Sample: AI Cost & Usage Review
What the report looks like: where the money goes, what each option would save, and where a cheaper model would answer worse.
This is a sample, not a client. The business in this report is made up, and so are its numbers, to show the format. It is not a real company and not a result we achieved.
Summary: Example Software Co., support assistant
- Requests a month
- 40,000
- Prompt size
- 4,000 tokens
- Answer size
- 300 tokens
- Current model
- Claude Opus 5.5
- Current monthly cost
- $880.00
- Share from repeated prompt text
- 70%
The usage is made up. The costs are calculated from vendors' published prices, checked 12 Oct 2026, using the same method as our free calculator. Savings are estimates, not promises.
Options, in the order we would try them
High priority
Turn on prompt caching: about $454.40 a month (saves about $425.60)
70% of each prompt is the same instructions and reference text. Answer quality should not change, because the model stays the same.
What to do: Cache the fixed part of the prompt, and check the vendor's cache-write charges against this estimate on the first bill.
Medium priority
Move to Claude Sonnet 5.5 with caching: about $227.20 a month (saves about $652.80)
A mid-range model may handle routine support questions. Accuracy has to be tested on your own questions first.
What to do: Test 50 real questions on both models, and switch only if the results hold up.
Low priority
Claude Haiku 5.5 with caching: about $11.92 a month. Cheapest, and the most risk
Said plainly: we would expect a small model to do worse on complicated questions. The saving is not worth wrong answers to customers.
What to do: Consider it only for simple steps such as sorting questions, and keep the larger model for the answer itself.
Want this for your own system?
$99 · £79. Usually 3–5 working days. You keep the report whether or not we work together afterwards.