PerUnit
Track AI costs by customer, feature & tier for better margin
We’ve shipped products with AI and watched margins disappear without knowing why. The AI bill was one number — we couldn’t see which customers or features were driving it.
OpenAI and Anthropic dashboards show totals, not per-customer or per-feature costs. We couldn’t answer: Are free users profitable? Should we gate AI behind paid plans? Who should we charge more?
So we built PerUnit. It connects to your AI provider APIs (OpenAI, Anthropic, Google) and Stripe, and shows cost per customer, per feature, and per pricing tier. No data warehouse, no pipelines — just direct API sync and a dashboard.
We’re in early access and building with founders who care about AI unit economics. If that’s you, we’d love your feedback: perunit.ai
What would make this actually useful for you? What’s missing?
3 Comments
This is a real problem. We hit the same wall at our company - AI bill was one number, no visibility into which customers or features were driving it. Per-customer and per-feature attribution is the unlock. We use aicosts.ai for the multi-provider cost aggregation layer and it gives us per-model, per-team breakdown. The per-customer and per-feature piece is something we built on top by tagging our AI usage events with customer/feature metadata before pushing to the dashboard. Took maybe a day of dev work. Curious if PerUnit handles the multi-provider case natively or if it starts from a single provider. The hardest part for us was normalizing billing across Claude, GPT, and others into one schema before we could do the per-customer attribution. Good luck with the launch.
Very real pain point. I built a smaller consumer version of this for myself called TokenBar because I kept getting surprise Claude/OpenAI bills and just wanted live usage visible in the menu bar while I worked. Different angle, same problem. The cost is way too invisible until it is too late.
Here is the link
tokenbar.site