We built a computer vision system for a resort group's buffets that tracks tray levels and live demand, then tells kitchens when to cook more or hold back. Food waste fell 15%, and margins rose 11%.
Industry: HospitalityService: AI & AutomationUpdated
Overview
The challenge
Kitchens cooked from fixed prep sheets based on past occupancy. What they produced had little link to how fast guests were actually eating, and buffets wasted 18-22% of their food as a result.
Expensive proteins were over-produced with no way to trace the waste back to a prep cycle. New sustainability rules also called for accurate waste reporting that the group could not yet provide.
Swings in consumption at peak dining times meant either heavy waste or costly last-minute prep.
Over-producing valuable proteins hit profit, and no data linked waste to specific prep cycles.
Without accurate waste weights, the group was slow to meet new mandatory sustainability disclosure rules.
The solution
Privacy-gated cameras use computer vision to track serving-tray fill levels and waste bin weights as service happens.
AI compares how fast food is being eaten with live POS data and event schedules, then sends Stall or Fire instructions to the kitchen.
Every production change is saved with a reason code, such as Low Tray Speed, giving a fully traceable record of waste reduction.
Before and after
Before
Manual checks
Kitchen staff checked the buffet lines now and then, so they reacted late.
After
Live monitoring
Each tray's fill level is tracked continuously, with automatic alerts when production should slow down.
Before
Fixed schedules
Prep lists were set 24 hours ahead from guest headcounts alone.
After
Live feedback
Prep volumes adjust automatically to live consumption and guest arrival rates.
Before
End-of-day estimates
Staff weighed waste buckets by hand, which was error-prone and gave no item-level detail.
After
Traceable ledger
Vision sensors track waste item by item and feed automated regulatory reports.
Key features
Video is processed at the edge to measure how fast trays empty, without overloading the cloud connection.
Business impact
Zero response lag
Guest faces are masked at the camera itself, in line with GDPR and SOC 2 privacy expectations for hospitality.
Business impact
100% privacy aligned
Fire tasks put high-demand dishes first and automatically hold back prep for slow-moving buffet stations.
Business impact
15% waste reduction
Results
15% Waste Reduction
Prep now follows what guests are actually eating, which cuts the surplus left at the end of service.
Before: High WasteAfter: 15% Lower
11% Margin Lift
Better-timed protein prep and automated yield tracking improved gross profit margins.
Before: CongestedAfter: Optimised
400ms Edge Inference
Edge processing gets Fire alerts to kitchen displays in under 400ms.
Before: < 1sAfter: 400ms
Faster delivery
Reusing tested components saved about 8 weeks of setup work, and we spent that time on the demand model and the kitchen alerts.
Our tested vision module gave us camera capture and on-device face masking, ready for the buffet lines.
A reusable forecasting module combined tray depletion, POS data and event schedules into each Stall or Fire decision.
An existing monitoring platform tracked cameras, edge devices and alert times across outlets.
Pre-built cost controls tracked compute spend as the system rolled out across properties.
Trust and control
Processing happens on the device, so no guest facial data leaves the local network.
PRIVACY-GATEDAudit-ready waste logs meet mandatory ESG and local municipal reporting requirements.
VERIFIABLE DATACloud-managed edge clusters let the system scale reliably across 100+ properties worldwide.
READY TO GROWThe resort group owns 100% of the vision models and the demand prediction system.
100% OWNEDEdge cameras track how full each serving tray is and how fast it empties. AI compares that with live POS data and event schedules, then tells the kitchen to Fire more of a dish or Stall its prep. Food waste fell by 15% because cooking now follows real demand.
Food waste fell by 15% and margins rose by 11%, largely through better-timed protein prep. Fire alerts reach kitchen displays in under 400ms. Item-level waste logs with reason codes also gave the group audit-ready data for mandatory sustainability and municipal reporting.
Guest faces are masked on the camera itself, so no facial data leaves the local network. The system only needs serving-tray fill levels and waste bin weights, and all video processing happens at the edge. This keeps the design in line with GDPR privacy expectations for hospitality.
If your kitchens still cook from fixed prep sheets, we can help you match production to what guests are eating right now.