AI adoption in Italian SMBs: the 2026 data and where they're really stuck

Every week a CEO asks us the same thing in one form or another: “are we behind the others?”.

That’s the wrong question. The right one is “behind whom, and on what?”. Because the 2026 data paints an Italy where average adoption is low, but the real gap isn’t between who bought AI and who didn’t. It’s between who has the skills to use it and who doesn’t.

Key takeaways:

• According to 2026 Istat data, roughly 16.4% of Italian companies use at least one artificial intelligence technology: the average is low and pulled up by larger firms.
• The number-one brake isn’t cost or technology, it’s internal skills: without someone who knows where to apply it, AI stays an individual ChatGPT experiment.
• From August 2026 the AI Act is fully applicable, but for most SMBs the direct obligations are limited and don’t justify waiting.
• The benchmark that matters isn’t “how much AI you use”, it’s “how many high-volume processes you’ve automated with a measured result”.
• Before buying anything you need a baseline: without starting numbers you can’t tell where AI pays off, nor verify whether it worked.

What “AI adoption” means for an SMB

AI adoption, in an SMB, is the stable, repeated use of artificial intelligence systems inside real operational processes, not a subscription switched on nor a personal chatbot experiment. It’s measured not by the number of licences bought, but by how many high-volume processes produce output used in the business with a quantified result.

That distinction is everything. A company where 30 people use ChatGPT their own way has “adopted AI” in the statistics, but hasn’t changed a single process. A company with one agent processing invoices autonomously adopted fewer tools and got far more.

The 2026 Istat numbers: where we really are

According to Istat data cited in the 2026 report on AI skills, roughly 16.4% of Italian companies use at least one artificial intelligence technology. It’s an average, and averages mislead.

The share moves with size:

• Micro firms (under 10 employees) stay well below the average.
• Small firms (10-49) sit around or just below.
• Medium and large firms clearly exceed the average, doubling it in some sectors.

Translated: if you’re an SMB of 10-200 people, your real comparison isn’t with the national average, it’s with companies in your size band and your sector. And the mid band is exactly where the gap is widening fastest.

Why Italian SMBs are stuck (it’s not cost)

Here’s the most honest data point: the main brake isn’t software price. It’s internal skills.

We see it in every assessment. The company has budget, has obvious repetitive processes, sometimes already pays for an enterprise licence. What’s missing is someone who knows how to:

• recognise which process is worth automating;
• map how it actually works today, not how the manual describes it;
• measure before and after.

Without that skill, tools sit idle. The ChatGPT licence turns into scattered individual use, the automation starts and stalls at the first edge case. It’s not a technology problem. It’s an adoption problem, and it’s exactly the gap we tackle with our AI Adoption programs: we don’t sell an abstract course, we train the team to work on the company’s real processes.

And the AI Act? Not the alibi it seems

From August 2026 the AI Act is fully applicable, and many CEOs use it, knowingly or not, as a reason to delay. That’s a misreading.

For most SMBs the direct obligations are limited: transparency toward chatbot users, human oversight on critical cases, an art. 28 GDPR DPA with vendors. Only some uses (automated recruitment, credit scoring) fall into the high-risk tier with heavy obligations.

The AI Act isn’t a reason not to adopt AI. It’s a reason to adopt it with governance from day one: audit logging of decisions, a human in the loop on sensitive cases, EU hosting where needed. Things that, built well from the start, cost little and take the problem off your plate.

The benchmark that actually matters

Forget the national percentage. If you’re a CEO or COO, here are the questions that capture your real position:

1. How many high-volume repetitive processes have you identified and quantified? (Not sensed, quantified.)
2. How many of those currently have a system running them autonomously with human oversight?
3. For each one, do you have a before/after number that proves the result?
4. How many people on the team use AI on high-value tasks, not just out of curiosity?

If the answer to 2 and 3 is “zero”, it doesn’t matter how many licences you bought: you’re in the stuck part of the statistic. And it’s a more crowded place than you’d think, which means even one well-automated process moves you ahead of your direct competitors.

To measure your position in a structured way we wrote an operational guide to an internal AI assessment: you can run it yourself before you ever talk to us.

Where to start (and when NOT to)

The right move on a limited budget isn’t spreading across ten tools. It’s picking a single high-volume process with structured output that lands in your systems, and building an agent on it. At Soraia the first delivery lands in 4 weeks, so you get real data before investing further. That’s how APraise reached 100k+ candidates handled by the agent, equivalent to 4 extra recruiters, or how Numeraria handed roughly half a month back to management.

I’ll also tell you when not to start:

• If you don’t have a measured baseline, don’t buy anything. Time it first.
• If the team is under 10 people with highly varied tasks, an enterprise licence plus one or two automations is often enough, not a custom agent.
• If the process will change in three months, wait for it to stabilise.

These are the same rules we apply before proposing a custom AI agent: without starting numbers our “you only pay if it works” guarantee would make no sense, and we wouldn’t sign it.

The honest conclusion

Italy isn’t behind because technology is missing. It’s behind because the skill to apply it to the right processes and measure the result is missing. That’s also the good news: the gap doesn’t close with huge budgets, it closes with one well-chosen process, measured, and put into production.

Want to understand where you stand versus your size band? Let’s talk for 20 minutes, no pitch, or run the 3-minute check-up first.