FinOps guides
Margin playbooks for AI SaaS
15 working guides for finance and engineering teams who carry LLM tokens in cost of revenue. Each one shows its arithmetic and uses list prices from the LLM Price Index.
- 1SaaS gross margin math when LLM tokens are part of COGSPut tokens in cost of revenue correctly4m
- 2Designing per-tenant LLM cost attribution that finance will trustAttribute every token to a paying customer3m
- 3Hybrid pricingPrice so heavy users pay for heavy use4m
- 4Billing LLM usage with Stripe meter eventsBill the overage automatically3m
- 5Unit economics dashboards for AI SaaS finance teamsReport it every month3m
All guides
15 of 15 guides
- Normalizing token usage across OpenAI, Anthropic, Gemini and BedrockProviders report input, cached, output and reasoning tokens in different shapes and count them with different tokenizers. How to normalize usage so cost and limits compare fairly.Rohan MalhotraAttribution3 min
- Credit wallets and prepaid drawdown for AI productsDesign a prepaid credit wallet for LLM usage: credit value, per-action pricing, real-time drawdown, low-balance behavior, expiry and the accounting questions to settle early.Vineet SaresarPricing3 min
- Cost anomaly detection for LLM spend: building per-tenant baselinesHow to build hour-of-week baselines per tenant, score deviations with robust statistics, handle new tenants, and route alerts, alongside in-path circuit breakers.Kabir SethiReportingHow-to3 min
- Unit economics dashboards for AI SaaS finance teamsThe metrics, definitions and layout for a dashboard that shows LLM cost per tenant, per task and per dollar of revenue, and how to keep it consistent with the books.Ananya IyerReporting3 min
- Forecasting LLM spend from usage drivers, not last month's invoiceBuild a driver-based forecast of model cost: tenants, sessions, calls, tokens and price, with growth, model changes, caching and an incident reserve.Ananya IyerForecastingHow-to3 min
- Fail-open vs fail-closed for LLM gateways: a decision frameworkFail-open keeps your product up when the gateway is unreachable; fail-closed keeps every call metered and capped. How to choose per feature with expected-cost arithmetic.Rohan MalhotraReliability4 min
- Budget envelopes for multi-agent systems: giving each sub-agent a spending limitHow to allocate a dollar envelope per agent run, split it across planner and worker sub-agents, keep a reserve, and enforce it in the gateway as well as in the orchestrator.Meera KrishnamurthyAgentsHow-to3 min
- Controlling reasoning token cost on o3, o4-mini, Claude and GeminiReasoning tokens are billed as output but never shown. How to measure them per tenant, cap them with effort and budget settings, and route only the requests that need them.Meera KrishnamurthyAgents3 min
- Modeling LLM usage in Metronome and Orb: events, billable metrics and commitsHow to shape usage events for Metronome and Orb, define billable metrics over token classes and models, and sell prepaid commitments against them.Ananya IyerBilling4 min
- Billing LLM usage with Stripe meter events: a build guideCreate a Stripe meter, choose the unit, attach a usage price, send idempotent meter events per tenant, handle allowances and reconcile, with PALM SOUQ billing sync.Nisha PillaiBillingHow-to3 min
- Model downgrade policies: when to fall back to a cheaper model and how to roll it outA procedure for choosing fallback model pairs, setting budget and reliability triggers, scoping by feature, telling users, and measuring the quality impact.Nisha PillaiRoutingHow-to3 min
- Prompt caching economics: write premiums, read discounts and break-even reuseHow Anthropic's 1.25x cache write and 0.1x cache read, OpenAI's automatic caching and Gemini context caching change cost, with break-even math and prompt layout rules.Ananya IyerCaching4 min
- Designing per-tenant LLM cost attribution that finance will trustIdentifiers, headers, pricing at the serving model, cache and reasoning tokens, internal traffic and reconciliation: a build guide for per-tenant attribution.Rohan MalhotraAttributionHow-to3 min
- Hybrid pricing: base fee, token allowance and overage, designed from your cost dataA step-by-step method to set the base price, size an included usage allowance, and price overage so every tenant clears your margin target.Vineet SaresarPricing4 min
- SaaS gross margin math when LLM tokens are part of COGSHow to put model tokens, embeddings and gateway costs into cost of revenue, compute gross margin per tenant, and explain the change to your board with arithmetic that adds up.Ananya IyerMargin4 min
Step-by-step
Procedures you can hand to a team
These guides are written as ordered procedures, with each step stated up front so an engineer and a finance lead can split the work.
Reporting6 stepsCost anomaly detection for LLM spend: building per-tenant baselines
- 1.Choose the grain
- 2.Build hour-of-week baselines
- 3.Score each hour
- and 3 more
- 1.Choose the drivers
- 2.Measure each driver
- 3.Price per call from token classes
- and 3 more
- 1.Set a run envelope
- 2.Split it across roles
- 3.Propagate the run identity
- and 3 more
- 1.Choose the billing unit
- 2.Create a meter
- 3.Create a usage price on the meter
- and 3 more
- 1.Classify features by quality sensitivity
- 2.Pick fallback pairs
- 3.Evaluate the pair offline
- and 3 more
- 1.Pick the tenant identifier
- 2.Add tier, feature and user dimensions
- 3.Record every token class
- and 3 more
New guides in The Token Governor
One weekly email with new guides, price index changes and a worked margin example.
Turn the guides into controls.
PALM SOUQ meters every token per tenant, enforces budgets and fallbacks in the request path, and syncs usage to your billing system.
14-day free production trial · 100k requests included · no card required