Your AI pipeline, costed and mapped.

$1,500 fixed. Written report. Async. One week.

You are running documents, tickets, calls, or code through a model, and you cannot say what one unit of that work costs today, what it would cost on a different model, or which change to the architecture actually moves the number. Most teams answer this with a spreadsheet of list prices. That is not the same thing as counting your own tokens.

This is a fixed price engagement that counts them.

What you get

• Your sample documents, token counted. Not extrapolated from a typical page. Counted on the files you actually process, with the counts shown so you can check them.
• Your volumes priced across every model you run or are considering. Claude, GPT, Gemini, open weight models on hosted inference. The method is model agnostic and the report names every provider it priced, including the ones it ruled out.
• Three to four architecture options, each projected to a monthly cost at your volume. Model tiering, prompt caching, batching, and extraction strategy are the levers that usually move the number, so those are the ones priced against each other.
• Every number traceable. Each figure resolves to either your own samples or a dated, cited entry in our provider price ledger. No round numbers somebody remembered.
• A written report you can forward to whoever signs off, with the assumptions stated where you can argue with them.

How it runs

• You send the inputs. Your stack, the models you use or are considering, monthly volumes, and a handful of representative sample documents. Over email, on your schedule.
• We count and price. Tokens measured on your samples, volumes projected, architectures compared against a dated price ledger.
• A person reads it. Every report is read end to end by a human who writes the note on top before it ships.
• It lands in your inbox inside one week. No calls, no meetings, no slide deck.

Price

$1,500. Fixed, one payment, invoiced. No hourly rate, no scope creep clause, no retainer attached to it.

Disclosure

The analysis engine runs on Claude. When a recommendation lands on Anthropic, the report says so, in the report, next to the recommendation. You should discount any provider recommendation from a tool built on that provider unless it shows you the arithmetic. That is the reason every number in the report is traceable to a source you can check.

A person reads every report

The counting and the pricing are automated. The review is not. A human reads each finished report end to end and writes the buyer specific note on the front before it goes out. Nothing here is auto sent.

What happens to your data

Your documents and your numbers stay private. We keep an internal record of each engagement so later assessments are calibrated against measured runs instead of estimates, and only anonymized aggregates above a minimum count are ever published. No client is named, and no single engagement is identifiable in anything we put out.

After the report

Model prices move, and when they do your numbers go stale. We can re-run your profile against the updated ledger and tell you what changed, so ask about monitoring if you want that standing.