When people talk about artificial intelligence in cafés and bars, the image that comes to mind is a robotic arm making a cappuccino. It is also the least useful image. Coffee at the counter remains, for now, a job done with hands and relationships. What AI can change is everything else: how many croissants to order, how to plan shifts, how to answer reviews, how to train a new hire quickly.
And for an Italian bar today, everything else matters a lot.
The sector in numbers
According to FIPE’s Rapporto Ristorazione 2025 (FIPE is the Italian association of bars and restaurants), at the end of 2024 Italy had 127,667 active bars, 3.3% fewer than the year before. In 2024, 3,959 opened and 12,368 closed: a net loss of 8,409 businesses.
The problem is not only economic. The same report finds that about 70% of the bars that looked for staff had trouble hiring baristas, counter staff, waiters or dishwashers. Meanwhile innovation is slow: in 2024 only 7% of bars invested in digitizing their management processes.
Thin margins, missing staff, management still largely done by hand. That is exactly the ground where AI tools have something to offer.
Where AI does not reach (and where it has already failed)
Tyna Eloundou and colleagues (Science, 2024) estimated how much of each occupation’s tasks is exposed to language models like ChatGPT. In their analysis about 80% of US workers have at least 10% of their tasks exposed. Yet the list of occupations with no exposed tasks at all includes dining room attendants and bartender helpers, along with dishwashers. Preparing, serving and cleaning are physical activities that a language model does not perform.
Even where AI could replace a person, the results are not a given. Starting in 2021, McDonald’s tested automated voice ordering with IBM in about 100 US drive-thrus. In June 2024 it ended the test. According to a 2022 analysis cited by the trade press, order accuracy stayed just above 80%, while the company’s target was at least 95%. If it does not work well at a drive-thru, it is hard to picture it at the counter of a busy bar, among spoken orders, regulars’ habits and requests like “the usual”.
Where AI really changes the work
Sales forecasts and waste
This is the application with the most concrete evidence. A study published in the Journal of Industrial Ecology (Hübner and colleagues, 2024) analyzed a machine-learning sales forecasting service used by bakeries: in 2022 it cut unsold products by 30% on average, mostly bread and rolls.
A bar’s situation is very similar: croissants, pastries and sandwiches are ordered the evening before, and whatever is left gets thrown away. Globally, according to UNEP’s Food Waste Index, food service wastes about 290 million tonnes of food a year, more than a quarter of all food waste. Forecasting demand better (weather, day of the week, local events, sales history) means less waste and better margins.
Orders and inventory
Modern cash registers already know what sells and when. Connecting that data to supplier orders, with alerts when stock falls below a threshold, does not require big investments: often it is enough to configure the software you already have properly, or to use no-code automation tools.
Shifts and organization
The same footfall forecasts help build shifts: more people at the counter at peak hours, fewer in quiet moments. In a sector where finding staff is hard, using the available hours well is worth as much as hiring.
Communication and reviews
Answering reviews, writing social posts, translating the menu for tourists, describing a new item: these are writing tasks, and this is where generative AI tools are strongest. The guide on how to use ChatGPT at work explains how to do it without impersonal text, and the one on effective prompts how to get usable results on the first try.
Training new hires
This point is the least obvious and perhaps the most important. Brynjolfsson, Li and Raymond (Quarterly Journal of Economics, 2025) studied 5,179 customer support agents who were given an AI-based assistant. Productivity rose by 14% on average, and by 34% for the least experienced: the assistant passed the best colleagues’ practices on to newcomers.
The context is different, but the mechanism fits a bar with high staff turnover well: procedures, recipes, allergens and answers to frequent questions gathered in an assistant the new barista can check during a shift.
| Task | What AI can do | What stays human |
|---|---|---|
| Ordering croissants and goods | Forecast demand from sales history | Deciding the product range |
| Inventory | Stock alerts, draft orders | Choosing and dealing with suppliers |
| Shifts | Estimate footfall by time slot | Managing people and the unexpected |
| Reviews and social media | Draft replies, texts, translations | Tone, checking, customer relationship |
| Training | Assistant with procedures and recipes | Teaching the craft and hospitality |
| Counter service | Little or nothing | Almost everything |
The skills people working in bars need
For anyone who runs a bar, or wants to grow into running one, three skills are gaining value.
Reading sales data. Knowing which products pay off, at what times of day, with how much waste. It is the basis of any forecast, whether done by hand or with software.
Using AI tools for writing and organization. You do not need to code: you need to ask well and check the result.
Hospitality and relationships. This is the part AI does not touch, and for that very reason it becomes what sets one bar apart from another. The barista who remembers a regular’s name and order does a job no tool replicates.
Where to start
1. Measure waste for a week. Every evening, write down what is left unsold. Without this figure you do not know what better forecasting is worth.
2. Export your till data. Almost all electronic cash registers and point of sale software let you download sales. Two or three months of history are enough to see patterns by day and by time slot.
3. Start with writing. Replies to reviews and menu translation are the first use at almost zero cost, and they give a visible result right away.
4. Write down your procedures. Recipes, opening, closing, allergens. A tidy document helps new hires even without AI, and becomes the basis for an assistant when you want to build one.
5. Be wary of promises of full automation. The McDonald’s case shows that even large companies with large technology partners can stop. Judge tools on a specific, measurable problem.
My take: AI helps the bar that remains a place for people
My reading is that AI will not change the Italian bar from the counter, but from the back office. Bars that close do not close because the coffee is bad: they often close because of thin margins, waste and difficulty finding and training staff. Those are exactly the problems on which the evidence gathered here points to concrete help. The rest, meaning hospitality, remains the reason people walk in. It is a hypothesis, not a result measured on Italian bars, which does not exist yet.
Frequently asked questions
Will artificial intelligence replace baristas? In Eloundou and colleagues’ estimates, dining room and bartender helper jobs have no tasks exposed to language models. AI works on management, not on service.
Are there robots that make coffee? Yes, but they remain isolated cases. The largest documented experiment in automating orders, McDonald’s with IBM, ended in 2024 because of accuracy problems.
How much does it cost to bring AI into a bar? The first uses (reviews, social media, translations, analysis of till data) can be done with free tools or ones already included in the software in use. Costs grow with automatic forecasts connected to orders.
What is the most concrete benefit? Less unsold stock. In the bakery case studied, machine-learning forecasts cut returns by 30% on average.
What skills do you need to work in a bar that uses AI? Reading sales data, using writing and organization tools, and above all hospitality, which remains the skill that makes the difference.
Sources
- FIPE Confcommercio, Rapporto Ristorazione 2025, 2025 (in Italian). 127,667 active bars in 2024 (-3.3%), 3,959 openings and 12,368 closures; about 70% of bars had trouble hiring; 7% of bars invested in digitizing management processes.
- T. Eloundou, S. Manning, P. Mishkin, D. Rock, GPTs are GPTs: Labor market impact potential of LLMs, Science, 2024. About 80% of US workers with at least 10% of tasks exposed; list of occupations with no exposed tasks in the extended version.
- A. Kelso, McDonald’s is ending its AI drive-thru test with IBM, Nation’s Restaurant News, 2024. Test in about 100 restaurants since 2021; order accuracy just above 80% (BTIG analysis, 2022) against a 95% target.
- N. Hübner, J. Caspers, V. C. Coroamă, M. Finkbeiner, Machine-learning-based demand forecasting against food waste: Life cycle environmental impacts and benefits of a bakery case study, Journal of Industrial Ecology, 2024. Average 30% reduction in bakery unsold products in 2022.
- UNEP, Food Waste Index Report 2024, 2024. 1.05 billion tonnes of food wasted in 2022, of which 290 million in food service.
- E. Brynjolfsson, D. Li, L. R. Raymond, Generative AI at Work, Quarterly Journal of Economics, 2025. 5,179 support agents; +14% productivity on average, +34% for the least experienced.



