How to Track AI Spend Across Teams (A Simple FinOps Approach)
AI costs grow quietly. Different teams use different models, keys and tools, and the bill arrives as one number. A simple tracking habit fixes most of this.
What to record
For every usage entry, capture:
- Date
- Team and project
- Model used
- Input and output tokens
- Cost, calculated from current prices
Most providers expose token usage in their dashboard or in each API response.
Tag by owner
Use separate API keys or project labels per team. Without ownership, you cannot tell who should optimize what. Shadow usage, where people sign up with personal cards, is the usual blind spot.
Review monthly, inform before enforcing
Share a monthly summary by model and team before introducing limits. People change habits quickly once they see their numbers, and you avoid blocking useful work.
Set a budget and watch the ratio
Track spend against a monthly budget and look at the share per model. If one expensive model takes most of the budget, test a cheaper one for part of the traffic.
Keep prices current
Model prices change often. Keep one price table and let costs calculate from it, so updating a rate fixes every row.
A four-week plan to get visibility
Week 1: find the spend. List every place AI is paid for: API accounts, chat subscriptions, coding assistants and embedded features in other software. Collect the last two or three invoices. Ask teams what they use, since some tools will not show up in central billing.
Week 2: assign owners. Create separate keys or projects per team and name an owner for each. Every dollar needs a person who can explain it.
Week 3: start logging. Record usage in one place, with team, project, model and tokens. A spreadsheet is enough at first.
Week 4: review. Share a one-page summary and agree on two or three actions, such as testing a cheaper model for the biggest feature.
Metrics worth tracking
- Spend by team and by project: shows ownership
- Spend by model: shows where a cheaper option might work
- Cost per request or per task: shows efficiency, not just totals
- Budget used: the share of the monthly budget consumed so far
- Cost per successful outcome: the best measure, since it includes retries and failures
A total that rises is not necessarily bad. If cost per task falls while usage grows, the program is healthy.
Dealing with shadow AI
People adopt AI tools on their own because they are useful. Rather than blocking everything, find out what is in use, offer approved alternatives and make the approved path easy. Look at expense reports and browser extensions lists, and ask teams directly. Blocking without an alternative usually just hides the usage.
Inform first, enforce later
Start by showing teams their numbers. Visibility alone often changes behavior. Introduce budgets and alerts after that, and reserve hard limits for cases where overspending is clearly risky. This keeps trust and avoids breaking useful work.
What a good monthly review looks like
Spend 30 minutes. Look at total versus budget, the top three cost drivers, any big changes from last month, and any price changes from providers. Decide one action and one owner. Repeat next month.
Frequently asked questions
Do I need a special tool?
Not at first. A well-structured spreadsheet with a price table and a usage log covers a lot of ground. Tools become useful when you have many teams or need real-time alerts.
How do I get token counts?
From the usage field in API responses or your provider's dashboard.
What about subscriptions?
Track them separately as a fixed monthly cost per seat, and compare against equivalent API usage. For the structure, see how to calculate LLM API cost.
A ready-made tracker
The AI spend tracker spreadsheet has a price table for 34 models, a usage log with automatic cost, and a summary by model and team with a budget check. Get it from the template page. Want to estimate costs before you spend? Try the calculator.
Track spend by team and project: the AI Spend Tracker spreadsheet is launching soon.