Send one month of usage totals. Get a written estimate in 48 hours.
Want a rough number first? Try the AI bill calculator. Before any call or contract, I'll look at your AI usage totals and tell you, in writing, how much of the bill is likely avoidable and what I'd test first. No charge, and no prompts, customer data or credentials needed.
Totals only
- One recent month of usage from your provider's dashboard: tokens or spend by model, ideally by day. A CSV export or a screenshot both work.
- Roughly what the AI does in your product, in a sentence or two
- If you self-host: your GPU type and count, the model, and the monthly GPU bill
- Optional: one typical prompt with anything sensitive removed, so I can see its shape
Please don't send API keys, customer data or full prompt logs.
A one-page written estimate
- Where the spend goes: by model, input versus output, and repeated context
- The changes likely to cut it, such as prompt caching, a cheaper model for easy requests, shorter outputs or batch pricing, each with a rough range
- What would need measuring on your traffic before any change ships
- An honest answer if there isn't much to save
It's an estimate from totals, not a measurement. Measuring on your real traffic is what a paid engagement does.
Your numbers stay private
Usage totals are used only to write your estimate. They arrive in ILLATE's email account, where Claude, the AI assistant I use, can read them, as described in the privacy policy. They're deleted 90 days after the estimate is sent, or sooner on request, and I'll sign your NDA first if you want one.
For a sense of what an estimate is based on, see Teardown #1, where prompt caching alone cut the API cost of a support workload by about 40%.
- Cost
- No charge
- Reply
- Within 48 hours, on business days
- You send
- One month of usage totals
- You get
- A one-page written estimate
- Obligation
- None