Which process first. The method, not the list of tools.

The question that stops almost every small business isn't which tool to use. It is where to put it, the first time, so that it produces something measurable instead of a curiosity.

This piece is the method. The snapshot of the gap between those who use AI and those who have integrated it, with the numbers and the barriers companies report, is in 81% of Italian SMEs use AI and one in four has integrated it. Here we start from that figure and answer the next question: where to start.

Why the list of tools doesn't help

A list of tools answers the wrong question. It says what exists, and what exists changes every few months, while the structure of a company's processes rarely changes. Whoever chooses starting from the tool ends up adapting the work to the software instead of the other way round.

There is also a more concrete reason. The same survey that measures the 81% uptake finds that cost is named as the main barrier by only 6.6% of businesses: the real obstacles are the skills to use it independently and trust in how data is handled. Neither is solved by choosing another tool.

First step: map, don't imagine

The map is built by looking at what really happens in a week, not what should happen according to the organisation chart. You take a sheet and write down the processes that take up someone's time, one per row, with a description that ends with a verb: what that process produces, and in what form.

The rule that makes this step useful is just one: write only what you have seen or what someone has told you they did. An imagined process, or one described as it should be, produces a map that looks like a manual and doesn't help decide anything.

For a small business the list is short, between ten and twenty rows. If fifty come out, you are almost certainly writing activities and not processes, and it pays to go back: a process is a chain that starts with something that arrives and ends with something that goes out.

Second step: the four columns that decide

Next to each row you write four values. Precise numbers aren't needed: you need orders of magnitude comparable with each other, because the decision comes from the comparison and not from the exact measurement.

Column	What you write	Why it matters
Frequencyhow often it happens	Times per week or per month. An order of magnitude is enough: ten, a hundred, a thousand.	It is the multiplier. An improvement on a process that happens twice a year doesn't show, however good it is.
Repetitivenesshow much it is always the same	Whether whoever does it follows a stable rule, or decides case by case using experience they have never written down.	A process governed by a rule can be described, and what can be described can be delegated. Implicit judgement can't.
Cost of errorwhat happens if it goes wrong	Who notices, how quickly, and how much it costs to fix. A customer noticing is the worst case.	It is the brake. The first process is chosen where a mistake shows straight away and gets corrected, not where it leaves the company.
Ownership of datawhose information it is	Whether the process touches customer data, personal data, prices, or only material already public.	It is the second barrier companies report. A written answer on where the data ends up should be given beforehand, not afterwards.

The first process to automate is the one with high frequency, high repetitiveness, low cost of error and not very sensitive data. It isn't the most interesting, and it is almost never the one people in the company were talking about: it is the one where an improvement shows quickly and a mistake leaves no mark.

Third step: the baseline measurement

Before changing anything, a number must be taken by hand on the chosen process. It must be a number the company already has or can count in an afternoon: how many minutes pass between arrival and reply, how many files are closed in a day, how many enquiries go without follow-up in a week.

It is the step almost everyone skips, and it is the one that makes everything else verifiable. Without the before number, any result afterwards becomes an impression, and impressions of one's own processes are systematically generous.

An improvement without a baseline measurement isn't an improvement. It is a feeling with an invoice attached.

Fourth step: the human checkpoint

Before switching anything on it must be decided, in writing, who looks at what. The rule we follow is written in Ethics: the system prepares and flags, but no message that commits the company reaches a customer without a person having read and approved it.

On an incoming reply the rule applies differently, and it is worth knowing before choosing the process. Human approval is there anyway, but it comes earlier and concerns the content: information already approved by the owner, such as opening hours and availability, can go out on its own, while offers, prices and confirmations can't. The distinction, with its limits, is in how we reply to incoming enquiries.

There is also an obligation, and it concerns transparency. Article 50 of the European regulation on artificial intelligence applies from 2 August 2026 and requires that whoever interacts with an AI system be informed of it. What it means for a small business is in AI Act for Italian SMEs.

Fifth step: what you look at afterwards

After switching on you look at three things, and none is the satisfaction of those using it. The first is the baseline number, taken again in the same way and at the same point. The second is how many times the human checkpoint corrected something. The third is how many times the process stopped altogether.

The second is the most informative and the most neglected. A correction rate that falls in the first weeks and then stabilises says the system has learned the ordinary case; a rate that stays high says the chosen process was richer in judgement than it seemed, and that it pays to go back to the map.

It is also why the first process should be chosen where a mistake gets corrected. A high correction rate, on a process with a low cost of error, is useful information. On the same process with a high cost of error, it would have been damage.

What this method doesn't promise

It doesn't promise a savings percentage, because it depends on the process, the volume and how orderly it already was. Whoever gives you a figure before looking at your numbers is selling, and the figure they give you has no way of being true.

What can honestly be said is conditional and concerns you: if that process happens two hundred times a month and each time takes ten minutes of someone's time, those minutes are thirty-three hours a month, and the question becomes what thirty-three hours are worth in your company. You do the sums, with your own numbers, and they are the only ones that mean anything.

Where the map ends and the work begins

The map is an afternoon's work and needs no tool. It serves to move from the question «should we use artificial intelligence?», which has no answer, to the question «can this process, which happens a hundred times a month, be described well enough to delegate it?», which has one.

When that question has a positive answer, the next step is building the piece. The systems we put into production, one per process, are listed in the services, and each has its own page with what it does and what it doesn't. The map serves to choose which one, not to buy them all.

Questions and answers