What does a customer actually cost you?

Your OpenAI bill tells you the total. It doesn't tell you what one customer costs — or what happens when a heavy user lands on your "unlimited" plan. Set your numbers and find out before the invoice does.

Your numbers

Model — presets are approximate, edit to your provider's rate

Token price — per 1M tokens (input / output)

Interactions per active user / month — an "interaction" = one AI request

Tokens per interaction — input / output (include your prompt + context)

Heavy user vs. typical4.0×

Your p95 (power) user does this many times the usage of your median user. This is the one that blows up margins.

Target gross margin70%

The readout

Cost per interaction $0.021 1,500 in + 600 out tokens

P50 Typical user / mo

$6.30

median usage

P95 Heavy user / mo

$25.20

4× the median

Monthly cost to serve4× gap

Typical

Heavy

Price for the averagemargin target met on a typical user

$21.00

Price for the heavy user (p95)margin protected even on power users

$84.00

…

SET THIS ALERT

Flag any account above — interactions/mo

…

Then do one of three things within 48h: cap usage · reprice · talk to that customer

This is the number most AI kits never show you.

LaunchAI ships per-user token & cost metering built in — so you see this per customer, live, not once a month from your provider's invoice. One-time purchase, you own the code. Live on Product Hunt Aug 1.

Join the waitlist Follow on Product Hunt →

Model presets are approximate list prices and will drift — always set the price fields to your actual provider rate. This estimates the marginal AI cost to serve a customer; it doesn't include infra, storage, or fixed costs. p50/p95 model your usage spread, not a guarantee — the point is to see the shape of the risk before you price.