ForecastingBudgetAnthropicAI Agent Skill
forecast
Forecasts budget or cost based on real usage trends, with a confidence range instead of a single point estimate.
Install Options
Install into Claude Code, Cursor, Codex, or Antigravity with:
$ npx skills add anthropics/knowledge-work-plugins@forecast
Real Work Situation Solved
Answer the compute/API cost question on the spot, no engineer needed
“Leadership asks "how much will this AI feature cost per month" mid-meeting - what do you estimate with?”
5-Step Action Framework
- 1Estimate monthly API calls: active users × how often each one uses the feature.
- 2Estimate average tokens per call (input + output) - round numbers are fine, you don't need exact precision.
- 3Multiply calls × tokens × the model's token price to get the base cost.
- 4Add a buffer (typically 20-30%) for retried failed calls or traffic spikes.
- 5Compare the estimate against a known budget threshold - if it's over, say so directly instead of just handing over a number.
Before
"I'm not sure, let me check with engineering."
After
"10,000 active users × 5 calls/month × 1,500 tokens/call × $0.002/1K tokens ≈ $150/month, plus a 25% buffer ≈ $190/month - within the approved $500 budget."
Related Course Lesson
AI Product Management • Lesson 14
SKILL.md Source Instructions
SKILL.md • Read-only preview# Forecast Skill Project cost or usage forward from historical trend data, and express the forecast as a range with stated assumptions rather than a false-precision single number.