Weekly AI spend reviews are easier when the numbers come pre-formatted. Paste your per-model spend for this week and last, and this tool produces a clean Markdown digest — totals, week-over-week change, per-head cost — ready to drop into Slack or email.
How it works
You enter usage as simple model = amount lines for two weeks. The tool:
- Parses each line into a model name and dollar amount, ignoring anything that does not match.
- Totals each week and computes the overall week-over-week percentage change.
- Diffs per model — change for models in both weeks, plus notes for models added or dropped.
- Divides this week’s total by team size for a per-head figure.
- Renders a copyable Markdown report.
All of it runs locally in your browser, so internal spend data never leaves your machine.
Example digest
For gpt-4o = 142.50 and claude-sonnet = 88.00 this week versus gpt-4o = 120
and claude-sonnet = 95 last week, team of 6:
## Weekly AI Cost Digest
Total this week: $230.50 (last week $215.00, +7.2%)
Per head: $38.42
- gpt-4o: $142.50 (+18.8%)
- claude-sonnet: $88.00 (-7.4%)The +7.2% headline plus the per-model breakdown immediately shows where the increase came from.
What makes a good weekly AI cost review
A useful digest answers three questions in under two minutes of reading:
- Did total spend go up or down? The week-over-week headline tells you immediately.
- Which model drove the change? The per-model diff isolates the cause — a spike in one model is a different conversation than a distributed rise across all of them.
- Are we spending proportionately per person? Per-head cost spots the difference between a team that grew and a team where individuals simply used AI more.
Common patterns to watch for
A single model spikes sharply (+30% or more). Usually means a new feature landed that uses a powerful model for all requests, or someone left a debug loop running. Worth investigating before it compounds.
Total cost rises but per-head cost stays flat. The team is growing, not overusing AI — a healthy signal. No action needed unless the budget ceiling is near.
A model disappears week-to-week. Shows up in the digest as “dropped.” If it was intentional (a migration to a cheaper model), good. If it is unexpected, check whether an integration silently broke.
Per-head cost climbs steadily over several weeks. Gradual creep can be harder to spot than a spike. Weekly digests make the trend visible before it becomes a budget issue. Typically solved by reviewing which workflows actually need a frontier model versus a faster, cheaper one.
Tips
- Keep the same model names week to week so the diff lines up; renames show up as one model dropped and another added.
- Track per-head cost over time — a steadily rising figure is an early signal to review usage patterns or switch models.
- Paste the digest directly into a recurring Slack message or wiki page so the weekly compare is always in the same thread.
- Pair with the LLM API Cost Calculator when you need to forecast next week rather than report last week.