The short answer: in most Italian SMEs an artificial intelligence project doesn't stall over the model chosen or the cost of the licence. It stalls because the company's data isn't in a fit state to be read. The companies say so themselves: among those that considered investing in AI without going ahead, 45.2% name the unavailability or poor quality of data among the obstacles, and 58.6% name a lack of skills (Istat, the Italian national statistics institute, Imprese e ICT, 2025).
It's a point that gets little airtime, because it isn't what people sell. A conversational assistant demos well in a meeting; a customer record aligned between the ERP and the online store does not. And yet the second decides whether the first will produce anything useful.
This article lines up the available figures, explains what "the data isn't ready" actually means in a company of 10 to 50 people, and sets out what is worth sorting out before buying anything.
How many Italian SMEs really use AI?
Few, but the number is climbing fast. According to Istat (Imprese e ICT, 2025), in 2025 16.4% of companies with at least 10 employees used artificial intelligence technologies. The figure was 8.2% in 2024 and 5.0% in 2023: it doubles every year.
The average hides the size gap, though. Among small companies of 10 to 49 employees the share stops at 15.7%; among those with at least 250 employees it rises to 53.1%.
On future investment, the Politecnico di Milano's Digital Innovation in SMEs Observatory (May 2026) finds that 76% of SMEs have neither invested nor plan to invest in AI, and that only 7% have started structured training programmes on the subject.
Why projects stall: what companies say
Istat asked companies that had considered AI without adopting it which obstacles they had met. The answers, in order:
| Obstacle | Share |
|---|---|
| Lack of skills | 58.6% |
| Lack of legislative clarity | 47.3% |
| Unavailability or poor quality of data | 45.2% |
| Privacy concerns | 43.2% |
| High costs | 43.0% |
| Ethical considerations | 25.7% |
| Judged not useful | 14.8% |
Source: Istat, Imprese e ICT, 2025.
Two observations. First: only 14.8% think AI isn't useful. So the problem isn't conviction, it's the starting condition. Second: the top three — skills, the regulatory picture, data quality — are organisational obstacles, not technological ones. None of them is solved by buying a better tool.
What "the data isn't ready" actually means
"Data quality" is a phrase that means nothing inside a company until you translate it into everyday facts. In an SME it sounds like this:
- The same customer exists three times: in the ERP under the company name, in the online store under the personal email of whoever ordered, in the salesperson's spreadsheet under a nickname.
- There is no field saying where a lead came from, so nobody can say which campaign produced which order.
- The status of a deal lives in the head of the person handling it. If that person is on holiday, the information doesn't exist.
- Quotes are documents in a shared folder, not rows in a system: they can't be counted, searched or added up.
- The monthly report is rebuilt by hand every time, and every time with slightly different criteria.
An assistant built on a language model and connected to this base returns exactly what it finds: three versions of the same customer and no attribution. It isn't getting it wrong — it's reporting. That's why so many pilots close after three months with a sense that "AI doesn't work".
What comes first: four foundations
Before any artificial intelligence project, a company of this size needs four things. They aren't optional, and they don't cost as much as people expect.
1. A single customer record. One definition of "customer", shared between the ERP, the online store and the CRM, with a written rule on who creates it and how. It's the least glamorous item on the list and the one that unlocks all the others.
2. A CRM that follows the real process. Not the one in the manual: the one your salespeople actually follow. If the system asks for steps nobody takes, the data doesn't go in and the CRM becomes a dead archive. It's worth remembering that, according to Istat (2025), a CRM is used by 21.1% of SMEs against 56.5% of large companies: a gap of more than 35 points. On how to connect it to the rest of your systems, I've written a separate article: integrating CRM and ERP.
3. Tracking from the campaign to the closed order. Every lead carries its source with it, and that information survives all the way to the invoice. Without it, every discussion about ad budget is an opinion.
4. One view that somebody actually looks at. A dashboard containing only the numbers decisions are made on. If it holds forty metrics, nobody opens it and we're back where we started.
With those four things in place, AI becomes an inexpensive addition: the assistants have something to read, the automations have defined steps to act on, and the results are measurable because there is a before and an after.
How to tell whether your company is ready
Six questions. If you answer "no" or "I don't know" to three or more, the problem isn't AI.
- Do you know how many leads came in last month, and where from?
- If a salesperson were away for two weeks, could someone else pick up their deals from the system?
- Does the same customer have the same identifier across all your tools?
- Can you say which euro of advertising produced which order?
- Does the report the owners look at update itself?
- Is there someone inside the company responsible for data quality?
What it costs to put the foundations in order
It depends on the mess you start from, and anyone quoting a figure without having looked is guessing. What can be said is how the spend is usually structured: an analysis phase with fixed scope, then a build, then light oversight so the system doesn't slide back into disorder.
In my case the first phase is an analysis: two weeks, from €2,500, fixed scope. It produces a map of how leads, orders and data move, the points where the flow breaks quantified, a 90-day plan with priorities and costs, and an honest assessment of what AI would actually deliver in your case and what it wouldn't. If the conclusion is that you should stop there, I write that in the document — which is yours, and which you can have anyone carry out.
Frequently asked questions
Do you necessarily need a CRM before using AI?
Not necessarily a formal CRM, but you need one reliable place where leads and deals live. In practice, for a company of 10 to 50 people, that place is almost always a CRM. If you already have one sitting unused, making it work costs far less than replacing it.
How long does it take to sort the data out?
The analysis takes two weeks and two hours of the owner's time. The build, in most cases, six to ten weeks. It isn't a multi-year programme: companies of this size have a contained number of systems, which is exactly why the work is feasible.
Can we start with AI and sort the data out later?
You can, but you pay twice: first for the pilot that produces nothing, then for the tidying up that was needed at the start. And you spend internal goodwill, which is the hardest resource to recover when you try again.
Who should own this inside the company?
You need one point of contact, not a department. Usually it's whoever already holds the pieces together informally. They need training and they need the authority to decide on customer records: without that responsibility assigned, the mess reforms within months.
In short
Adoption of artificial intelligence in Italian companies doubles every year, but it still sits at 15.7% among small companies. Among those who tried and stopped, almost one in two names data quality as an obstacle. The practical conclusion is simple: the foundations aren't a dull preliminary to get through quickly, they're the part that decides whether everything else will work.
If you want to work out where you stand, let's talk for thirty minutes. There's always something to improve: in those thirty minutes I'll tell you where I'd start. You'll find how I work and who I am on the rest of the site.
Sources
- Istat, Imprese e ICT — 2025, press release of 15 December 2025. Population: companies with at least 10 employees.
- Digital Innovation in SMEs Observatory, Politecnico di Milano, press release, May 2026.
The figures quoted are primary-source data with the year stated. The assessments of cost and timing are based on my own direct experience and should be checked against your case.