Open now — create your account and start measuring in minutes
Some of your customers are losing you money. Find out which.
Your AI SaaS charges flat plans, but token costs vary 10–20x between users on the same tier. TokenMargin shows revenue vs. real AI cost per customer — so you price on data, not hope.
Customers by AI cost — July
2 unprofitable customers| Customer | Plan | AI cost | Margin |
|---|---|---|---|
| cus_9f2k | $29/mo | $41.20 | −$12.20 |
| cus_x81a | $29/mo | $22.70 | +$6.30 |
| cus_m3p0 | $79/mo | $61.40 | +$17.60 |
| cus_t55r | $29/mo | $33.90 | −$4.90 |
| cus_b7qe | $9/mo | $1.80 | +$7.20 |
cus_9f2k costs 142% of what they pay you. Alert sent to Slack.
Your billing page shows spend. It doesn't show who.
Flat plans, variable costs
One 'generation' can cost $0.02 or $0.80 depending on context size, retries, and model. Two users on the same plan can differ 20x in what they cost you.
Caps set by gut feel
You picked '100 generations/mo' by guessing. Even Anthropic mispriced Claude Code and had to pull it from the $20 plan. Solo founders misjudge this constantly.
Margin erodes silently
You think you run 85% margins. Your power users quietly drag it to 60%. You find out when the invoice lands — or never.
Connect in 10 minutes
Tag your AI calls
Add our drop-in wrapper around your OpenAI or Anthropic client and pass a customer_id. One line of code. No proxy — your requests go straight to the provider.
See cost vs. revenue
Enter what each customer pays. The dashboard ranks every customer by real token cost against their monthly revenue.
Get alerted
Email or Slack ping the moment a customer crosses into unprofitable territory — before the month ends, not after.
Find your first margin leak today.
Create an account, follow the guided setup, send a test event, and see a real customer margin in your dashboard.
FAQ
I already cap usage per tier. Why do I need this?
Caps limit requests, not dollars. A single request can cost 40x another depending on context length, retries, and model. TokenMargin tells you whether your caps and prices are actually right — most founders discover they aren't.
Is this another proxy that sees my prompts?
No. The wrapper reports token counts and cost metadata only — never prompt or completion content. Your API calls go directly to OpenAI/Anthropic with zero added latency.
How is this different from Helicone or LLMeter?
Those answer 'how much did I spend?'. TokenMargin answers 'which customer is unprofitable?' — spend matched against revenue, per customer. It's a pricing tool, not just a meter.
What if I buy and it's not useful?
Full refund, no questions asked. You're buying early access from a solo founder building in public — the risk should be mine, not yours.