Resources
AI cost management (FinOps)
AI costs are scattered (per-user licences, pay-as-you-go API consumption, AI options charged as extras in existing software, projects), hard to forecast and rarely set against measured value. This pillar applies the four domains of the FinOps Framework — understand, quantify value, optimise, manage the practice — to an organisation that consumes AI without running its own platform.
General recommendations: the assessment selects those that match your situation and ranks them by priority.
AI spend visibility
The first step in FinOps: knowing how much AI costs, where, and who consumes it. In an organisation that consumes AI, spending hides in four places: per-user subscriptions, AI options in existing software, pay-as-you-go API consumption, and services. Individual subscriptions paid by card often escape all tracking.
What we recommend
- List all AI spending over the last 12 months using the budget template provided: extract invoices and subscriptions (accounting, expense claims, cards), and link each line to a use in the register and to a team.
- Create an “AI” cost code or category in the accounts and require every new AI subscription or contract to be charged to it; prohibit AI subscriptions on expense claims without a reference to the register.
- Tag each API key by application and export consumption monthly from the provider's console; reconcile it with the invoice and set up a tracking sheet per application.
Budgeting, forecasting and alerts
AI costs are less predictable than traditional IT costs: pay-as-you-go consumption, adoption that suddenly takes off, price or model changes by the provider. The FinOps Framework recommends short forecasting windows, budgets per use and overspend alerts rather than a fixed annual budget.
What we recommend
- Enable the consumption caps and alerts available from each provider (API budgets, per-key limits, notifications at 50 / 80 / 100 %); it is free and immediate.
- Set an AI budget per use in the dedicated tab of the template provided, with a rolling quarterly forecast rather than an annual one, and a monthly review of variances.
- Add a cost section to the AI use assessment grid (Security pillar): unit cost, expected volume, 12-month cost if rolled out widely, exit cost.
Value and unit economics
A cost only makes sense against what it produces. The FinOps Framework calls this unit economics: cost per resolved ticket, per processed case, per active user, per document produced. This is what lets you compare uses, stop those that deliver nothing and defend the budget of those that do.
What we recommend
- For each major use, define a value indicator and a unit cost (cost per active user or per unit produced) in the “value” tab of the template provided; start with the three most expensive uses.
- Extract the number of active users of each per-user AI tool from the admin consoles and cancel or reassign licences that have been inactive for more than 60 days.
- Set up a six-monthly review of the value of AI uses (cost, value indicator, adoption, risks) before the AI committee or lead, with an explicit decision: continue, expand, stop. Initial scoping helps choose indicators that stand up to scrutiny.
Rate and usage optimisation
Once spending is visible, it can be reduced: choose a model suited to the task (a smaller model is often ten times cheaper for an equivalent result), negotiate commitments, consolidate subscriptions, limit costly uses that add no value. The FinOps Framework distinguishes rate optimisation (paying less for the same thing) from usage optimisation (consuming less).
What we recommend
- Consolidate scattered AI subscriptions under an enterprise contract and renegotiate based on the actual volumes found in the inventory; include an annual review clause.
- Audit applications that use APIs to match the model to each task, enable caching and reduce context size; savings of 30 to 70 % are common on unoptimised uses.
- Review the AI options enabled in your business software (CRM, office suite, HR) and disable those that are not used or that duplicate a tool already in place.
AI procurement and spend governance
Who can commit AI spending, above what amount, with which approvals (security, legal, finance)? This topic connects finance with governance: the tool request process (Security pillar) must also be the budget process, otherwise one of the two will be bypassed.
What we recommend
- Merge the AI tool authorisation request and the purchase request into a single form (use, data, provider, unit cost and volume) with simple approval thresholds based on the amount.
- Appoint someone responsible for tracking AI costs (often management control or the AI lead) and add an “AI costs” item to the periodic governance review.
- Set up quarterly showback: each department receives the AI cost it generates, set against its value indicator; move to chargeback if the organisation's culture lends itself to it.
Frameworks
- FinOps Framework
This pillar builds on the FinOps Foundation's FinOps Framework (CC BY 4.0), adapted to an organisation consuming AI services.
Where does your organisation stand?
The assessment evaluates these points for your organisation and ranks the actions by priority. Free, about 15 minutes, no account needed.