Large foodservice operations depend on an extensive technology stack. Procurement platforms manage purchasing, inventory systems track ingredients and supplies, scheduling tools coordinate labor, and point-of-sale systems record transactions. Waste-management platforms add another layer by measuring discarded food. Together, these systems give operators a detailed view of costs, staffing, sales, and reported outcomes across their locations.
Physical activity creates another category of operational data. Food arrives, moves into storage, enters production, reaches a serving station, depletes throughout service, and eventually leaves as a purchase, return, or discard. Determining how much food moved through each stage and how much remained unused can still be difficult across hundreds or thousands of locations. Computer vision in foodservice can make that activity measurable and connect physical observations with existing enterprise data.
Compass Group reported $46.1 billion in fiscal 2025 revenue and estimates its addressable global foodservices market at approximately $360 billion. The company operates at a 7.2% underlying operating margin. Sodexo reported €24.1 billion in fiscal 2025 revenue across approximately 27,000 client sites, with an underlying operating margin of 4.7%.
These figures illustrate the size and complexity of global foodservice. Each location must coordinate food, labor, equipment, service levels, and consumer demand within a specific operating environment. Small variations in purchasing, production, replenishment, and waste can become meaningful when repeated across a large network. Physical visibility can help the systems operators already use reflect the way work unfolds onsite.
The value of that visibility extends across the foodservice ecosystem. Distributors can use physical demand signals to improve forecasting, delivery schedules, pack sizes, and SKU planning. Operators can connect production and service activity with inventory, labor, availability, and waste. Enterprises can better understand how attendance, menu selection, and consumption patterns affect the programs they fund across their locations.
Each foodservice platform records a defined set of events. Procurement systems show what a location ordered, while delivery and inventory records show what should have arrived and what should be available. Production systems capture recipes and planned quantities. Point-of-sale data shows what customers purchased, and waste tools document some of what was discarded.
Several physical events occur between those records:
Teams can capture some of this information through manual checks, scales, audits, and spreadsheets. The quality and consistency of those records may vary by location, shift, and employee workload. A blended weekly number can also hide the moment when performance began to change, a common challenge when comparing operations across multiple locations. An operator may see excess food usage or a high waste total without seeing which production or replenishment decision contributed to it.
Across a network, these gaps appear as operational variance: the difference between the plan recorded in software and the activity observed onsite. Connecting those views helps operators locate the workflow behind a reported outcome and compare it across locations.
Corporate dining provides a clear example of this challenge. Hybrid schedules have made workplace attendance more variable, with cafeterias serving a large population on one day and a much smaller group the next. Weather, internal events, holidays, team schedules, and office policies can also change demand from one meal period to another.
Kitchen teams still need to make production decisions before service begins. Preparing too little can lead to empty stations, limited selection, and slower service. Preparing too much increases food costs, labor requirements, and end-of-service surplus. The margin between those outcomes becomes harder to manage as attendance patterns change.
That tradeoff also shapes operator behavior. In many corporate dining programs, running out creates an immediate and highly visible service failure, while surplus may carry a less visible or less direct consequence. Teams therefore have a rational incentive to build a cushion into production. Better physical demand data can help operators size that buffer with greater confidence, protecting availability without systematically paying for excess capacity.
Historical sales and attendance forecasts provide useful planning signals. Physical data can add detail about how demand developed during service, including when a station began to deplete, when another required frequent replenishment, and which items remained after the meal period ended. Those observations give the next forecast a stronger operational baseline.
ReFED estimates that U.S. foodservice generated 12.5 million tons of surplus food in 2024, valued at approximately $157 billion. Plate waste accounted for 69.6% of the total, while overproduction represented another 11.9%. These industry-wide figures cover many types of foodservice operations, but they show how purchasing, preparation, portioning, menu selection, and consumer demand can affect the final waste number.
Waste data shows what remained unused and where the process ended. Physical Operations Intelligence can help explain how the operation arrived at that result by tracing physical activity and the decisions that preceded it. This context can turn a total weight or cost into a specific question about production, replenishment, or menu planning.
For example, a cafeteria may regularly discard a particular entrée after Friday lunch. A waste record captures the item and quantity, while serving-line activity, production records, and attendance data provide a fuller view:
The resulting analysis can support changes to batch size, replenishment timing, menu planning, or purchasing. Operators can then monitor the same workflow to determine whether the change improved the outcome and prevented the same operational cost leak from recurring.
Cameras already produce a visual record in many foodservice environments. Computer vision models can convert selected activity into structured data about products, quantities, movement, timing, and workflow events. That data can give operators a consistent way to observe the same process across different locations and service periods.
A deployment may monitor a receiving area, production zone, serving station, or return area. Models can be configured around specific operational questions:
These signals can be aggregated by location, station, meal period, menu item, or shift. Deployments can focus on food and workflow activity while excluding personal identity. The most useful implementation begins with a defined operating problem because a team studying overproduction needs different observations from a team studying service-line availability or receiving accuracy.
Plainsight refers to this connected view as Physical Operations Intelligence: a continuous data layer describing physical execution, connected to the enterprise systems that already manage the operation . It gives teams a consistent way to compare execution across locations while preserving the business context behind each event.
Physical observations gain value when operators connect them with the systems already in place. A rapid depletion event becomes more useful when matched with the menu, transaction volume, attendance, and production quantity. A repeated end-of-service surplus becomes easier to investigate when paired with ingredient cost, batch timing, day of week, and building occupancy.
Observations from one cafeteria can help explain a local event. When the same physical signals are measured consistently across many locations, operators can compare patterns in menus, production, demand, returns, and waste. Those comparisons can reveal recurring relationships that may be difficult to identify within a single location or system. Examples may include:
This context helps regional and corporate teams compare similar operations, investigate outliers, and give local managers a specific workflow to review. It can also help teams determine which practices should remain local and which improvements can scale across the network.
Foodservice companies have spent years improving the systems that manage purchasing, inventory, labor, transactions, facilities, and waste. Computer vision in foodservice can extend that technology stack into the physical operation, creating a clearer view of how food moved, when service conditions changed, and where execution began to vary.
Plainsight makes physical workflows observable and measurable across locations. Operators can connect activity in receiving, preparation, presentation, service, and returns with the enterprise data they already use. Those connections support more specific decisions around production, availability, staffing, purchasing, and waste.
A practical implementation can begin with one workflow and one operating question. Teams can define the physical events that would help answer it, compare those events with existing system data, and use the combined view to improve the next decision.
At enterprise scale, the value comes from repeating those improvements across locations and meal periods. A more accurate production decision, better-timed replenishment, or clearer demand signal may create a modest result at one cafeteria. Applied consistently across a large network, the same improvement can have a meaningful effect on food costs, labor utilization, availability, waste, and the customer experience.