FinOps for AI assessment checklist

1. Establish what the data covers

• Collect provider billing exports, service-level usage, invoices, credits, discounts, and any relevant internal cost records.
• Record provider, account, region, service, currency, reporting period, export date, and latest observed billing date.
• Confirm whether the source is complete for the period and whether data is final, delayed, estimated, or still changing.
• Separate measured billing and usage from modeled scenarios; document the assumptions behind any estimates.

2. Connect spend to accountable teams and work

• Identify the department, team, application, project, environment, and service associated with each cost where the source supports it.
• Document allocation rules for shared services, pooled infrastructure, or costs that cannot yet be assigned.
• Track missing or inconsistent tags and assign an owner and correction path.
• Make clear whether allocations are direct, shared, estimated, or unallocated.

3. Define useful unit economics

Choose a denominator that reflects the value the workload is meant to deliver. Examples include cost per successful request, completed document, inference, active user, or business transaction. Define the metric before comparing systems or teams.

• Record the unit, counting method, success criteria, and data source.
• Compare costs over the same scope and reporting period as the output metric.
• Include relevant infrastructure, software, data, and operational costs where they are available.
• Show both total spend and unit cost so efficiency gains do not obscure growth in overall demand.

4. Review budgets, forecasts, and optimization options

• Compare actual spend with budget and forecast; identify variance, drivers, and accountable owners.
• Model workload, rate, utilization, and capacity scenarios with visible assumptions and time horizons.
• Review committed-use discounts, reserved capacity, credits, contract terms, and upfront fees against expected usage.
• Separate alternatives clearly: for example, pay-per-use API and fixed GPU capacity may be alternative scenarios, not combined charges.
• Show which changes affect usage cost, which affect a fixed commitment, and which require operational or quality review.

5. Turn analysis into a decision

• Identify who can act on each finding: engineering, FinOps, finance, product, or procurement.
• State the decision, quantified cost or unit-metric impact, assumptions, tradeoffs, and evidence behind it.
• Assign an owner and review date; monitor whether the expected result appears in later billing and usage data.

AICS FinOps reviews connect cost evidence with allocation, unit economics, forecasting, and optimization decisions. The assessment should state which data was measured and which outcomes remain modeled.