When to use it
Use this prompt when your AI API bill has grown and you want a short, prioritized list of cost cuts. It works best with a simple table exported from your provider’s usage dashboard. Our list of seven ways to cut your AI API bill explains each lever the prompt asks about.
How to customize it
- [PERIOD] should be long enough to include normal peaks, such as 30 days.
- [PASTE USAGE TABLE] can be a CSV or a pasted dashboard view. Remove customer names and any personal data before sharing it.
- [HOW I JUDGE A GOOD ANSWER] is the most important field. Without a quality bar, every cost cut looks safe.
- Add the providers you already use if you want a comparison of their caching and batch options.
Why it works
The prompt forces a ranking before any advice, so effort goes to the workflows that matter. Separating levers that cannot change answers, such as caching and batching, from levers that might, such as routing and lower reasoning effort, produces a sensible order of work. The instruction not to quote prices avoids invented numbers, since prices change often, and the ASSUMPTION label shows which statements need checking.
Tips and variations
- Run it separately for each provider, then compare the plans.
- Ask for a spreadsheet formula to track cost per successful outcome rather than cost per token.
- Treat the answer as a hypothesis list and test each change on your own evaluation set before rolling it out.
- Never paste API keys or customer text into a chat tool unless your policy allows it.