.png)
How well do AI sales forecasts really work in the catering industry – not on slides, but in everyday operations? This is precisely the question we are addressing. GoNina and the SV Group, one of Switzerland's leading gastronomy groups, investigated this in a joint pilot project. Over two months, the GoNina -Forecasts tested in practice at ten community catering locations - from staff restaurants to educational institutions to public companies.
This article shows how the pilot was set up and what the result was.
The structure: Two phases, one clear scale
Phase 1 - Lab. In the first month, the AI forecasts ran in parallel with daily operations without interfering with planning: For each location and each day, the GoNina The forecast of actual sales volume was compared with the planning of experienced restaurant managers. This resulted in a fair comparison under real-world conditions – without risk to the business.
Phase 2 - Live. In the second month, the forecasts were available to the teams for daily production planning. At the same time, the models were continuously developed further – every additional day of sales contributed new training material.
The standard was deliberately chosen to be tough: not a comparison with a simple rule of thumb, but the real-world planning experience of professionals who have known their businesses for years.
The results: Highly precise – with a steep learning curve.
The key result of the pilot project: The AI forecasts proved to be very accurate across all locations – from the staff restaurant to the cafeteria, under completely different demand profiles, weekly rhythms and guest structures.
Just as important as the initial level was the development: The models were continuously improved throughout the pilot project and became significantly more precise over time. This is precisely the strength of machine learning in practical applications – while experiential knowledge grows over years, a model improves over weeks, and this is done individually for each location.
Important for context: The aim of the forecasts is not to replace the experience of the restaurant teams, but to relieve them - the machine calculates the patterns across all locations and menu lines, the professionals concentrate on what is not in the data.
What the pilot shows about community catering
The fluctuations can be learned. Home office rhythms, holidays, weather, and location profiles appear chaotic, but follow patterns – and the model has learned precisely these patterns from the sales data, per location and menu line.
The learning curve is crucial. A predictive model is weakest on the first day and improves with each passing week – anyone who measures its performance at the starting level systematically underestimates where the system will be after just a few weeks.
Humans remain involved. The forecast is a suggestion, not an automatic process: Restaurant teams retain the final say – and contribute what isn't reflected in the data, from major events to construction work outside.
Frequently Asked Questions
Is it possible to conduct such a pilot project in smaller companies?
Yes - the principle of laboratory comparison, then live deployment works from a single location: The forecasts first run parallel to the existing planning and prove themselves on the company's own data before being incorporated into everyday practice.
What data is needed to get started?
The recorded sales or point-of-sale data from the business provides the training material. This is supplemented by the planned menus and dishes on offer. The connection sets up the GoNina -Team in the onboarding process.
Conclusion
The pilot project with SV Group demonstrates the potential of AI forecasting in the catering industry: precise predictions under real-world conditions, with a learning curve that becomes visible over weeks – validated at ten locations in two months of practical operation. For an industry that produces ahead of demand daily, this provides the foundation for less overproduction and more stable planning – supporting the teams, not replacing them.
To learn how AI sales forecasts work in the restaurant industry as a whole, check out our comprehensive guide for the restaurant industry.
