Which customers lose us money?
Your observability tool shows tokens. It doesn't know what anyone pays you.
AI cost accounting for finance teamsKnow what AI
Know what AI
actually costs you
Costling turns your OpenAI and Anthropic bills into cost per customer, gross margin impact, and a clean month-end close. Built for finance teams, not engineers.
See your unprofitable customers in 10 minutes. No engineering required.OpenAIAnthropic
Illustrative data
Overview
AI share of COGS31%
Unprofitable customers14
Sept forecast$56,900
| Customer | AI spend | Revenue | Gross margin | Status |
|---|---|---|---|---|
| Northline Freight | $4,784 | $4,752 | (0.7%) | Unprofitable |
| Harbor Vet Group | $2,855 | $588 | (385.5%) | Unprofitable |
| Brightcove Labs | $1,850 | $7,200 | 74.3% | Profitable |
The problemYour AI bill is one line item.
Your AI bill is one line item.
Your questions aren't.
Is AI eating our gross margin?
The invoice lands in COGS as a lump sum. The board deck needs it split by product.
What will next month cost?
Usage-based pricing means forecast and actual drift 30 to 50 percent apart.
How it works1 2 3
Three steps from invoice to close
Connect
Pull spend from your providers, tags from LangSmith, Helicone or Langfuse, revenue from Stripe.
Allocate
Attribute every request to a customer, product and team. Write rules for anything untagged, the way you allocate shared costs today.
Close
AI COGS by product, margin by customer, journal entries mapped to your chart of accounts. Export to Excel or push to your ERP.
For finance leaders at companies spending $20k+ a month on AI
If you're a CFO, VP Finance, or fractional CFO whose AI spend just became a board question, Costling gives you the answer without asking engineering to build a dashboard.
Pricing
Simple pricing, sized to your spend
Starter
$499/mo
- Up to $50k AI spend
- CSV imports
- Excel export
Growth Most popular
$1,499/mo
- Up to $250k AI spend
- API connectors
- NetSuite and Stripe
FAQ
Frequently asked questions
Do we need engineering to set this up?
No for the CSV version. Tagging requests in your logs makes attribution sharper, but allocation rules cover the rest.
Which providers do you support?
OpenAI and Anthropic today, plus any spend that flows through LangSmith, LiteLLM, Helicone or Langfuse. Google is next.
Is this an observability tool?
No. Those tell engineers where tokens go. Costling tells finance what they cost, per customer and per product, in the format your close needs.