25 October 2026
What a panzerotti shop at the station taught me
Between January and March 2018 I was the manager of a daPrette shop in Padua. It was not a long assignment and I will not inflate it. It was the opening of a shop in a busy spot at the station: hiring and supervising staff, the accounts, and a marketing campaign so people knew the shop existed. We sold fresh panzerotti, cooked and heated, made with Italian products. Three months. Long enough to see what a day is like when the whole firm fits into a few square metres.
Before that shop I had already founded Frank & Jhon, in 2017: importing and customising condoms, a project later sold. Afterwards I would lead Universal, with the Yellow Innovation brand, and other companies. But the station shop stayed with me as a blunt lesson. In a slightly larger firm you can still tell yourself the problem is strategic. In a shop that is opening, the problem is the shutter, the person who did not show up, the receipt, and the fact that nobody yet knows you are there.
Opening is a sequence, not an announcement
Opening is not a post. It is a sequence of material things. The shop has to be set up. People have to be chosen, hired and followed, because a counter does not run itself. The books have to be written, not remembered. And someone has to tell passers-by that there is something to eat behind that window. My role held all three: staff, accounts, campaign. I am not listing them for a résumé. I am listing them because they are the three places where a small operation jams, even when the product is simple.
People, books, being seen
Staff, in a new shop, have no long memory to lean on. Every exception — a shift, an absence, a customer question you did not expect — climbs back to the person who opened. If that person is also the one doing the books and thinking about how to be seen, the day fills with interruptions. This is not theory. It is the shape of the work when there are few heads.
The accounts of a small shop are not a sustainability report. They are knowing what came in, what went out, what was ordered. If that lives in a notebook, or a chat, or the head of whoever closes the till, the next day is a reconstruction. It can be done. It wastes time, and the mistakes land on the boring things, which are the things that later do not add up.
The marketing, in that case, was letting people know we existed. I do not have a customer count to tell, and I will not invent one. I have the operational memory: without a simple way to say “we are open, we are here”, a counter stays a counter. A campaign, even a modest one, is already a process. It has to be repeated, not reinvented every morning.
What I would look at today
Today, when I walk into a company, I do not start with software. I start with a list of what repeats. If I look back at those three months with today’s eyes, the list would have been concrete. Imagine — and this is an exercise, not a project I ran in 2018, when these tools were not discussed this way — a shop with a few people at the counter. The customer questions that return. The shifts. The reorder. The small end-of-day note. None of these items is “innovation”. They are the work. Custom AI, if it were ever useful, would sit there: not to replace the person at the counter, but to take the repetitions off the person who has to hold opening, people and books together.
A line
I will draw a line. In 2018 I did not put an agent behind the panzerotti counter. That would be a convenient lie. What I did was open, hire, follow, count and try to be seen. The lesson is not technological. It is that the daily problems of a minimal firm are already all the problems I later find, enlarged, in more structured companies: communication that scatters, exceptions that climb, paper or messages instead of a clear handoff.
At Universal, where I was chief executive from 2018 to 2021 and then general manager until March 2022, the perimeter was different: innovation in agriculture, software, supply chains. Yellow Innovation was the brand of that project. The station shop was the opposite extreme, and that is why it is useful to me. If you cannot run a counter, be wary of anyone who talks about processes as if they were an org chart. The org chart does not open the shutter.
From there, the step to AI is shorter than it looks. Not because a panzerotto has anything to do with a language model. Because the question is the same: what, today, is done again by hand every day, and who is the person holding it in their head? If the answer is “the owner”, you have a switchboard, not a process. If the answer is a short, true list, then a tool can be designed. And someone, afterwards, checks the result. Always.
Three months of a shop do not make a career. They make a measure. When a company tells me it is too small to put things in order, I think of a counter at a station. It was small. Order was needed anyway: people, books, and a way not to start from zero every morning.