A smarter AI budget starts with actual usage.

An AI budget can look tidy in a spreadsheet and still be difficult to explain. A company may pay for several general assistants, coding tools, and specialized applications. Some are purchased centrally. Others belong to a department. Their renewal dates rarely line up.

The first useful question is simple: what are people actually using?

Here is an operational approach to reviewing that question. It is a framework for a conversation, rather than a formula for cutting every quiet license.

Connect the purchase to the person and the tool

Start with approved products, purchased seats, contract owners, and renewal dates. Then connect that inventory to observed activity over a period that represents the team’s normal work.

Keep your definitions explicit. A purchased seat, a provisioned seat, and an active user are different things. An employee opening an application once is also different from someone using it regularly.

For each tool, build a view of:

• The team responsible for the purchase and the intended workflow.
• The number of paid seats and the people assigned to them.
• The people using the tool during the review period.
• Other AI tools those people use for similar work.
• The next point when the company can change the commitment.

This creates a starting point for review. It does not decide the outcome on its own.

Treat overlap as a question

Two assistants may appear interchangeable in a product category while serving different purposes for a team. A developer might prefer one for code completion and another for working through a difficult architectural question.

Ask what the second subscription contributes. Is it tied to a distinct workflow? Does the team need both? Would a smaller allocation preserve the useful capability?

The aim is to distinguish intentional overlap from accidental duplication. The answer should come from usage context and a conversation with the team, rather than the similarity of two product descriptions.

Give quiet licenses some context

An inactive seat is a review opportunity. Before reclaiming it, check for recent onboarding, leave, seasonal work, or a workflow with a longer cycle.

A useful process gives the license owner a chance to explain the need. Where a tool is no longer needed, recover the seat or adjust the next renewal. Where adoption has stalled, decide whether enablement is more useful than removal.

Keep the decision visible so that the same unused allocation does not quietly reappear next quarter.

Keep cost estimates and invoices distinct

Usage information and contract information answer different questions. Observed activity can show whether a tool is used. The contract and invoice establish what the organization actually pays.

For a meaningful spend review, record the assumptions behind any estimate: the seat count, pricing basis, contract term, and period covered. Reconcile those assumptions with the owner of the contract before making a financial commitment.

This makes a potential opportunity easier to assess and avoids treating an estimated saving as a realized one.

Make renewals a recurring operational decision

The best time to understand a tool is before the renewal deadline. Bring its adoption trend, overlapping products, and outstanding license reviews to the owner early enough to act.

A useful renewal conversation ends with a clear choice: retain, expand, consolidate, reduce, or investigate. Give that choice an owner and revisit whether it had the intended effect.

Themisto’s approach is to connect AI usage with those decisions. A clearer view of the stack gives finance, IT, and the teams doing the work a shared place to start.