For teams with an AI product in production, or about to ship one
I put a number on cost per request, per active user, and per subscriber, while it can still change the model, the architecture, or the price.
Model selection, infrastructure, and margin are the same decision. Most organizations find that out after launch, when the inference bill arrives and the pricing page is already public.
The usual ways of answering come up short. The cloud bill gives a total with no unit behind it. The vendor’s pricing page times a token estimate is a guess that stops being true the first time usage shifts. Finance has the spend but not the drivers, and engineering has the drivers but not the spend. Nobody can say what one more subscriber costs.
This is FinOps for AI products: usage mapped to cost at the feature level, a model that runs before a line of code is written, and a forecast the finance team keeps running after I leave.
Unit economics
Cost per request, per active user, per subscriber, tied back to feature-level usage so pricing and roadmap calls have a number behind them.
Built for the top three AI products in a global financial data company’s portfolio, with FP&A now running the pacing and forecasting unassisted.
Pre-build modeling
Model selection, token splits, caching efficiency, and break-even, worked out before engineering commits to an architecture.
An LLM cost calculator built at the request of a global financial data company’s CFO, CTO, and CPO, used to model a feature’s economics before a line of code is written.
The cost floor
When the AI product sits on a larger cloud estate, I work the forecast from the drivers down, so the infrastructure decisions that set the floor get made with the same numbers as the product ones.
Five cloud platforms and 600+ engineering teams under one forecast, held to within 5% of actual.
in data, ML, and platform product
of annual AI and cloud spend under measurement
users served by an ML platform I owned
Financial data · media and streaming · higher education · regulated gaming · energy markets
I run two businesses of my own. It’s where the economics stop being theoretical.
A digital agency for home-service businesses across Maryland, Pennsylvania, and Virginia. I founded it and run it with two cofounders.
A meal-planning app I built for my wife. Other people started using it, so I kept going.
If you are shipping something with a model behind it and nobody can tell you what it costs, that is the conversation I want.
Full work history on LinkedIn.
Currently consulting. Open to the right problems.