Gastronomy·AI · Gastronomy · Operations
AI in gastronomy: what actually runs in a real bar.
AI in gastronomy is not a question of the future. In my own bar in Berlin an AI reads the till report every evening, calculates staff costs and result and proposes the schedule. It answers inquiries, writes event texts from templates and checks what ChatGPT says about the venue. Here is what of that takes work off your business and what does not.
What runs.
Numbers: photograph the till report, result in the morning. Schedule: proposal from revenue curve and availability, approved by the business. Inquiries: reservation and event inquiries by email are read, sorted and answered with a draft. Texts: events, menus, newsletters from your templates, reviewed. Visibility: what AI engines say about the venue, measured monthly.
What does not run.
A phone assistant that turns guests away because it does not understand the venue. Review replies everyone recognises as a machine. Robots at the counter. AI takes over administration, not hospitality. And everything it writes is reviewed by a human before it goes out. Chatbots and generated texts are labelled, as the AI Act requires since August 2026.
Why now.
Minimum wage 13.90 euros, 14.60 in 2027. Real revenue in Berlin below the previous year. 18,785 open positions in Berlin hospitality in May 2026. Every hour of administration a machine takes over is an hour with the guest or an hour of sleep for the owner.
Steps.
- 01
Visit
One evening in the venue, a look at till, inbox and schedule.
- 02
One use case
The one with the most hours. Fixed price, in use within weeks.
- 03
Next
By usage, not by catalogue. Monthly operation.
Questions
- What does AI for gastronomy cost?
- A first use case, the close-out for example, has a fixed price for setup and a monthly amount per venue. No fee per user. You know both figures before the start.
- Do I have to change my till?
- No. The tools work with the close-out receipt or export of any till.
- Where is our data stored?
- On servers in the EU, in your possession. No training data for foreign models, data processing agreements.
- Can I see it beforehand?
- Yes, in live operation at Honey Lou in Berlin.