Annual AI budget planner
A single monthly API bill tells you little about where your AI spend is heading. Usage grows, but optimization claws some of it back. This planner projects a full 12-month forecast from your starting spend, your expected growth, and the efficiency gains you plan to ship — so you can put a defensible number in next year’s budget.
How it works
Each month combines two forces: growth pushes spend up, efficiency pulls it down. The planner applies both as compounding monthly rates:
month[n] = month[n-1] × (1 + growth_rate) × (1 − efficiency_rate)
annual = sum of all 12 months
exit_run_rate = month[12]Because both effects compound, small monthly rates produce large differences by December. A 10% monthly growth with no efficiency gain nearly triples spend across the year; a few points of monthly efficiency meaningfully flattens it.
Worked example
Suppose your current spend is £2,000 per month, you expect 8% month-on-month usage growth as your product ships new AI features, and you plan to squeeze 2% monthly efficiency gains from prompt trimming and response caching:
| Month | Spend (illustrative) |
|---|---|
| 1 | £2,000 |
| 3 | ~£2,300 |
| 6 | ~£2,800 |
| 9 | ~£3,400 |
| 12 | ~£4,100 |
| Annual total | ~£36,000 |
| Exit run rate | ~£4,100/mo |
Without the 2% efficiency offsets the annual figure would be materially higher — demonstrating why even modest optimization efforts have outsized year-end value.
When to use this tool
- Annual planning cycles — give finance a credible monthly cost curve rather than a flat extrapolation of today’s bill.
- Product roadmap decisions — model the cost impact of a new AI feature before you commit to building it.
- Comparing model switches — combine with a model-pricing tool to layer in the cost reduction (or increase) of moving to a different model midyear.
Tips and notes
- Plan to the exit run rate, not the average. If usage is growing, December’s spend is what carries into next year — budget for that number.
- Treat efficiency as a project, not a wish. Only count efficiency gains you have a concrete delivery plan for; otherwise the forecast flatters you.
- Re-forecast quarterly. Replace projected months with actuals as they land so the back half of the year stays honest. A Q3 actuals-vs-forecast diff often reveals a growth rate assumption that needs upward revision before December.
- Model the discontinuities. If you are switching models in month 6, run the planner twice (months 1–6 and 7–12 with different baseline costs) and add the two annual totals.
- Compound growth is non-linear. Eight percent per month is not 96% per year — it is roughly 151% because each month’s growth compounds on the previous month’s higher base. The planner captures this correctly; a spreadsheet extrapolation often does not.