AI in the enterprise: 5 questions you should be able to answer

Direction financière et DSI échangeant sur l’usage de l’IA

AI in the enterprise: 5 questions you should be able to answer

AI is already everywhere in your company. Employees have adopted it faster than IT, Procurement and Security could keep up. As a result, five simple questions often go unanswered. Take the test.

Adoption moving faster than the means to manage it

In France, for example, 47% of employees already use generative AI in their work, but only 23% have been trained to use it responsibly (Observatoire de l’IA responsable, Impact AI, KPMG and Bouygues, February 2026). Every team tests its own tools, every developer picks a subscription, and vendors are adding AI to the software already in place. The five questions below help you gauge where your company stands. The same rule applies to each one: an answer based on measured data is better than an estimate.

1. Usage: which AI tools are actually used, by whom and how often?

This is the starting question, and often the hardest. Inventories are usually based on purchased tools and on what teams report. Yet the gap with reality is significant: according to Productiv (2026, US companies), a company uses an average of 14 AI tools, while its IT department knows about only 4 or 5.

To answer it, you need to measure usage on employee computers, tool by tool and plan by plan, in the office and when working remotely, with indicators on chat usage that vendor consoles don’t show. Frequency matters as much as presence: a tool opened once a month doesn’t call for the same decision as a tool used every day. And the breakdown by team shows where AI is already creating value.

Warning sign: your inventory is based on a survey or on the list of contracts, with no measurement of actual usage.

2. Spending: how much do we spend on AI, across seats, APIs and expense reports?

AI spending no longer flows through a single channel: per-seat subscriptions, usage-based consumption, tools paid for through expense reports or corporate cards, AI built into existing software. Each channel often falls under a different budget, and no single line gives you the total. The trend is accelerating: in the United States, enterprise spending on generative AI more than tripled in one year (Menlo Ventures, 2025). We detail these four sources of spending in a dedicated article.

Warning sign: nobody can give you the annual amount of AI spending, expense reports included, within a day.

3. Licenses: which seats are paid for but unused, or duplicated?

The invoice tells you how many seats are paid for, never how many are used. Phased rollouts leave seats idle, some employees accumulate several equivalent tools, and licenses remain assigned to people who have left. Conversely, other users regularly hit their limits. Matching the usage of each seat with the HR directory lets you give everyone the plan that suits them, without jumping to conclusions: a rarely used seat may meet an occasional need.

Warning sign: your AI licenses are renewed based on the previous year’s seat count.

4. Data: what is leaving through unauthorized tools or personal accounts?

Text pasted into a consumer assistant, a contract to summarize, code submitted for correction: with shadow AI, company data leaves without a trace. In a PagerDuty survey from June 2026 conducted in Australia, Japan, the United Kingdom and the United States, two out of three office workers say they have already used AI tools they believed were unauthorized.

Answering it requires identifying unauthorized tools, personal accounts, the populations involved and events involving sensitive data, without exporting the content of conversations. With one more question about data control: where are the providers you use located, and therefore where is the data going? To go further: detecting shadow AI and bringing it under control.

Warning sign: your usage policy bans certain tools, but nobody can tell you whether they are still being used.

5. Models: are we using the right model for each task, in terms of cost and CO₂?

This is the question least often asked, yet it weighs on both the bill and the carbon footprint. Rewording an email, translating a paragraph or summarizing a document doesn’t require the most powerful model. Without guidance, choices are made out of habit, with higher cost, energy and water consumption for an equivalent result.

The answer depends on how the AI is accessed. For API flows, some companies use an LLM gateway that routes each request to the model best suited to the task. For per-seat subscriptions, the choice remains in the user’s hands: the challenge is to give them guidance on the cost and footprint of the different models. These are estimates expressed as ranges, based on frameworks such as those of ADEME (the French Agency for Ecological Transition) and Boavizta. They don’t have the precision of a measurement, but they are enough to compare models and shift practices, provided the quality of the result is also assessed.

Warning sign: you don’t know which models are used in the company, or in what proportions.

From questions to responsibilities

These questions don’t add up to a compliance score. They help organize the work: for each one, assign an owner, identify the information available and choose a first action.

TopicFunctions to involveFirst useful result
UsageIT and business unitsAn inventory with documented coverage
SpendingFinance and ProcurementA reconciled total over a defined scope
LicensesIT, Procurement and business managersA list of adjustments to check
DataSecurity, privacy officer and business unitsUsage rules and the situations to examine
ModelsUser teams, IT and FinanceSelection criteria and tests on real tasks

How many answers do you have?

If you answered all five questions with certainty, your company is among the most advanced. If not, you’re in good company: these answers require cross-checking data scattered across IT, Finance, Procurement and Security. The good news is that all five rely on the same approach: measure actual usage, then match it with spending. The steps of an assessment are detailed here.

How Saaswedo Can Help

With sobrAIvo, Saaswedo answers these five questions with measured data. Its AI Trace software, deployed in a few days through your unified endpoint management (UEM) tool, passively measures actual AI usage on employee computers, in the office and when working remotely, without the content of conversations ever leaving the computer. These measurements are cross-checked in the portal with your invoices, expense reports, network logs and other sources to deliver, in about six weeks, a costed action plan that also covers risky usage and the estimated carbon footprint.

1600 900 SAASWEDO | Telecom Expense Management & Managed Mobility Services