What did your AI actually buy you?
Most teams can’t say. The bill lands once a month, after the money is gone, and someone splits it across teams in a spreadsheet and calls it attribution. We think that is a strange way to run a business.
This is how it works today
every month someone takes the total bill and applies a distribution factor in a spreadsheet. I'm now fairly sure a couple of our biggest accounts are unprofitable and I can't prove it.
The UI for all of that spending was our credit card statement. Once a month. After the fact.
a team at 5k could be idle gpu boxes or 50 million tokens of real work. One is waste the other is the product and once its dollars i cant tell them apart.
Real posts, quoted as written. We collected a few hundred of them before writing a line of code.
Two things we do differently
The number ties out to the invoice
Nearly everyone computes cost as tokens times a price map. That drifts, and it flatters you, because a cost report derived from your own estimate can only ever confirm your estimate. We reconcile against what the provider actually billed. It is unglamorous plumbing and it is the difference between a number your CFO accepts and one they don’t.
Over budget shouldn’t mean stop
Everyone else’s answer to a spending cap is to kill the request. Two different people told us in one week that a cap which stops the work is unusable in production. We’d rather finish the job on a cheaper model. Your provider can’t do that — deciding what to complete more cheaply means knowing what it was for, and they have no idea who your customers are.
Where this actually is
Early. You can create an account, make an organization and send us usage data today. The per-customer margin reporting is not built yet, and we’re not going to pretend otherwise. We’re looking for a handful of companies who have this problem badly enough to help us get it right.