Inflight catering demand forecasting that orders meals for the passengers who board, not the ones booked
Meals ordered for the passengers who board, not the ones booked, and no cabin left short.
See one case, screen by screen ↓Inflight catering demand forecasting predicts how many meals each flight and cabin needs — and which dishes — from bookings, no-shows, upgrades, missed connections and pre-orders, so the caterer produces for the passengers who board, not the ones booked. The result is a meals per flight forecast by cabin and dish. Catering Demand Forecasting adds seven agents that forecast every cabin, propose the final order before each caterer cut-off with its reasons and sources, and grade their own accuracy after each flight. The catering planner approves every change.
How to choose AI for airlines: a buyer's guide to agentic AI solutions →
Updated
Inflight catering demand forecasting at most airlines still means a catering uplift of booked passengers plus a fixed buffer. The catering planner sends a standing order a day out and final catering numbers three hours before a widebody leaves, about 90 minutes before a narrowbody, for 140 departures a day. There is no time to check no-show history, upgrades waiting to clear or the inbound flight that is running 85 minutes late, and the dish split is a fixed percentage from the menu planners that rarely changes.
So the caterer produces for seats that fly empty, and the meal a passenger wants is not the one loaded. Untouched trays come back and, on international flights, are destroyed, not donated. Business passengers miss their first choice when a new menu sells better than planned. Caterer acknowledgments get lost in email, return-load reports are never compared with the order, and no one measures forecast accuracy, so the buffer rule never changes.
Estimated split of a planner's working hours at a network hub, by hand and with the solution.
Seven agents forecast every cabin and dish and propose the final order; the catering planner approves every change.
Thursday, October 15, 2:32 p.m. at Chicago O'Hare. Corvane Airways CV 118 to Frankfurt, a 787-9 with 48 business seats, leaves at 17:55. The caterer cut-off is 14:55, in 23 minutes. Here is what happens next, screen by screen, in the working solution.
Renee's day starts with one line: CV 118 to Frankfurt, the agents propose 40 business meals, not 46. Below it, six decisions wait by cut-off — a caterer counter-offer on CV 1422, CV 118, a low-confidence Denver flight and three more — and what the agents found since 06:00, from a late Boston inbound to a short rib that ran out yesterday. 142 catered departures today; 203 meals not made so far, $4,742.
“CV 902 from Boston will misconnect 10 passengers · 85 minutes late; 10 connecting passengers rebooked off CV 118, so their meals come out of the forecast.”
CV 118 opens as a forecast cone for business meals: booked rising from 14 days out to 41 of 48, the forecast narrowing to 37 (range 35–39), and the standing caterer order of 46 as its own line. The proposed 40 sits dashed at the cut-off mark. Beside the chart, five agents explain the flight in a conversation Renee can join.
CV 902 from Boston is now due at 17:10, 85 minutes late. Four business passengers connecting to CV 118 would have 45 minutes against the 75-minute minimum connection time, and they are already rebooked on CV 126 at 21:40, so the agent removes them. Renee clicks the citation and the flight status message opens with the delay highlighted.
Limit: the Connection Watcher never removes a passenger who is not rebooked.
41 booked, minus 4 rebooked off the late CV 902, minus 0.8 expected no-shows (2.1% in October), plus 0.7 of 3 waitlisted upgrades expected to clear: 36.9, so 37. The spare policy adds one meal per choice, 3 in business, for an order of 40. Every line on the flight detail cites its source, and the passenger movements show who the meals follow.
“Special meals: loaded by name, never forecast.”
The Choice Mix Forecaster expects 61% to choose the braised short rib, 27% the halibut and 12% the risotto; nine pre-ordered. The Order Planner proposes in one sentence: cut business from 46 to 40, loaded as 24 short rib, 11 halibut and 5 risotto, and economy from 214 to 204. The Forecast Auditor re-derives the forecast from the raw booking and flight-status data only, gets 36.9 → 37 and agrees, confidence 92%.
“Your decision, the reason and the agents' sources go on the audit trail. The Caterer Liaison sends only what you approve.”
Renee approves 40 with the reason “Misconnects confirmed”. The Caterer Liaison sends the change to Northmere Flight Kitchen as NFK-118-1015-3, and at 14:36 Northmere acknowledges 40 business (24 short rib, 11 halibut, 5 risotto), 24 premium economy and 204 economy. The cone's order line drops to 40 and the flight moves to sent and acknowledged.
PM 412 to Denver is an acquired Pellmar flight, and the Pellmar booking feed has not updated since 11:30. The Load Forecaster's range is wide, 9–12, at 54% confidence. The Forecast Auditor works from departure control instead — 7 of 11 checked in — expects 11 to board, not 10, and disagrees with a cut. The Order Planner keeps the order at 12 and hands the flight to Renee: keep 12, set a number or ask the station.
“I will not propose a cut on data this old. The order stays at 12 unless you set a number.”
An approved cut on CV 1422 to San Francisco reached Northmere after hot-meal production started at 14:10. Luis Romero replies that they can reduce 2 first-class meals at a 50% charge under §4.4, $16.00 each. The Caterer Liaison works out the net saving, $32, and recommends accepting; Renee decides before 14:40.
Limit: the Caterer Liaison never accepts a caterer's charge itself.
The reviewer read each caterer's return-load report and the crew's notes: 131 of 138 flights measured, 3.7% of meals returned untouched (2,184 of 58,460 uplifted), 2 run-outs. The miss that matters is CV 86 to London, where the new autumn short rib ran out at row 8 — chosen 68%, forecast 52%. The reviewer proposed raising its share; Owen Mercer approved at 09:12, and today's CV 86 proposal uses it.
Accuracy is a screen anyone can open: 96.8% of cabins within 2 meals, an average error of 1.4 meals against 4.9 for the old buffer rule, business run-outs 0.9%, untouched returns 3.8% against 7.1% before. Each big miss is explained, and the agents propose rule R-12 — use a coach group's own show-up history — for Owen Mercer to approve or reject.
“CV 140 HND · Oct 11 · premium · 3 fewer boarded than forecast: misconnecting passengers were not yet rebooked at the cut-off. Correct by design: never removed until rebooked.”
The same catering day, seen by five of the people who carry it — what it looked like on booked plus a buffer, and what it looks like now.
Seven agents forecast every cabin and dish and propose the final order; the catering planner approves every change.
Forecasts how many passengers will board each cabin from booking snapshots, a passenger no-show forecast built on each cabin's history, the upgrade waitlist, standby and departure control, with a range and a confidence.
Watches inbound flights, predicts misconnects against the minimum connection time at the hub and reads the rebooking record.
Forecasts dish shares from pre-orders, route and menu history and crew run-out notes, and learns new menus from similar dishes.
Proposes the final order by cabin and dish before each caterer cut-off, with the spare policy, the cost and a one-sentence reason.
Re-derives every forecast from the raw booking and flight-status data, never from the other agents' working, agrees or disagrees in writing, and scores accuracy after each flight.
Sends approved changes to the caterer with a reference, tracks each acknowledgment, chases, and reads replies and counter-offers against the caterer agreement.
Reads return-load reports and crew notes after each flight, measures untouched returns and run-outs against the forecast, explains the misses and proposes rule changes.
Approve every order change and every rule. The agents propose; a named person decides.
Ask about any flight, cabin, caterer or forecast — or give an instruction. Answers cite their sources, and an instruction becomes a rule someone approves.
Why 40 business meals on CV 118?
Four business passengers on the late CV 902 are rebooked to CV 126. October no-shows on that cabin run 2.1%, and 0.7 of 3 waitlisted upgrades are expected to clear, so 37 will board; the spare policy adds 3. That cuts 46 to 40 and saves $612 in business.
Which flights could run out of a dish tonight?
One: CV 86 to London. The short rib ran out there yesterday, so the agents propose 58 business meals instead of 54, and the increase waits for Owen Mercer.
How accurate were you on Frankfurt?
On ORD–Frankfurt business, 97.4% of forecasts were within 2 meals over 30 days, with an average error of 1.2 meals against 4.6 for the old buffer rule.
Keep one extra halibut on Frankfurt flights
Done as a draft: rule R-14, “Frankfurt business: one extra halibut above the spare policy”. Rule changes need Owen Mercer, so it is waiting in Settings and no order changes until he approves.
13 screens from the working solution, on its sample data. Pick one to see it large.
The rules every proposal is built on are settings, numbered and versioned. An edit becomes a pending change, and the inflight services manager approves it before the agents use it.
| Setting | Default | Choose from |
|---|---|---|
| Spare meals, business (R-03) | 3 meals | One per main-course choice; edits go to the inflight services manager |
| Spare meals, economy (R-03) | 5% | Share of the forecast, rounded up |
| Smallest change sent to a caterer (R-04) | 2 meals | Per cabin |
| Planner approval limit (R-05) | $1,000 or 5% increase | Above this, the inflight services manager approves |
| Confidence floor (R-06) | 70% | Below this, the agents do not propose a cut |
| Booking data freshness (R-08) | 60 minutes | Older than this widens the range and lowers confidence |
| Caterer cut-offs, ORD | T-3 h widebody · T-90 min narrowbody | Per caterer contract and station; FRA, HND, NRT, ICN and GRU at T-4 h |
| Notifications | Cut-off in 30 and 15 minutes | Caterer not acknowledged in 10 minutes · accuracy under 94% for 7 days · return report missing after 24 hours · daily digest |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (IATA cabin waste: 0.94 kg per passenger from 25 audits, 3.6 million tonnes in 2023 and doubling by 2040, and 23% of cabin waste untouched food and drink in its Heathrow audit). Corvane Airways, Pellmar Air, the caterers, flights and the people named on this page are fictional — characters and sample data in the working solution.
Inflight catering demand forecasting predicts how many meals each flight and cabin needs — and which dishes — from bookings, no-shows, upgrades, missed connections and pre-orders, so the caterer produces for the passengers who board, not the ones booked. The result is a meals per flight forecast by cabin and dish. Catering Demand Forecasting adds seven agents that forecast every cabin, propose the final order before each caterer cut-off with its reasons and sources, and grade their own accuracy after each flight. The catering planner approves every change.
Most still send booked passengers plus a fixed buffer. Here the Load Forecaster starts from the booking snapshot, removes passengers already rebooked off late inbounds, subtracts expected no-shows from the cabin's history and adds upgrades expected to clear; the Order Planner then adds the spare policy. On the sample Frankfurt flight that is 41 booked, 37 expected to board and an order of 40, not 46.
The Connection Watcher compares each connecting passenger's time with the minimum connection time at the hub and reads the rebooking record. Passengers who will misconnect and are already rebooked come out of the forecast; those not yet rebooked stay in and are flagged as at risk. No-shows are forecast from the cabin's own history, such as 2.1% in October.
Inflight food waste reduction starts with ordering for expected boarding instead of booked plus a buffer, forecasting the dish mix, and measuring what comes back. After each flight the Waste & Short-Load Reviewer reads the caterer's return-load report and crew notes, measures untouched returns and run-outs against the forecast and proposes rule changes. In IATA's Heathrow cabin waste audit, 23% of cabin waste was untouched food and drink; the design goal is a third less of it.
The accuracy screen shows this airline meal forecasting openly, misses included. On the sample data, 96.8% of cabins were forecast within 2 meals over 30 days, with an average error of 1.4 meals against 4.9 for the old buffer rule, and business run-outs fell to 0.9% from 3.1%. Each big miss is explained by route and cabin.
The agents will not cut. Booking data older than 60 minutes widens the range and lowers confidence; under 70% confidence no cut is proposed. The Forecast Auditor re-derives every forecast from raw data, and if it disagrees the flight goes to the planner to keep the order, set a number or ask the station.
Yes. The agents propose and never send on their own: the catering planner approves, changes or rejects every order change with a reason, the inflight services manager approves changes over $1,000 or a 5% increase and every rule change, and the Caterer Liaison sends only what was approved and never accepts a caterer's charge itself.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your caterer agreements and price lists, spare policy, approval limits and past return-load reports — and it goes live once every quality gate has passed. In the working solution, newly added flights first run a two-week shadow forecast compared with actual loads: the agents forecast but never propose a cut until every setup step is done.
Every caterer line checked against what was loaded, and every missed penalty claimed.
1-min demoGalley PlanningEvery galley loaded to spec: no cabin short of a meal, no weight flown for nothing.
1-min demoOnboard Retail ManagementEvery flight's sales reconciled, and every sell-out turned into the next load.
1-min demoWe'll run Catering Demand Forecasting on a sample of your own departures, bookings and return-load reports.