How to track AI costs per team, a practical framework for SMEs

Learning how to track AI costs per team is one of the fastest ways to keep AI spend predictable at SME scale. Without team-level attribution, a single €7000 monthly AI invoice looks like a mystery until someone digs through logs to figure out which department drove which line item. This guide describes three practical approaches to track AI costs per team, what to capture at team level, where provider dashboards break down, and when a dedicated governance layer becomes worth the setup.

The audience is finance leads, IT owners, or founders at European SMEs who have moved past the stage of "one shared account for everyone" and want visibility into how each team consumes AI. In practice, this problem tends to appear around 20 to 30 employees, when informal awareness stops scaling and monthly invoices start surprising the CFO.

Three approaches to track AI costs per team

There are three practical approaches to track AI costs per team, in order of setup effort and reliability.

Approach 1: separate provider accounts per team. Each team signs up for its own team-level account with the AI provider. Sales has one Anthropic Team account, engineering has another, support has a third. The provider's admin dashboard shows spend per account. Simple to set up, but breaks down when teams share tools or when one employee sits on multiple teams.

Approach 2: tags or labels within one shared account. Some providers allow tagging requests with a team identifier passed as a header or metadata field. Reports in the admin dashboard filter by tag. Cleaner than multiple accounts, but requires that every code path passing prompts includes the tag correctly. In practice this is easy to enforce for API traffic and hard to enforce for interactive UI use.

Approach 3: governance layer with team-scoped identity. All AI traffic flows through one central layer that maps each request to a team based on user identity, not on developer discipline. The layer produces per-team reports across every provider in one view. Highest setup cost, but the only approach that captures both API and interactive UI spend without asking users or developers to remember to tag things.

Most SMEs move through the sequence: approach 1 during the first year, approach 2 as tooling grows, approach 3 once compliance or finance requires airtight attribution.

What to attribute at team level

Once you can track AI costs per team, the next question is what to record. Four dimensions matter for SME AI cost decisions.

Provider and model. Which provider ran the request and which model within that provider handled it. Different models within the same provider can differ 10x in unit cost. Knowing team X uses GPT-4 for tasks that team Y solves with GPT-4o-mini is actionable.

Volume and unit cost. Number of requests and average token cost per request. Volume growth is normal when teams adopt AI; unit cost growth without a matching value increase is a signal that the wrong model is being called for the wrong task.

Time distribution. Weekly and monthly patterns. Some teams have spiky usage tied to reporting cycles or client work. Knowing this makes forecasting realistic instead of based on a moving average that hides seasonality.

Compliance jurisdiction. For European SMEs, where the model ran matters as much as how much it cost. A cheap US-hosted model call for a workflow that legally requires EU jurisdiction is not a saving, it is a violation waiting to be discovered.

Track these four consistently and the monthly review conversation shifts from "where did the invoice go up" to "which team's growth is proportional to output".

Where provider dashboards break down

Native provider dashboards work well until three specific situations appear. Recognising them is worth more than picking any specific tool.

Multiple providers in use. Once you use OpenAI plus Anthropic plus Mistral, no single provider dashboard shows the total picture. Building a spreadsheet that stitches them monthly is possible but degrades quickly as the number of teams grows.

Bring-your-own-key deployments. Some teams use the provider's model through a third-party tool that carries the request but pays for the licence separately. Provider dashboards attribute cost to whichever entity holds the API key, not to whichever team actually issued the prompt.

Compliance evidence needed. Provider dashboards show cost, but rarely show the full audit trail needed for GDPR Article 30 records or EU AI Act Article 12 deployer logging. Reconstructing this after the fact is expensive; capturing it inline is not.

When any of these appear, the cost of a dedicated governance layer typically pays back in one to three months. For a decision framework on when to make the jump, see the AI cost management for SMEs guide.

The governance-layer approach in practice

The governance-layer approach to track AI costs per team works differently from provider dashboards. Instead of asking each provider "what did team X spend", the layer knows which team made every request because it authenticated the user before routing anywhere.

Concretely for an SME using Ciralgo, each request carries the user identity, which the layer resolves to a team through the identity provider integration. The report writes itself. No tagging discipline required from developers, no missed attributions from UI traffic, no reconciliation between provider dashboards at month-end.

The trade-off is setup cost. Standing up the governance layer takes a few hours plus some policy decisions. For an SME under 15 people, this is usually not yet worth it. For 25 people and up, it typically is, because the manual reconciliation work at that scale exceeds the setup cost within the first quarter.

For the full breakdown of Ciralgo pricing at team scale, see the Ciralgo pricing page. For the Dutch-language cost breakdown per company size, see Kosten AI implementatie MKB.

Frequently asked questions

How do you track AI costs per team when employees use both API and interactive tools?

The only reliable answer is a governance layer that authenticates every user regardless of channel. Tag-based approaches work for API traffic where developers control the code but rarely capture interactive UI traffic accurately. This is why once mixed usage becomes common, the layered approach tends to win over the tag-based one.

Can I track AI costs per team using just Excel and the provider dashboard?

For a small SME with one provider and clear team boundaries, yes. It works up to about 15 employees and one provider. Beyond that, the manual reconciliation at month-end starts costing more staff time than a dedicated tool would.

Do I need to track AI costs per team for the EU AI Act?

Not directly. The EU AI Act requires deployer records of AI systems in use and audit logs of AI interactions. Per-team cost attribution is not a legal requirement. However, the same infrastructure that produces per-team cost reports usually also produces the AI Act audit records, so most SMEs get both from the same layer.

How does per-team tracking work when one employee is on multiple teams?

The best implementations allow multiple team memberships and attribute costs based on the primary team unless a specific workflow flags otherwise. Attribution by primary team is not perfect but is stable enough for finance conversations. Perfect attribution across matrixed teams is often not worth the complexity for an SME.

What is a reasonable frequency to review per-team AI spend?

Monthly is the standard for financial reporting. For teams growing AI usage quickly, weekly informal reviews for the first month after enabling a new workflow are helpful. Anything more frequent tends to consume time without producing new decisions.

Further reading

• AI cost management for European SMEs: the 2026 practical guide
• Kosten AI implementatie MKB: concrete prijzen per bedrijfsgrootte
• AI governance voor MKB: 30-dagen setup framework
• Ciralgo pricing

Last reviewed 17 July 2026.