How to become an AI-first company: a practical path

"AI-first" has become a label everyone uses, from Silicon Valley start-ups to press releases about a chatbot. For a company with 50 to 1,000 employees, the useful question is simpler: what has to change, concretely, for AI to become part of how the company runs, and in what order?

The context helps. In 2025, 20.0% of EU enterprises with 10 or more employees used AI; Romania was last, at 5.2% (Eurostat, reference year 2025, published December 2025). Among medium-sized firms (50–249 employees) the share was 30.4% across the EU and 7.8% in Romania (Eurostat, KS-01-26-009). Most mid-sized companies, in other words, have not started either. Becoming AI-first in the next few years mostly means getting ahead of competitors who have not moved yet.

What does an AI-first company actually mean?

A company is AI-first when, for any new or reorganised process, the first question is "which part of this can software with AI do, and which part needs a person?". That takes three things: the data the processes need can be reached, people know how to work with AI systems, and every system has an owner who keeps it running.

What it does not mean: ChatGPT licences for everyone, a chatbot on the website, a model of your own, or a redundancy plan. Licences without changed work show up in the numbers quickly. IBM's 2026 study of 2,000 CEOs found that only 25% of the workforce uses AI regularly as part of their job (IBM, May 2026).

What are the stages of becoming AI-first?

Four stages, in this order. Each has a clear result and a signal that you can move to the next one.

Stage 1: foundations

Someone in the leadership team owns AI. There are written rules about which tools and which data staff may use. You have a list of the main processes and where their data lives. People get basic training, because since February 2025 the EU AI Act has required organisations using AI systems to take measures to support the AI literacy of their staff (Regulation (EU) 2024/1689, Art. 4). The bar is effort, not certification, but having done nothing is not a defence.

The signal to move on: you can name the first use case, with reasons, and the person in the business who wants it.

Stage 2: one use case in production

A single process, chosen because it repeats often, has reachable data and an owner who wants the change. You take it all the way: pilot, integration into the real systems, measured adoption. How to choose it is in how to choose your first AI use case; why most stall on the way to production is in from AI pilot to production.

The signal: the system runs every day without the project team, and the adoption number is measured.

Stage 3: three to five use cases on shared foundations

The second and third use cases should cost less than the first, because they reuse what was built: data access, integrations with the ERP and CRM, the way you evaluate systems, cost monitoring. This is usually where you need one internal technical person who owns all the AI systems, not just one.

The signal: each new use case starts faster than the one before, and people in the business come forward with proposals on their own.

Stage 4: processes redesigned around AI

Only now does the work itself change. Instead of adding AI to an old process, you redesign the process knowing what the system can do. BCG puts 70% of AI value in people and processes, 20% in data and technology, and only 10% in algorithms (BCG, The Widening AI Value Gap, September 2025). This stage never finishes. It is simply how an AI-first company changes its processes from then on.

What changes in roles and processes?

Roles change less dramatically than the headlines suggest, but they do change:

• Process owners carry an AI measure in their objectives: how much the system is used and what changed in the process.
• Every AI system has a technical owner, internal or external, who keeps it running and decides what changes.
• A review role appears: people who check AI output on high-stakes cases and correct the system.
• IT takes on integration, access rights and evaluation, not only infrastructure.
• Training becomes role-specific: the accountant learns something different from the sales rep.

The last point is the one most often skipped. In 2024, 73% of large EU enterprises provided training to develop their staff's ICT skills, against 21% of SMEs (Eurostat, isoc_ske_ittn2). How to do it in practice is in how to train your team to use AI.

Processes gain a few new parts too: a clear path for exceptions, a person in the loop where mistakes are costly, documentation of what the system does, and, for high-risk systems, the deployer duties of the AI Act.

What should you not do first?

• Do not buy licences for everyone and call it a strategy. A tool without a changed process produces an invoice, not a result.
• Do not start with a company-wide data platform. Build data access for the first use case, then extend it.
• Do not train a model of your own. Almost every use case in a mid-sized company works with existing models, through an API, over your data. See custom model vs API.
• Do not hire a Head of AI with no use case and no team. We cover this in should I hire an AI engineer.
• Do not start with systems that make decisions about people, such as hiring or credit. The AI Act treats them as high-risk, and your first use case should build trust, not a compliance file.
• Do not announce the transformation before the first result. People believe a system that shortens their work, not a memo.

How long does it take to become AI-first?

The first use case in production is measured in months, not years, if the data can be reached and the use case is well chosen. Stage 3 usually takes a budget cycle or two. Stage 4 is permanent. Companies that get there do not "finish" the transformation; they change how they make decisions about processes.

Where should a mid-sized company start?

With three things this week: name the person in the leadership team who owns AI, write down on one page the ten processes that take the most hours, and pick one of them for a first serious conversation. Vlad Tudor has a practical list of first changes in what to change in your business to use AI, and if you are not sure where you stand, read how to tell if your business is AI-ready.

At Sapio the first step is the Tech Call: a free 30-minute conversation in which you describe one process and get an honest answer on whether AI belongs in it. For a picture of the whole company, the next step is the Tech Audit, our starting package, on your site. For stages 2 and 3 we work with an embedded AI engineer in your team, who takes one use case at a time into production alongside your people. Request a Tech Call.

Sources

• Eurostat, "20% of EU enterprises use AI technologies", 11 December 2025 (reference year 2025, enterprises with 10+ employees, financial sector excluded)
• Eurostat, statistical report KS-01-26-009 on AI use in enterprises, 2026 edition, published 26 March 2026
• IBM, 2026 CEO study, 4 May 2026
• BCG, The Widening AI Value Gap, September 2025 (BCG sells AI consulting services)
• Eurostat, enterprises providing ICT training, isoc_ske_ittn2, reference year 2024
• Regulation (EU) 2024/1689, the AI Act, Art. 4, as amended by Regulation (EU) 2026/1744