The price of a subscription is easy to find. A company’s AI spending is much harder to pin down: central contracts, individual purchases, API consumption and built-in features don’t go through the same budgets. Before trying to reduce it, you need to reconstruct it.
And the issue is growing. In the United States, enterprise spending on generative AI more than tripled in one year, from $11.5 billion to $37 billion (Menlo Ventures, 2025).
Define what you want to measure
The budget for an AI project may include consulting, development, infrastructure and training. Usage spending, on the other hand, covers subscriptions, API calls and add-ons consumed day to day. The two answer different questions: to prepare a license renewal, there’s no need to reconstruct development costs. So start by setting a period, the entities involved and the categories included.
Four sources of spending to reconcile
1. Per-user subscriptions
This is the visible part, billed monthly or annually. It must be matched against assigned accounts, activity and needs: accounts of departed employees, rarely used licenses, multiple subscriptions. Low activity isn’t enough to draw a conclusion, since an occasional need can justify a seat. Conversely, some profiles constantly hit the limits of their plan. Optimizing also means giving them more capacity.
2. API consumption and credits
Applications, agents and development tools are often billed by usage. Spending depends on the models, the volumes and how calls are chained, with different rates for input, output and cache tokens. To explain an increase, match the invoice to the activity: more requests, longer contexts, a change of model, repeated processing. Rising consumption may also go hand in hand with highly valuable usage.
3. Corporate card purchases and expense reports
A subscription reimbursed to an employee doesn’t necessarily appear in Procurement’s portfolio, and several teams may sign up separately for the same service. This is far from marginal: in the United States, SaaS spending submitted through expense reports grew 267% in one year, and ChatGPT became the most frequently expensed application (Zylo, 2026 SaaS Management Index, large enterprises). Watch out for double counting: the same payment can appear on a card statement, an invoice and an expense report.
4. AI features built into software
An AI feature may be included in a suite, sold as an add-on or consumed as credits. If it has its own price, include it. Otherwise, allocating part of the bundle to AI requires a documented rule: the total price of a suite doesn’t become AI spending. Free tools count in the usage inventory, without weighing directly on the bill.
A consolidation example
A fictitious company reconciles its amounts over one month:
| Item | Illustrative amount |
|---|---|
| Central subscriptions | $6,000 |
| API consumption | $1,500 |
| Reimbursed subscriptions, after deduplication | $600 |
| Total identified spending | $8,100 |
Annualized, that comes to $97,200 at constant activity: a projection, not a forecast, since volumes and prices change. AI features included in other software are tracked separately, as they have no separate price.
Three metrics not to confuse
- Actual spending. The amounts billed or committed. This is what the budget and savings tracking rely on.
- API-equivalent cost. Consumption valued at API rates. For a subscription, it is neither an additional cost nor a recoverable sum.
- Business value. The work done and its quality. A large volume of tokens doesn’t, on its own, prove a productivity gain.
Why a cap doesn’t replace management
A cap contains variable consumption. It doesn’t detect subscriptions bought elsewhere, accounts of departed employees or commitments that renew without review, and it can interrupt legitimate usage. Over time, you need budgets with alerts, a watch on supplier offerings and, for each action, an expected gain, a deadline and its conditions. A saving expected at renewal shouldn’t be presented as a done deal.
Where to start?
Choose a manageable scope: one entity, one contract portfolio or one team. Gather the financial sources, reconcile licenses and review the main gaps with the people responsible. This first baseline is enough to make initial decisions and spot missing information. To structure the approach: the steps of an AI usage assessment.
How Saaswedo Can Help
With sobrAIvo, Saaswedo reconstructs your AI spending by matching your invoices, expense reports and vendor consoles with the actual usage measured on employee computers by its AI Trace agent, in its portal. In about six weeks, you have a costed action plan, followed by ongoing monitoring with budgets, alerts and a pricing watch.



