspaces burned through forty dollars of tokens last week and the other burned forty cents. Until you put those two truths in the same row, "profitable" is a vibe, not an answer.
You do not need a perfect data model on day one. You do need to stop pretending that one big AI line in the P&L means the business is healthy. It might be. Or you might have a thin slice of accounts doing to your margin what one noisy neighbour does to a shared wall — you feel it everywhere, but you cannot point to the exact spot yet.
The minimum math (really)
Pick a month. For each customer account — however you define "account" in your product — you want two rough numbers: what you spent on AI to serve them, and what they paid you. The AI side should come from usage you can tie to a stable ID, the same one you would use if you had to refund someone or answer a billing dispute. The revenue side is whatever Stripe (or your billing tool) says: subscription, seat, usage line items, invoices, all of it.
The first pass will be messy. Some events will be untagged. Some costs will land in a bucket called "unknown" and make you wince. That is still better than only looking at the company total, because the company total cannot tell you whether your free tier is a marketing expense you can afford or a leak you cannot.
Free accounts are zero subscription revenue by definition — so the only question is whether the rest of the business subsidises them at a level you are willing to live with. Paid accounts should, in theory, carry more revenue than AI cost per account. In theory is doing a lot of work here. We have seen teams where a handful of "Pro" seats were quietly underwater because those customers treated the product like an unlimited API playground.
Why the blended average lies to you
Leadership loves a single KPI. "AI is eight percent of revenue" sounds manageable until you realise that number is a blender. Twenty thousand people costing you two dollars each in AI and twenty people costing you two thousand each can produce a tidy headline in the middle. The strategy you need for those two situations could not be more different — and the average will not tell you which world you are in.
What you actually want is the shape of the curve: cost and revenue by tier, and within each tier, the tails. Who are the expensive accounts with almost no revenue attached? Who are the quiet, profitable ones you should not annoy with sudden limits? Those names rarely show up in a provider console. They live at the intersection of usage metadata and billing — which is exactly where most teams have not wired anything yet.
What we usually do once the picture is clear
Nobody needs a consultant to say "cut costs" or "raise prices" once they can see the map. The uncomfortable accounts float to the top on their own. Sometimes the fix is tighter free limits. Sometimes it is moving a heavy workflow behind a paid plan. Sometimes it is realising your "Enterprise" discount was generous in a world before AI usage exploded, and the contract needs a conversation, not a feature flag.
You still will not have perfect attribution on week one. That is fine. What is not fine is another quarter of arguing from totals while the bill keeps climbing. A rough honest split beats a precise fiction.
If you want to stress-test your gut before you buy anything, our free AI margin calculator and AI cost benchmark help. If you are ready to keep AI cost and Stripe revenue in one place without building a warehouse first, PerUnit is what we built for that — customer, feature, and tier, updating as your providers do.