# Budgets, Quotas and FinOps for AI — AI Safety, Evaluation and Cost Control

Source: https://www.geekswithgeeks.com/en/ai-safety/cost-budgets

> Set spend limits, allocate costs to features and alert before budgets are exceeded.

## Make spend visible and bounded

Tag each request with the **feature, team and environment** so you can see who spends what. Set **hard limits** (provider-side spend caps, per-key quotas, per-user token limits) and **soft alerts** at, say, 50%, 80% and 100% of the monthly budget. Use separate API keys for development, testing and production so a test script cannot drain the production budget. Review the bill regularly and ask whether each large cost is buying value.

## A budget policy

Write it down so alerts and limits have owners. Numbers here are illustrative.

```text
Monthly AI budget      1,000 USD  (support assistant 600, search 250, internal 150)
Soft alerts             50% / 80% / 100%  -> Slack #ai-costs + feature owner
Hard limits             provider spend cap at 110%; per-user 200k tokens/day
Keys                    dev, staging, prod are separate; prod key not in CI
Monthly review          cost per resolved ticket; top 5 costly prompts; routing hit-rate
```

**Quiz:** Why use separate API keys for development and production?

- [x] So a test script cannot drain the production budget, and spend is attributable
- [ ] Keys must be unique by law
- [ ] It makes the model faster
- [ ] It removes the need for budgets

*Answer:* So a test script cannot drain the production budget, and spend is attributable. Separate keys isolate environments and make costs traceable.