Cost attribution
Tie every AI charge to the key, model and provider that caused it, so every dollar has an owner.
What is Cost attribution?
Cost attribution maps AI spend to the API key, model and provider that caused it, so each charge has a clear owner. On Spro every request is recorded with its key, model and provider, so spend can be grouped by any of them and compared across date ranges. That makes AI cost something you can charge back and explain.
A monthly invoice that says only the total does not tell finance which team or feature to charge. Attribution breaks the total down by who made the requests and which model and provider served them.
Because the breakdown comes from the same request records as the logs, any line item can be traced back to the calls behind it.
Key features
What this feature does for your requests.
- Spend grouped by API key, which maps to an app or team
- Spend broken down by model and provider
- Compare spend between two date ranges
- Trace any line item back to its requests
How it works
4 steps, from setup to the first request.
- 01
Give each owner a key
Use one API key per app or team so spend has an owner.
- 02
Send requests as usual
Every request is recorded with its key, model and provider.
- 03
Group the spend
Break the total down by key, model or provider.
- 04
Compare periods
Find the owner behind a month-over-month change.
Is it the right fit?
Specific situations where this feature helps, and where it does not.
Good fit if
- Finance needs to charge AI cost back to teams or products
- Several apps share one account and each needs its own cost figure
- A cost jump needs to be traced to a specific model or provider
Not a good fit if
- One app with one budget owner, where the total is enough
- Keys are shared across apps, so the breakdown cannot separate their costs
- You need the exact amount on the provider's invoice. Costs use listed model prices
Cost attribution FAQs
Common questions about Cost attribution.
Route every AI request through SPRO.
One OpenAI-compatible API for 350+ models, with intelligent routing, 29 built-in guardrails, observability, and automatic failover.
