81% of Italian SMEs use AI. One in four has integrated it.

There are fifty-six percentage points between the two numbers. In that gap sit most of the Italian businesses that have already paid for artificial intelligence and haven't yet got anything back.

The number that matters isn't 81%

The Observatory says it in a single line: only one in four has built it into its processes on an ongoing basis. The same survey adds that about one in three still uses it occasionally and without structure.

The 81% is quoted everywhere, and it should be read for what it is: a measure of uptake. It says almost every Italian business has opened a chat window, tried getting an email rewritten, asked for a translation. It's a threshold already crossed, and as such it no longer describes a competitive advantage: if eight businesses out of ten do it, doing it doesn't set you apart from anyone.

The number that describes a real difference is the other one. One in four has brought those tools into a process, meaning a point the work passes through every day, with a stable rule and a verifiable outcome. The rest have a tool open in a browser tab, used by the people who got curious, each in their own way.

The difference between the two is one of kind, not degree. A tool used at people's discretion produces results that depend on who's using it, don't add up, can't be measured and vanish when that person changes role. A process, on the other hand, produces the same outcome no matter who is on shift, and it's the only form in which an improvement stays with the business rather than with the person.

Why it stops right there

The same survey answers this question too, and the answer contradicts the objection we hear most often. Cost is named as the main barrier by only 6.6% of businesses. There are two real obstacles, and both are human: the skills to use it independently, and trust in how data is handled.

That turns upside down the way most suppliers frame the conversation. If price isn't the problem, a discount solves nothing. If the problem is that nobody in the business knows where to start, and nobody knows where the data goes once it leaves, then what's missing isn't a cheaper subscription, but someone to write down the process and state where the data goes.

Then there's a figure that makes the picture more interesting, not less: 60% of those using it say they save at least five hours a week per person. Five hours a week, for a single person, is about twenty-five working days a year. So the saving already exists, and it's reported by people using AI only occasionally. The question that follows is the only one that matters: if this is what you get without structuring anything, what's being left on the table?

The European data says the same thing in different words

It isn't an Italian quirk. An OECD survey of over 5,000 SMEs in seven countries (Austria, Canada, Germany, Ireland, Japan, Korea and the United Kingdom) finds that 65% of SMEs using generative AI report improved staff performance, far more than those saying they've used it to scale up (35%), to compete with larger companies (29%) or to increase revenue (26%).

That ranking deserves attention. The benefit businesses recognise first is internal: people work better. What they recognise much less is the effect on turnover. That doesn't mean the effect isn't there: it means that, in most cases, nobody has measured how one leads to the other. You can feel the time saved; where that time went, you can't.

The same survey adds a detail that works as an operating instruction: the benefits (time saved, quality, job satisfaction) turn out 10% to 40% greater when the employer encourages its use, rather than leaving it to individual initiative. That says nothing about the technology. It says a lot about how the business is run.

What, in practice, sets the 25% apart from the rest

It isn't how sophisticated their tools are. The businesses that have integrated almost always use exactly the same tools as the others. What they have in addition are four things, all of them dull:

• A written process. They can say, without hesitating, at which point in the work the system steps in, what it receives and what it produces. Businesses that haven't integrated answer this question with the name of a tool.
• A baseline measurement. They know the before figure: how many hours, how many documents, how many minutes of waiting. Without that number, the improvement is a feeling with an invoice attached.
• A stated human checkpoint. They know exactly what the system can do on its own and what has to go through a person. Not out of mistrust: because it's the only way responsibility stays where it belongs.
• A written answer on data. Where it's processed, who the supplier is, whether the supplier can use it to train its own models. It's the survey's second barrier, and it's overcome with a contract, not a verbal reassurance.

None of these four things is technology. They're all decisions, and they're all made before anyone builds anything.

What it means for the people who decide

If your business is in the 81%, you're not behind: you're where almost everyone is. The gap to close isn't with the businesses using AI, because that race has already ended in a draw, but with the quarter that has built it into a process, and that gap is closed with work that has nothing technological about it: looking at where the work gets stuck, measuring it, and building a single piece where the loss is most obvious.

How to look, which four criteria to use in choosing the process, and what to measure before switching anything on, are covered in the method: which process to automate first.

It's also why you won't find packages with a price next to them on this site. Which piece comes first depends on a number we don't have today, and that can't be guessed from the outside.

Questions and answers