AI audit for companies: choosing the right platform

An AI audit for companies is how you choose the right platform before signing: Microsoft 365 Copilot, Dust, Claude (Anthropic), or a mix of the three. It combines leadership's needs, employees' actual usage and the state of your data. Without it, the choice rests on an existing contract or a demo, and adoption rarely follows.

What is an AI audit for companies?

An AI audit for companies is an assessment that lets you choose which AI platform to deploy, for whom and with which data. It starts from your usage and your information system, not from a vendor's catalogue.

The term also covers another topic: auditing AI systems already in place, to check their compliance and risks. That kind of audit relies for example on the ISO/IEC 42001 standard on AI management. This article covers the first case: the audit that comes before choosing the tool.

It is step 3 of our method, after awareness and governance.

What problem does an AI audit solve?

An AI audit solves a choice problem. The IT department has to commit the company to a platform for several years, without knowing which one will actually be used. Four situations come up:

• Personal licences in circulation. Employees use ChatGPT or Claude with their personal subscriptions, outside the IT department's control: this is shadow AI.
• Licences paid for but little used. A company-wide rollout, then low adoption after a few months.
• Data the tool cannot read. The ERP, the CRM or the file servers are not connected. The AI's answers then stay generic.
• Vendor pressure. Each supplier presents its platform as the right answer to every need.

The IT director then has to decide alone, with sales arguments rather than facts about the company.

Why audit before choosing your platform?

You should audit before choosing because an AI platform is only worth what it can read and who uses it. Both are specific to your company: no demo reveals them.

A useful audit is independent of the platform it recommends. At Idoption, the audit costs the same whether the recommendation is Claude, Copilot or Dust. We are certified on every platform mentioned here (Dust, Claude, Microsoft Copilot, OpenAI, Google Gemini, Mistral).

What does a good AI audit look at?

A good AI audit looks at three things. What teams need, what your systems allow you to connect, and what each scenario will really cost.

1. Leadership, the field and use cases. The needs of each executive, but also employees' actual usage. The two often diverge. These interviews surface the first use cases for each department, and those use cases are what separate the platforms.
2. The stack and the data. The stack is all of your software. The audit checks what each platform can read (office tools, ERP, CRM, files), who has access, and which data is sensitive.
3. Total cost. Licences, usage consumed, prerequisite licences, connectors to build and support. The right indicator is cost per active user: total annual cost divided by the number of employees who use the tool each month. A tool that is cheaper per seat can cost more per active user.

Copilot, Dust or Claude: which AI should a company choose?

To choose your company's AI, start from your user groups and your data, not from announced features. Microsoft 365 Copilot fits office assistance in a company fully on Microsoft 365. Dust fits business agents connected to several sources. Claude fits advanced users and technical teams. The outcome of an audit is often a mix of the three.

Here are the trends we see in our audits:

Other options exist depending on the context: Mistral Vibe (French vendor, formerly Le Chat) when sovereignty or hosting in your own infrastructure is a criterion, Gemini when the company works on Google Workspace.

An agent is an AI assistant connected to the company's tools and data, able to carry out a task. Agentic AI is not always the right answer: for many teams, an assistant connected to the right documents is enough.

What does an audit change in the decision?

An audit changes the decision on four points:

• One tool per user group rather than one tool for all. In our audits, the final recommendation is often a mix. For example: Copilot for office work, Dust for business agents, Claude for advanced users.
• A total-cost budget. Leadership compares costed scenarios, not prices per seat.
• Prerequisites known before signing. Data to clean up, access to review and missing connectors are listed before deployment, not discovered during it.
• A decision you can defend. In front of the executive committee as in front of the works council (CSE, the body that represents employees in French companies), the IT department relies on facts specific to the company.

A medical device manufacturer with 450 employees

At a medical device manufacturer with 450 employees, project scoping included a decision matrix between platforms. It was combined with a map of data and access rights.

This work surfaced two points a demo would not have shown. First, redundant, poorly structured files on file servers, which need structuring work before connecting agents. Second, an ERP with no direct connector, worked around with a data export to the platform.

Dust was chosen for three reasons: simple connections to the company's tools, a French vendor, and fast uptake by business teams. The deployment reached 90% adoption.

Takeaway: the platform choice depends as much on the state of your data as on the tool's features.

How do you prepare an AI audit?

To prepare an AI audit, bring together three things. A sponsor in leadership, the list of your software and data, and access to employees in each department. Who owns what:

Which mistakes should you avoid when choosing an AI platform?

The four most frequent mistakes come from choosing before looking at the company.

• Taking the tool already under contract. Choosing Copilot because the company has Microsoft 365 makes sense for office work, much less for agents that combine the ERP and the CRM. Check usage before extending a contract.
• Choosing on a vendor demo. A demo runs on prepared data. Test at least one real case, on your own documents.
• Not listening to the field. Interviewing only IT and leadership misses actual usage. Employees often already use another tool.
• Forgetting connectors. A tool that cannot read your data gives generic answers. Count your sources and check each connector before choosing a plan.

When is an AI audit not needed?

An audit is not needed when the platform is already chosen and adopted. In that case, the useful step is the use-case diagnostic, to find where AI pays off, then a pilot on your data.

It is also unnecessary for a small organisation with a single office use and a single user group. A few-week test of the tool you have in mind is enough.

And after the audit?

After the audit come the use-case diagnostic, a first pilot on your data, then deployment. These steps are described in our method and illustrated by our client cases. If leadership is not yet aligned on the topic, an executive AI training can come before the audit. Also think about the framework: the AI policy and, in a company of 50 employees or more, consulting the CSE (Works council (CSE) and AI).

To find out which platform suits your company, contact us.

Sources: ISO/IEC 42001:2023, AI management system · Microsoft Learn, Microsoft 365 Copilot overview · Dust, security and hosting page · Anthropic, Claude for Microsoft 365 and Claude Code · Mistral, Mistral Vibe · Google Workspace, AI tools (accessed 2 November 2026).