Why 81% AI Adoption Creates Almost No Value

According to Exact’s KMO-barometer, 81% of Belgian SMEs now use artificial intelligence, up from 8% a year earlier. Yet more than half do nothing with it beyond writing text, translating, or producing content.

Adoption is not the problem. It has already happened. The problem is that the tool is used to run old tasks faster, without any process, reporting line, or decision right being rethought. The history of general-purpose technologies makes the point: value appears only when the organisation redesigns itself around the new capability. That is a leadership choice, not a technical one.

The Shift

Exact’s figure is striking. In twelve months, AI use among Belgian SMEs rose from 8% to 81%. Adoption is no longer the question. It is settled.

The second figure from the same barometer is the real story. More than half of companies never go beyond three uses: writing text, translation, and content generation. The tool is everywhere, the value nowhere.

This gap is not particular to Belgium. McKinsey’s State of AI 2025 shows the same fracture worldwide. Adoption is broad, but fewer than four companies in ten report a measurable effect on operating results, and most often below 5%. Only a minority has moved past the pilot stage.

The conclusion holds from one country to the next. Owning AI produces nothing on its own. What produces value is the decision to reorganise the work around it. That decision belongs to the executive committee, not the IT function.

For a chief executive, this is good news. If 81% adopt and a handful gain an advantage, the competitive gap is not about access to the technology. It is about the operating model. That is precisely where a company can still differentiate.

The Problem: The Task Changes, The Process Does Not

Most organisations take a new engine and use it to turn the same old wheel slightly faster. Last year an employee drafted the email by hand. Today the machine drafts it. The task has changed. The process has not. The sequence of steps, the allocation of responsibilities, the approval points, all of it has stayed identical.

Two reasons lie behind the stall, and both are organisational.

The first is ease. Leaders let AI settle on the visible, low-stakes tasks, because they are safe. Drafting a note, summarising a document, translating a letter. It feels like progress while nothing is questioned. The gain scatters in thin slices across large populations, too diffuse to register on a profit-and-loss statement.

The second runs deeper: leaders do not know where to start. Faced with a general-purpose technology that applies to everything, most fall back on the nearest small task, because the real question, how to redesign the way work is done, has no obvious entry point. So they never enter.

As long as AI sits on top of processes designed for humans, it corrects neither their slowness nor their redundancy. It simply makes them faster. An organisation that accelerates a bad process gets a bad result faster.

The Framework: From Task to Process

Economic history offers the best guide. When the steam engine, and later electricity, entered industry, the first factories simply replaced the water wheel with an engine and kept producing the same way. Productivity gains came only a generation later, when owners redesigned the entire plant around the new power source: layout of the workshops, assembly lines, organisation of labour.

The economist Paul David documented this lag in his landmark study of the dynamo and the computer. Robert Solow summed it up in a line that became famous: you could see the computer age everywhere except in the productivity statistics. Brynjolfsson, Rock, and Syverson named the phenomenon the productivity J-curve: every general-purpose technology first demands a reorganisation effort before it delivers its gains. AI follows exactly the same path.

The lesson is directly actionable. Before any pilot, the executive committee must answer three questions, in order:

1. What business outcome are we after? Not which tool to deploy, but which lead time to cut, which cost to remove, which decision to improve. If the answer is “draft faster,” the ambition is too low.

2. If this capability is now free and instant, what should we stop or redesign? This question moves attention from the task to the process. It forces you to identify the steps that existed only to compensate for human slowness, and that no longer have a reason to exist.

3. Who holds the decision, and should that decision right change? A redesigned process redistributes authority. As long as every AI action requires routine human approval, there is no scale, only an expensive assistant.

A company that answers these three questions stops automating gestures. It redesigns a process. That, and only that, is where value appears.

Sector Lens

In recruitment and business services, the temptation is to plug AI into writing job ads or screening CVs. Real gain, marginal gain. The value is elsewhere: redesigning the process from the client request through to the candidate shortlist, removing the re-keying steps, redefining who approves what. The business changes structure, not just speed.

In banking, drafting a credit report faster changes nothing about the approval time if the validation chain stays intact. Redesigning that chain, by contrast, moves the time from several days to a few hours, subject to clear model governance.

The technical context differs from one sector to the next. The requirement is the same everywhere: rethink the process, not just tool the task.

For Belgian leaders, the regulatory deadline adds useful pressure. The EU AI Act requires traceability and oversight for high-risk systems from August 2026. An organisation that has redesigned its processes with governance built in from the start will move ahead of those that have stacked tools without a framework.

The Question to Ask Monday Morning

At the next executive meeting, ask a single question: for every AI use in place today, have we redesigned the process around it, or have we simply made the old gesture faster?

If the answer is “faster” everywhere, you are part of the 81%, not the handful gaining an advantage.