Deploying AI in your company: a four-step method

Deploying AI in a company is not just about buying licenses. The tools are there, often already used by your teams on a personal basis. What is missing is organization: leadership choices, prioritized use cases, shared rules and tools connected to your data.

The method we apply has four steps, run in a loop:

• Inspire leadership: the leadership team sees concrete cases from its own business lines and decides where AI will be useful, and where it won’t.
• Map operations: we identify the bottlenecks and rank use cases by gain and difficulty.
• Bring teams on board: in a hands-on session, each person comes with a real problem and leaves with the AI agent that solves it, within the framework of an AI policy.
• Build the tools: we connect AI to the ERP, the CRM and your documents, or rebuild the business tool that is holding things back.

Each step can be done on its own. Together, they turn individual use into team AI. Each step is detailed on our method page.

The surveys agree. According to McKinsey, nearly two-thirds of organizations have not yet begun scaling AI across the enterprise. According to BCG, 72% of companies cite integrating AI with their systems as a major challenge.

On the ground, the causes are almost always the same:

• No clear decision: no one has said which processes are priorities, or with what resources.
• An isolated pilot: the tool works for a test team, but it is not connected to any company software. Using it means copying and pasting.
• Use without a framework: everyone uses the tool of their choice, with data that leaves the company’s perimeter.
• Theoretical training: a general presentation, then nothing. Usage drops off within a few weeks.

We detail these blockers, and how to get past them, in our guide on moving AI pilots into production.

Inspire your leadership team

A talk built on cases from your business lines, not a general presentation of AI. The goal is a decision: which processes to explore first, which limits to set, who leads. Without this mandate, initiatives stay scattered.

Map your operations

Interviews with each business department bring out the tasks that consume time without adding value: re-entering data, searching for documents, follow-ups, summaries. Each use case is scored on two axes: the expected gain and the difficulty of implementation. The result is a short list of cases to launch, and the list of tools to adopt with the IT department.

Bring your teams on board

The hands-on sessions start from participants’ real problems. A buyer leaves with an agent that compares supplier offers; an executive assistant, with an agent that prepares meeting minutes. An AI policy sets out what is allowed, what is not, and who approves.

Build with the AI Factory

When AI needs to read and act in your software, it takes connectors, approved by the IT department. When the business tool is too old to be connected, our engineers rebuild it, AI-ready, in a few weeks. The code and the data stay with you.

Here is the sequence we recommend for a mid-sized or large company:

• Appoint a sponsor on the leadership team and a point person in each business department. The subject cuts across every function: it needs someone in charge.
• Take stock of current use: which tools, for which tasks, with which data. An anonymous questionnaire is enough to start.
• Choose the shared tools with the IT department: conversational assistant, data hosting, access rights. We work with Claude, Microsoft Copilot, ChatGPT, Dust or Mistral, depending on your environment.
• Write a short AI policy: authorized data, mandatory review, excluded uses. Our AI policy guide sets out its structure.
• Launch one or two use cases with high gain and low difficulty, with a volunteer team.
• Measure on an indicator you already track: processing time, number of cases handled, rework rate.
• Extend to other teams, then connect AI to business software once the need is confirmed.

This progression avoids two pitfalls: the big program that delivers nothing for months, and a string of tests that lead nowhere.

The best first cases have three things in common: a frequent task, information already available in your documents or software, and a person who validates the result. A few examples found in most mid-sized companies:

Conversely, avoid starting with a case that requires data you don’t have yet, or that would have AI decide in place of a person. More examples, sorted by function, are in our use cases.

Data. The main risk is not the tool, it is the information you give it. Choose tools whose contract rules out training models on your data, with hosting suited to your requirements, in France or on your premises if needed.

The legal framework. Since February 2, 2025, Article 4 of the AI Act has required companies that use AI systems to ensure a sufficient level of AI literacy among their staff. The GDPR applies as soon as personal data is processed. Involve your DPO from the mapping stage.

Human review. AI prepares: it sorts, checks, summarizes, writes a first draft. A named person reviews and decides. For anything sent outside the company or any decision that commits it, this rule must be written into the AI policy.

Access. An agent connected to the CRM or the ERP has only the rights the IT department grants it. Every action must be logged, so you know who did what, and based on which source.

Long-term adoption. One-off training is not enough. Plan for point people within teams and a regular review of the uses that work, so they can be shared.

We run the four steps with you, together or separately: a talk for the leadership team, mapping of operations, hands-on sessions by business line, then tool building by the AI Factory. We work with the tools chosen by your IT department, with no exclusive tie to any vendor. Your data is not used to train models, and you keep ownership of the code and the data.