📧 Understand and Estimate LLM Costs
A pile of finance email, a language model, and nobody who has yet worked out the bill.
Your Brief
Your finance team reads a steady stream of email that somebody has to summarize — month-end close updates, expense reports, regulatory notices. Roughly ten thousand a month at the volume you will start with. Someone has asked what it would cost to hand that job to a language model, and nobody has priced it before.
You will build the answer from the bottom up: one email, then a month, then the decisions that change the number. By the last exercise your CFO has constraints and you are designing to them.
What you will do
- What It Consumes — you are here. Price one email summary in tokens, input and output counted separately, then turn it into a monthly bill.
- Right-Sizing — route three kinds of email to the model each one actually needs, rather than the best one available.
- Volume — watch the bill scale, and check you can forecast a volume you have not measured.
- Cost of Being Wrong — find the point where the cheap model stops being the cheap option.
- Your Call — design a routing mix that satisfies a budget and a quality floor at the same time.
All five are scored, eleven questions in all. The sidebar keeps the count, and your answers are remembered if you leave and come back.
Exercise 2 · Right-Sizing Your Model Choice
Route each email type to a model
Drag a model chip onto each email type — or select a chip and press Enter on a target.
Exercise 3 · What Volume Does
Exercise 4 · The Cost of Being Wrong
Cheaper models miss things. A missed detail in a finance email costs someone time to catch and fix. Token cost is only half the bill — total cost = tokens + remediation. Move the sliders and watch the winner change.