For catering directors, contract catering operators, and enterprise restaurant managers.
Every morning, or several days before service, we turn your meals-served history, opening calendar, and available site signals into operational attendance forecasts. The goal is simple: adjust what can still be adjusted, reduce avoidable overproduction, limit last-minute fallback meals, and protect service quality.
Why Now
Attendance varies with hybrid work, public holidays, school holidays, strikes, weather, and site-specific events. Since remote work has developed, forecasting diner counts has become a more visible operational issue. The French Ministry of Agriculture cites meal reservation as one way to fight food waste in collective catering.
In ADEME’s 2025 report, the complete food-waste cost in workplace catering averages EUR 0.66 per meal. For a mid-sized workplace restaurant, this is around EUR 400 per service. A forecast does not automatically recover that full amount: it targets the part that can be avoided through better anticipation of attendance.
Sources: ADEME food-waste costs 2025, French Ministry of Agriculture.
What You Receive
Morning Forecast: an estimate of diner count around 9:30-10:30 on open days, to adjust late-morning preparation, prioritize replenishment, decide safety cooking, or call in reinforcement.
Week Planning: forecasts by day or by week for purchasing, menus, production, staffing, and reinforcements.
Multi-site Steering: standardized feeds, site-specific models, consolidated reporting, and comparison of biases or abnormal days.
Forecasts are delivered in the channel that fits operations: SMS, email, file, API, dashboard, or Teams/Slack. For Morning Forecast, open-day deliveries are included in the monthly fixed fee. Pricing is not calculated per message.
Getting Started
- Data Feasibility Diagnostic: remote data analysis, file-exchange design, first return-on-investment estimate, and test plan. Indicative price: EUR 4,500 excl. tax/site, with a EUR 2,500-6,000 range.
- Morning Forecast Test: historical test, comparison with your current forecast, 4-8 week parallel run, and error tracking. Indicative price: from EUR 10,000 excl. tax + EUR 900 excl. tax/month.
- Production: fixed-fee operation, error tracking, recalibration, and support. Morning Forecast starts at EUR 900 excl. tax/site/month; reference tiers: EUR 900 for 200-300 covers/day, EUR 1,200 for 300-500, EUR 1,500 for 500-700. Larger sites or heavy integrations: quoted separately. Week Planning starts at EUR 600 excl. tax/site/month.
Production is sold as a fixed monthly fee. Metrics are used to verify value and decide whether to move to production; they are not a performance-based billing mechanism.
Data
Two years of history give the best starting point. One year supports a cautious pilot. With no usable history, a start remains possible: scoping, reference forecast, structured collection of actual meals served, then progressive learning with adapted promise and pricing. The initial forecast without history is less reliable until actual meals served have been collected.
For a measured pilot, you need meals-served counts at daily grain over a usable period, the opening calendar, known exceptional days, and a contact who can describe available exports. After launch, meals served must remain available at daily grain; monthly or quarterly transmission is enough for fixed-fee production if the daily grain is preserved.
Presence signals such as badge counts, parking, WiFi, water, or occupancy can help, but they must be aggregated or anonymized. Identifiable data or small-volume aggregates require a separate privacy review.
Proof
The pilot compares your current forecast, simple baselines, and the candidate model. Measurement covers average error in covers, bias, frequent P80/P90 errors, and operational impacts: waste caused by volume errors, fallback meals, stockouts, substitutions, satisfaction, adjustable hours, and urgent orders.
The local prototype proves modeling know-how; it is not a performance guarantee for your site. Recent work on restaurant customer-flow forecasting confirms the relevance of historical data, calendar variables, weather, public holidays, and ML models for this type of problem.
Source: AIMS Press, 2025.
What Is Validated Before Production
- annual value that can be steered and measured by forecasting compared with the fixed fee;
- data quality and export stability;
- eligible days and excluded days;
- decisions that can actually be changed at 9:30, D-1, D-3, or weekly horizon;
- performance against current forecasting and simple baselines;
- IT constraints and specific integration needs.
Next Step
Test feasibility on your data.
An export of meals served, an opening calendar, known atypical days, and a description of available exports are enough to scope the diagnostic.