How Much Is Your Company Really Spending on AI?

There’s one question that makes more than a few executive committees uncomfortable: how much are we spending on AI every month? Almost nobody has the answer at hand. Marketing pays for ChatGPT. Development teams use GitHub Copilot or Claude Code. Microsoft 365 Copilot licenses are scattered across different departments. And API credits are being consumed by projects that few people actively monitor.

Each decision made sense on its own. The problem appears when you add everything together: AI spending keeps growing, it’s spread across the organization, and nobody is looking at the full picture. Even less are they connecting that spending to the value being generated.

Where the AI Bill Really Comes From

When you look closely at AI spending, money tends to leak through four areas. And none of them are simply “the tool is expensive.”

• Licenses for everyone, with low adoption. This is the biggest drain. Microsoft 365 Copilot costs approximately €28 per user per month. For a company with 500 employees, that represents €168,000 per year. And with the underutilization rates reported by Zylo (46% of paid applications see little or no usage), nearly €77,000 annually could be spent on seats that are rarely used.
• Overlapping tools. ChatGPT, Copilot, and Claude being used for the same day-to-day tasks. GitHub Copilot, Claude Code, and Cursor doing similar things in software development. When tools are purchased department by department, without central coordination, companies often end up paying twice for the same capabilities.
• The wrong pricing model. Paying for a per-user license when the actual use case would be better served through API consumption. We’ll come back to this point, because it’s where some of the largest savings opportunities exist.
• Unmonitored consumption. APIs and AI agents scale with usage. No one pays attention to the bill until it arrives much higher than expected.

User Licenses vs. API Consumption: The Decision That Has the Biggest Impact on Costs

A license is paid per person, whether that person actually uses AI during the month or not. An API is paid based on consumption: the amount of data sent to and generated by the model. For open-ended, everyday tasks (writing, summarizing, searching through emails and documents) a user license often makes sense.

For high-volume, repetitive, well-defined tasks, it usually doesn’t. Consider a practical example. Classifying or extracting data from 50,000 documents per month using an efficient model such as Claude Haiku (priced at roughly $1 per million input tokens and $5 per million output tokens) costs around $100 per month, and potentially half that when using batch processing. The exact total depends on document size, but the order of magnitude remains the same. Trying to handle that same volume manually, with a team supported by AI tools, would cost significantly more and would still depend on someone performing the task every day.

This doesn’t mean APIs are always the better choice. It means that every use case requires a different approach, and treating them all the same is what drives costs up.

1. Which model fits each use case? To make the right decision, evaluate five factors. How many people use it, and how often? A large number of users working with AI daily can justify licenses. A small number of users, or occasional usage, rarely does.
2. Is the task open-ended or well-defined? Creative, variable tasks often benefit from tools such as Copilot. Structured, repetitive tasks are usually better suited to API-based automation.
3. What is the volume? The greater the repetitive volume, the more attractive API consumption becomes compared to per-user licensing.
4. Does it need integration? If the process must connect to your business systems and data sources, a custom solution built around your environment is often the best option.
5. How sensitive is the data? The more sensitive the information, the more important it becomes to keep data within your own perimeter (your Microsoft tenant or your own API keys) instead of relying on consumer-grade tools. And sometimes the best answer is to add nothing at all. You may already be paying for something that does exactly what you need.

How to Regain Control of AI Spending

You don’t need another tool. You need visibility.

• Create a single inventory. Document every AI tool the company pays for, who uses it, and what it's used for. Most organizations don’t have this inventory today.
• Measure cost per active user, not cost per license. The Microsoft 365 admin center shows enabled Copilot users versus active users, along with the ratio between them. If you're paying for 200 licenses but only 80 people actually use them, you immediately have 120 licenses that can be reassigned or cancelled.
• Track API consumption by team. Platforms such as Anthropic and OpenAI allow costs to be segmented by API key or project. This makes it possible to understand which teams are consuming what amount of budget and establish spending limits where necessary.
• Review before renewal. Most subscriptions renew automatically. A simple review before each renewal cycle often frees up budget without requiring any additional changes.

Conclusion

Your company is probably not overspending on AI because the tools are too expensive. It is overspending because that spending is distributed, unmeasured, and managed in isolation. Measuring actual usage, eliminating overlaps, and choosing between licenses and APIs according to each use case often unlocks significant savings before adding anything new.

At Itequia, as a Microsoft 365 partner, we help you get to grips with your AI expenditure: measuring actual usage, identifying overlaps and deciding, on a case-by-case basis, when a licence is the best option and when a bespoke API-based solution is more suitable. Email us at [email protected] and we’ll carry out a quick review with a detailed assessment. You can see how we apply AI to bespoke software here.

You can also learn more about how we apply AI to custom software development.

Frequently Asked Questions

How much does Microsoft 365 Copilot cost per user per month?

The Enterprise version costs approximately €28 per user per month (the euro equivalent of Microsoft's $30 price), in addition to a qualifying Microsoft 365 license such as E3, E5, or Business. For organizations with up to 300 users, a Business version is available at approximately €18.20 per month under promotional pricing. Because Copilot is an add-on to an existing Microsoft 365 subscription, the actual total cost per user is typically much higher.

If my team already pays for ChatGPT or Claude, do we also need Copilot?

Not necessarily. These tools overlap significantly. Copilot makes sense when you want AI to work directly with your Microsoft 365 data (emails, documents, Teams conversations...) while respecting existing permissions. ChatGPT and Claude are often sufficient for more general-purpose use cases. Providing multiple AI assistants to every employee frequently results in paying twice for similar capabilities. The right choice depends on the use case, not on which tool is considered “better.”

How can I tell how many of my Copilot licenses are actually being used?

The Microsoft 365 admin center includes a Copilot usage report showing: Enabled users, active users and adoption rate between the two. The report can be viewed over periods ranging from 7 to 180 days. This is the best starting point for identifying licenses that can be reassigned or cancelled and for calculating cost per active user rather than simply cost per license.