AI cost data can show which model was used, how many tokens were consumed, and how much the usage cost. But those details do not always answer the question a FinOps or Finance team needs to resolve:

Which team, product, or cost center should this spend be associated with?

The answer depends on the identifiers available in your AI usage data and how consistently those identifiers connect to your organization’s business structure. This guide walks through a practical allocation process using user tags in Umbrella.

Choose the business dimensions you need

Start with the decisions you want the cost data to support. Common dimensions include:

  • Team: Which teams are driving AI usage?
  • Product: Which products or workflows use AI?
  • Department: Where does usage sit in the organization?
  • Cost center: Which budget should the spend be associated with?

Choose dimensions that someone will actually use in a review, forecast, or budget conversation. Adding a long list of labels is less useful than maintaining a small set of reliable ones.

For each dimension, decide who owns the mapping and how often it should be updated. Team membership, product ownership, and cost-center assignments can change over time.

Match AI user identifiers to business context

AI cost feeds may identify usage by email address, API key name, or workspace member ID. Those identifiers are useful for tracing activity, but they may not explain which business group is responsible for it.

Create a mapping between the identifier in the cost data and the business dimensions you selected. In Umbrella, User Tags use a CSV with three columns: user_id, tag_key, and tag_value.

Example mappings:

Use the identifier exactly as it appears in the AI cost feed. Tag keys and values are case-sensitive, so Team and team are treated as different labels.

A user can have more than one value under a tag key. For example, one user may contribute to two teams. That can be useful when it reflects the real operating model, but it affects how totals should be interpreted.

Upload and maintain the mapping

Upload the completed CSV from the User Tags page. Once processed, Umbrella joins the tags to AI cost records using the user identifier. The resulting dimensions are available in AI cost views.

Plan updates carefully: uploading a new file replaces the existing mapping; it does not add rows to it. To make a change, download the current mapping, edit the complete file, and upload it again. Otherwise, mappings omitted from the new file will be removed.

This makes the CSV a useful, auditable source of business context, but it also means that someone should own the file and keep it current.

Analyze the results in AI Cost & Usage Explorer

In the AI Cost & Usage Explorer, choose a time range and a measurement, then group or filter the data by the dimensions you need.

For example, you can:

  • Group spend by Team to see how costs are distributed.
  • Add a secondary grouping such as Model to understand which models each team uses.
  • Filter by Product or CostCenter to focus on a specific area.
  • Compare cost with tokens, requests, or active users to understand changes in usage.

The Explorer’s chart and details table use the same grouping and filters, so you can move from a broad view into a more specific breakdown.

Watch for multi-value mappings

If a user is mapped to multiple values under the same tag key, that user’s cost appears under each value in a group-by view. The same spend can therefore appear in more than one group, and adding those group totals may exceed the actual total.

Use multi-value mappings when they reflect how work is shared. When you need totals that reconcile cleanly to the full bill, choose a single-value allocation rule or review the mapping before summing across groups.

Investigate gaps before treating the allocation as complete

A useful allocation depends on the quality of the mapping. Review records that remain unassigned or do not match the expected business dimensions. Common causes include:

  • A new user, API key, or workspace member has not been added to the mapping.
  • The identifier in the CSV differs from the one in the cost feed.
  • A team or cost-center value has changed.
  • The mapping has not been updated after an organization change.

Treat these as data-quality issues to resolve, rather than assuming that every remaining cost belongs to a generic “AI” category.

When user tags are not the right allocation method

User-based mapping is useful when the available usage data can be associated with users. Some AI workloads are better understood through other dimensions, such as application, model family, customer tier, or a custom workflow. Umbrella’s AI Business Mapping supports allocation across AI products and LLM workflows using rules, model attributes, and usage metrics.

The right method depends on the question you are trying to answer and the dimensions available in the underlying data.

A practical starting point

Choose one AI provider or workload. Identify the business dimension needed for the next cost review, map the available user identifiers, and compare the resulting allocation with the ungrouped total. Then check for missing users, stale mappings, and multi-value assignments before using the breakdown for budgeting or reporting.

Good allocation does not come from adding more labels. It comes from connecting cost data to business dimensions that are clearly defined, maintained, and useful for a real decision.