AI use case

AI meal and catering demand forecasting for flights

Airline catering demand forecasting is machine learning that predicts how many passengers will board each flight, and in some systems which meal choices they will take, so the caterer loads fewer surplus meals without running short. It plans the quantity loaded per flight; it does not design menus or time the catering truck.

Last verified 18 November 2026 · 2 public deployments

63%
Reported food waste reduction
KLM Royal Dutch Airlines, organization claim.
USD 384,000 to USD 5.4 million
Indicative value per year
An airline operating 50,000 catered flights a year. Worked example, see how it is calculated.

What problem does it solve?

Airlines cater for the bookings they hold, but not every booked passenger boards. KLM puts the share of booked passengers who do not show up on time at between 3 and 5% depending on the class, so a meal count based on bookings loads food that nobody eats. Loading too little has a cost as well: passengers find the choice they wanted is gone. Flight delays, cancellations and late bookings change the count after the order is placed.

Caterers have traditionally set quantities from experience, historical averages and rough estimates. Forecasting models replace that guesswork with a per flight prediction that updates until close to departure. The two deployments on this page measure the effect on food waste: KLM against catering for every booked passenger, and Santan, the caterer for AirAsia, over a year.

How does it work?

  1. Collect history. Flight level loading, sales and uplift data, bookings and check in counts are joined with route, season, departure time and the passenger mix.
  2. Forecast per flight. A model predicts how many passengers will board, by cabin class, and in some systems the demand per meal choice. The forecast starts weeks before departure and is refreshed as bookings, check in and delays change.
  3. Set the load. The forecast feeds the meal ordering and loading system, which sets the quantity per cabin and holds back popular items for the flights that need them.
  4. Learn from the flight. Meals returned or discarded after each flight are compared with the forecast so the model improves.
Audience
Back office
Autonomy
Assist
Adoption
Emerging
Channels
API and system to system, Internal tools

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

Value benchmarks for AI meal and catering demand forecasting for flights
KPIMedianReported rangeData pointsClaimed by
Food waste reductionToo few to pool
20% to 63%
22 organization

Value drivers: Lower cost to serve, Speed and cycle time.

Indicative value

An airline operating 50,000 catered flights a year

USD 384,000 to USD 5.4 million

Annual meal cost avoided per year

How this is calculated

Formula: flights * mealsPerFlight * wasteShare * wasteReduction * costPerMeal. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Catered flights per year flights, flights50,00050,000The reference airline.
Meals loaded per flight mealsPerFlight, meals120180Editorial assumption, replace with your own average load.
Share of loaded meals that go to waste today wasteShare, fraction of meals0.080.15Editorial assumption, replace with your own waste audit.
Reduction in wasted meals wasteReduction, fraction of waste0.20.4Conservative against the evidence on this page (Santan reports 20% over a year; KLM reports 63% against catering for every booked passenger).
Cost of one meal loaded costPerMeal, USD per meal410Editorial assumption, replace with your own food, packaging and handling cost.

What it leaves out: Food and handling cost only. It leaves out the cost of the forecasting build, the cost of any shortage that the forecast prevents or causes, fuel burned by carrying less weight, and disposal and sustainability reporting benefits.

Which organizations already do this?

2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.

This library is not a list of Blits.ai customers. It documents public AI deployments from any vendor, each with its source.

KLM Royal Dutch Airlines

Netherlands · Travel and hospitality · 2024

ProductionGrade B

KLM uses an AI model called TRAYS to predict how many booked passengers will actually board each flight, with separate forecasts for Business, Premium Comfort and Economy. The Meals On Board System takes these forecasts, updated from 17 days before departure until 20 minutes before the flight leaves, to set purchasing and loading. KLM reports the result from a three month analysis against catering for every booked passenger.

  • Food waste reduction: 63%, three month analysis
    "A three-month analysis shows that 63% less food is wasted compared to catering for every booked passenger."
    Claimed by: organization

Santan

Malaysia · Travel and hospitality · 2025

PilotGrade C

Santan, the inflight caterer for AirAsia, built an AI demand planning tool in house, trained on more than three years of flight level sales and loading data plus forward looking pre booking forecasts. It recalculates expected demand after delays, cancellations or late bookings so loading volumes and popular items can be adjusted. The chief executive says it was live on flights and being prepared for rollout across the fleet.

  • Food waste reduction: 20%, over the past year
    "Inflight food waste has dropped by 20% over the past year"
    Claimed by: organization

How do you implement it?

A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.

Data you need

  • Per flight history of meals loaded, consumed and returned by cabin and meal choice
  • Booking, check in and no show history by flight, route and season
  • Flight schedule changes, delays and cancellations

Systems to integrate

  • Meal ordering and loading system at the caterer
  • Airline reservation and departure control systems
  • Flight operations feed for delays and cancellations

Complexity: Medium

The build needs booking and check in data per flight, the caterer's loading history, a connection to the ordering and loading system that sets quantities, a data sharing agreement between the airline and the caterer, and an agreement on who carries the risk of a shortage.

  1. 1

    Measure waste per flight first

    Weigh or count returned meals for a few weeks on selected routes, so the baseline is what leaves the aircraft and not what the order sheet says.

  2. 2

    Forecast passengers before meals

    Start with the number of passengers who board, by cabin, as KLM does, then add meal choice demand once the passenger forecast holds. Compare it with the current rule such as catering for every booked passenger.

  3. 3

    Run in shadow, then on selected routes

    Show the forecast beside the current order for a period, then load by the forecast on routes where a shortage is cheap to fix, for example short flights with a buy on board option.

  4. 4

    Update until departure

    Refresh the forecast as check in and delays arrive, and define the last moment at which the loading plan can still change.

Guardrails

  • A minimum meal count per cabin and a buffer for premium cabins and special meals
  • Safety critical items such as special meal requests are never cut by the model
  • A named owner who can override the forecast for disruption days

KPIs to instrument

  • Meals returned or discarded per flight, by cabin and route
  • Share of flights with a reported shortage of a meal choice
  • Forecast error for boarded passengers by cabin
  • Meal cost per passenger

Human in the loop

Catering planners own the loading rules and review the forecast on irregular days such as strikes, diversions and large group bookings. Cabin crew report shortages so the model and the buffers can be corrected.

Common failure modes

Shortages on premium cabins
A model that cuts the load too far leaves business class without a choice. Keep floors for premium cabins and track shortages by cabin.
Forecast trained on a different operation
History from before a schedule change or from a different route mix misleads the model. Retrain after network changes and watch error by route.
Waste measured at the wrong point
Counting what the caterer unloads, not what is thrown away across the supply chain, misses waste before loading. Define the measurement boundary up front.

What are the risks and rules?

EU AI Act

Minimal risk

A back office forecast of passenger numbers and meal quantities does not fall under an Annex III category and does not interact with people, so the AI Act sets no specific obligations beyond AI literacy for the staff who use it.

Rules that apply

Controls to put in place

  • Forecast inputs limited to counts and routes, without passenger identities
  • Logged forecast, order and actual quantities per flight for audit
  • Retraining and error review after every schedule change

Frequently asked questions

How much food waste can AI forecasting remove on flights?

KLM reports 63% less food wasted over a three month analysis, measured against catering for every booked passenger. Santan, the caterer for AirAsia, reports a 20% drop in inflight food waste over a year. The baselines differ, so compare a new program with your own current method.

What does the model actually predict?

At KLM it predicts how many booked passengers will board, per cabin class, from historical data, starting 17 days before departure and updating until 20 minutes before the flight leaves. Santan's tool also uses meal time, route, passenger mix and festivals to predict which items will be in demand.

Does AI catering forecasting risk running out of meals?

It can if the load is cut without floors. Santan describes both overstocking and understocking before its tool, and reports a fall in both after it. Keep minimum counts for premium cabins and special meals, and track reported shortages next to waste.

How to cite this page

Blits.ai AI Use Case Library, "AI meal and catering demand forecasting for flights", last verified 18 November 2026, https://www.blits.ai/ai-use-cases/airline-catering-demand-forecasting. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 18 November 2026: First published

Related use cases

Travel and hospitality

AI demand forecasting for restaurant labor and prep scheduling

Restaurant demand and labor forecasting is a machine learning model that predicts sales and guest traffic per restaurant, per daypart and often per menu item, from point of sale history, local events and weather, so restaurant managers can build staff schedules and prep lists against a forecast instead of a manual guess or last week's numbers. In retail stores, where the AI builds the staff schedule itself from foot traffic, labor rules and employee preferences, the job is AI workforce scheduling for stores.

Deployments
5 public, best grade B
Reported forecast accuracy
83%
McDonald's Georgia, vendor claim
Travel and hospitality

AI for aircraft turnaround and ground handling optimization

AI ground handling and turnaround optimization watches each aircraft turnaround in real time, often with computer vision, and alerts ground operations staff when a step falls behind. It uses cameras aimed at the gate and apron, covers fueling, catering, baggage, boarding and pushback, predicts the departure time as soon as the aircraft arrives, and raises the alert the moment a subprocess falls behind schedule, so staff can intervene before a small delay becomes a missed slot.

Deployments
4 public, best grade C
Autonomy
Assist
Travel and hospitality

AI decision support for airline operations control and disruption recovery

Airline operations control decision support is AI that watches an airline's day of operation and predicts where weather, delays, crew limits or technical problems will break the plan. It proposes recovery options across aircraft, crew and passengers, such as retiming flights, swapping aircraft or holding a connection, with the cost and passenger impact of each, for controllers to approve.

Deployments
4 public, best grade B
Autonomy
Copilot
Travel and hospitality

AI for airline fuel efficiency optimization

Airline fuel efficiency optimization is an AI system that analyzes flight data to coach pilots and flight operations engineers on fuel performance against a baseline. It analyzes fuel burn, routing, taxi procedure and auxiliary power unit use fleet wide, flags the highest value savings opportunities by route and procedure, and tracks the savings over time.

Deployments
4 public, best grade C
Autonomy
Assist