Every owner of a smaller company we have talked to about AI has, sooner or later, said the same thing: "that stuff is for people with an IT department and a budget we don't have". It is the most common thought among companies with 15, 30, 80 employees, and it is also why most small companies are still standing still. The problem is that it is only half true. The two cases that taught us the most on this subject do not come from enterprises with a CTO on the team: they come from an energy reseller with 12 people and a construction general contractor with around sixty employees. Neither of them bought "an AI project". They picked a small, specific process that was bleeding every day, and they fixed it.
The gap is real, but it is not what you think
The data is there and it is stark. According to the Istat survey Enterprises and ICT 2025, at least one AI technology is used by 53% of large Italian companies, 27% of mid-sized ones and just 14.2% of small ones. The distance by company size is growing, not shrinking, and that describes a real lag.
AI adoption in Italy: the gap grows with company size
But the most common reading of that number, "smaller companies are stuck because they lack the resources of large ones", is the wrong one. When Istat asks companies that evaluated AI without adopting it what the main obstacle was, the most frequent answer is not cost. It is a lack of skills, in around six cases out of ten. Budget is not the number one bottleneck. Not knowing where to begin is, and choosing badly when you try.
"Neither company started with AI in the abstract. They started from the process that was costing them the most, made it measurable, and only then chose the tool."
What the two cases that worked have in common
From the workbench. Two real companies, Morfeus clients, both small in the full sense of the word: no IT department, no enterprise budget.
The first is an energy reseller with 12 people. The problem was not "which AI do we buy": it was that incoming leads piled up in a spreadsheet and nobody knew for certain who to call first, or within how long. Not a technology problem, a process problem. We built a system aimed at that alone, second contact on hot leads. The second-contact rate went from around 50% to over 90%. We did not touch the rest of the company.
The second is a construction general contractor, between 15 and 80 staff depending on how many sites are open. They were sending quotes worth tens of thousands of euros and did not even know whether the client had opened them, let alone read them carefully. We hooked the sales follow-up to the actual opening of the quote: if the client opens it and does not reply within a certain time, a targeted contact goes out, not a generic reminder. The close rate has been rising since that mechanism went live.
What the two cases have in common is not the sector, which is completely different. It is that neither company started "with AI" in the abstract. They started from the specific process that was costing them the most, made it measurable, and only then chose the tool. This is also why the right order is almost always the opposite of how it gets told: first the number you want to move, then the technology.
Where to actually start: the process before the tool
The most expensive mistake we have seen smaller companies of every sector fall into is beginning with the question "which tool do I buy?". It leads straight to demos, licences and pilots that stay outside the real work, because nobody decided in advance what that tool was supposed to change.
The question that works is the opposite: where am I losing value every day, in a way the people working there can already see? These are the Value Leaks, the invisible losses in the processes. Not every process is worth the same effort: the right ones to start from have three characteristics together.
- High volume: it happens several times a day or a week, not once a month.
- Repetitive and rule-based: it follows predictable steps, not creative judgement case by case.
- Errors cost: a delay, a rework or a lost customer has a price you can write in euros.
High volume
It happens several times a day or a week. A process that runs once a month does not repay the setup, even when it is badly done.
Repetitive and rule-based
It follows predictable steps. If every instance needs different creative judgement, the machine does not hold and you need a person.
Errors cost
A delay, a rework, a lost lead have a price in euros. If the cost is not visible, the ROI cannot be defended in front of a CFO.
Where all three are present, as in the energy reseller's leads or the general contractor's quotes, the return comes sooner and is measurable from the first month. In smaller Italian companies, according to Istat, the areas AI most often enters today are marketing and sales, the organisation of administrative processes, and research and development: quoting, data entry and cleaning, recurring reporting and first-line customer support are the typical candidates. The criterion is not which one is most innovative, it is where the loss is largest and easiest to measure.
Skills are the real bottleneck
Back to the number that really matters. If the main brake is a lack of skills rather than costs, the practical consequence is that buying the right tool is not enough. You need somebody in the company who knows how to use it, understands when it works and when it does not, and spreads that to the others.
There is also a regulatory point that often goes unmentioned in smaller companies: since 2 February 2025, article 4 of the European AI Act requires anyone using AI systems to ensure an adequate level of literacy among the people working with them. It is not an obligation designed only for large groups, it applies to a company of 15 people too.
The answer is not sending everyone on a generic course about "artificial intelligence", which achieves very little. It is creating an internal reference point per department, a person who experiments on the chosen process, works out what works in the company's specific context and brings it to the others: an AI Champion. At the energy reseller and the general contractor this role already existed, informally, before we even gave it that name: somebody keeping an eye on the system and correcting course. Making it explicit is what makes the result last beyond the first month.
What it costs, and how to measure before spending
"What does AI cost a company" is almost always the wrong first question, because it has nothing to compare against. The useful question is: what is the problem I want to solve costing me today, in hours, in errors, in customers going cold while they wait for an answer? Only by comparing the cost of the problem with the cost of the solution do you see whether moving makes sense, and with what priority against other processes. The tool we use with clients to make that number explicit, department by department, is the ROIometer.
What is your worst-performing process costing you today?
The ROIometer translates the loss into euros per month, department by department, before you pick any tool.
The position to start from
You do not need a large-company budget to start with AI. You need to pick one process, a single one, small and specific, that is already costing you visibly today: a lead you do not call back in time, a quote nobody follows up after sending. You start there, you measure, you automate the repetitive part, and you train one person to keep the system alive. The rest of the company waits, and that is fine.
Start from the number, not the tool. Try the ROIometer and find out what the process wasting the most of your time is costing you today. If you want to understand the basics first, start from Learn AI.