Ai billing models
Skill alexclowe/awesome-claude-cowork-plugins/finops-practitioner/skills/ai-billing-models
AI vendor billing-model fluency — auto-activates when normalizing, comparing, or forecasting AI spend across seats, usage, credits, and reserved capacityFrom its SKILL.md
npx -y skills add alexclowe/awesome-claude-cowork-plugins --skill ai-billing-modelsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
2.6 KB, 513 tokens by cl100k_base, as published. Nobody here has run it
You have deep fluency in how AI products bill. When the user is working on AI spend tasks, apply this knowledge automatically.
The billing shapes
- Per-seat subscription — flat monthly/annual per user (chat assistants, copilots, embedded AI tiers). The waste mode is inactive seats; the control is seat-to-active-user reconciliation on a cadence
- Usage-based API — metered per token, request, image, or minute, usually with different input/output rates and per-model pricing. The waste modes are retries, oversized prompts/contexts, and defaulting to a bigger model than the task needs
- Prepaid credits — money in, credits out, often with expiry and sometimes with bonus tiers. The waste modes are expiry (breakage) and the false comfort of "already paid" masking a rising run rate
- Reserved / provisioned capacity — committed throughput (e.g., provisioned units) billed whether used or not. The waste mode is committing to peak instead of base load; the control is utilization tracking against the commitment
- Hybrid — seat + usage overage, platform fee + credits, or an enterprise minimum with drawdown. Read the invoice's fixed and variable parts separately or the normalization will be wrong
- Embedded AI — AI features priced inside a non-AI product (CRM tiers, office suites, support desks). Frequently invisible in an "AI spend" search because the line item doesn't say AI
Normalization rules
- Amortize annual prepay to monthly (÷12) and SAY it's amortized — cash timing and run rate are different questions
- Convert credit purchases to run rate from actual burn (credits used per month × price per credit), not from the purchase amount
- Count reserved capacity at commitment, with a utilization percentage alongside — 40% utilization of a committed pool is the finding, not a lower cost
- Keep one currency and one period; label estimates as estimates every time
Contract mechanics worth flagging
- Auto-renewal windows and notice periods (the negotiation clock starts at notice-minus-30, not at renewal)
- True-up clauses on seat growth; whether unused seats true DOWN
- Credit rollover and expiry terms
- Rate-lock or price-protection language, and what happens to grandfathered pricing at renewal
Always distinguish run-rate math (FinOps working numbers) from accounting treatment (the controller's domain) — never present the former as the latter.