You don't need more AI: you need to redesign a department

A year ago, the owner of a small manufacturer in the provinces — thirty people — did exactly what he'd been told to do: licenses for an AI assistant for everyone, from purchasing to production. Today, if you ask him, he'll admit something uncomfortable. Everyone has access. Some use it to write emails faster. But on the P&L — margins, delivery times, hours recovered — nothing measurable has changed. The question he carries around is: did I pick the wrong tool? Almost always the answer is no. He bought the access and stopped there.

What's new: the bottleneck is no longer the model

This is the good news, and it's recent. Until not long ago, AI's limit was the quality of the model: it hallucinated, it didn't understand context, it wasn't ready for serious work. That season is over. Today's models handle the real processes of an SME, and the constraint has shifted sharply: it's no longer what AI can do, it's how the work around it is organized. Giving everyone access is the easy part — the 90% that anyone does by now. The real unlock is the 10% almost nobody does: redesigning the process. And this isn't bad news, it's the best possible news — because the lever that matters is finally under your control, not inside a lab in San Francisco.

What the data says (and it isn't ours)

The gap between access and redesign is documented by three independent sources, and it's worth reading them as third-party evidence — not as our results, but as a snapshot of the problem we solve.

• Access is now everywhere, redesign almost nowhere. According to Deloitte, “State of AI in the Enterprise” (2026), in a single year the share of workers with access to authorized AI tools went from less than 40% to about 90%. But only 30% of organizations are redesigning key processes around AI. There's the hole: almost everyone has the tool, very few have changed the work.
• And it shows on the P&L. MIT NANDA, in the study “The GenAI Divide” (2025), finds that 95% of GenAI projects produce no measurable impact on the P&L. The mechanism MIT points to isn't the quality of the model — it explicitly rules that out — but a “learning gap”: tools that neither retain feedback nor adapt to context, and companies that don't integrate them into their processes, structures, and culture.
• Those who redesign, on the other hand, gain. McKinsey (QuantumBlack), “The State of AI in 2025” shows that AI “high performers” are about 2.8-3 times more likely to have deeply redesigned their workflows instead of laying AI on top of old processes (about 55% versus about 20%). Meanwhile nearly 9 in 10 companies use AI, but fewer than 4 in 10 see an impact on results.

Three different sources, one single line: access doesn't move the numbers, redesign does. The difference between those who win and those who stay stuck isn't having bought the right AI — it's having changed the way they work.

Why redesign is genuinely within reach for an SME today

Here comes the advantage a large enterprise doesn't have. “Redesigning processes” in a giant means a multi-million-euro project, committees, months. In an SME the surface to change is small: a single department, a single workflow. You don't have to rethink the company — you have to rethink how the order desk handles confirmations, or how administration records invoices. It's a perimeter that one person with a name can embrace in a few weeks, not the ocean that paralyzes multinationals. Your size, which is usually the disadvantage, here becomes the weapon: the redesign McKinsey associates with high performers is, for you, a single-department project, not a company-wide transformation.

What to do, in practice (and it isn't buying more licenses)

If your year went like the owner's at the start, the right move isn't to add tools: it's to remove and redesign. Concretely:

• Pick just one department, the one where the work is most repetitive and rule-based — that's where redesign pays off soonest and the proof is cleanest.
• Redesign the flow, not the people. Where AI comes in, where the human checks, which step disappears and which stays. This is where the return the three sources measure — or fail to measure — lives.
• Bring people into the new process, not in front of one more tool: it's the part that decides whether it takes root. How to do it concretely, one step at a time, we cover in how to get your team to adopt AI.
• Measure in hours and in euros before you start, so you know whether you're in the 30% that redesigns or among the many who stopped at access. What to really expect is in how much AI pays off in an SME.

Where we graft in

The part pilot projects skip — redesigning the process around AI — is exactly the part we do, inside the department, with the people who work there. We don't sell you the access: you already have that, like everyone. We redesign a real flow until the new way is simply the way, and the numbers confirm it. If the year of licenses didn't move your P&L, you don't need more AI — you need to redesign a department. The AI-readiness assessment is free and tells you which one to start with; then, if it makes sense, let's talk.

The data cited (Deloitte “State of AI in the Enterprise” 2026, MIT NANDA “The GenAI Divide” 2025, McKinsey “The State of AI in 2025”) are independent sources describing the market as a whole: they are not results of Innesti Digital nor outcomes from our clients. They serve to frame the gap between access and redesign. The real return of an intervention depends on the department, the use case, and the context of the individual company.