Brussels: predicting the pinch points in a working airport
Brussels Airport's operations plan looks across passenger and aircraft processes before problems reach the terminal. Machine-learning forecasting and baggage-reclaim estimates offer more specific tests of whether information changes an operational decision.
Aerial view of Brussels Airport. Photograph: Abxbay / CC BY-SA 4.0 (cropped to fit). Shows the airport, not an AI implementation.
The picture at Brussels Airport
An airport plan that reaches beyond the flight schedule
Security lanes, baggage belts and passenger assistance all compete for limited resources. Brussels Airport has been developing a shared Airport Operations Plan to bring forecasts and real-time conditions together, rather than leaving every team to react to its own queue.
The most persuasive part of the story is the question it raises about action. Forecasting passenger volumes matters when it changes staffing or how pressure is handled on the day. A separate baggage-reclaim prediction is aimed at passengers waiting for luggage, not the airport operations room.
The available sources describe years of work and several forecasting models. They do not demonstrate that all five proposed machine-learning modules entered production or that their combined effect has been independently measured.
Selected implementations
From the airport plan to the baggage belt
01 / Deployed
A forecast shared across the operation
The operational problem
Passenger volumes and ground processes change throughout the day. A security forecast becomes more useful when it can be interpreted alongside other airport constraints.
What the sources report
Brussels Airport describes an Airport Operations Plan covering passenger movement, baggage, aircraft handling and airport access. Its chief operations officer told International Airport Review in 2025 that AI-based security-volume forecasting had been used since 2017, with predictions in intervals as short as five minutes and later upgrades for real-time adjustment.
Why it matters
The airport is trying to make operational planning less fragmented. The significant test is whether the forecasts lead to better decisions on lane opening, staffing or passenger assistance.
What remains unproven
The reviewed sources do not provide attributable reductions in queues, operating costs or disruption from the shared planning programme.
After landing, passengers do not know whether to head straight for baggage reclaim or use the time for something else.
What the sources report
Superlinear says it developed the bTag machine-learning model with Brussels Airport, using flight origin, arrival time and other variables to estimate baggage delivery. The prediction can be sent to the passenger together with walking time and belt information.
Why it matters
This is a modest but tangible use of prediction: a specific estimate delivered to a traveller who can act on it. Superlinear reports prediction and passenger-response figures, but they are supplier-reported and historical.
What remains unproven
Brussels Airport has not independently confirmed current bTag usage or published a comparable current accuracy measure.
Evidence supports a reported claim; it does not automatically establish a measured operating benefit.
The wider register
What else is happening at Brussels Airport?
Other tracked work includes Superlinear's wider forecasting modules, a camera-based turnaround timestamping report and an autonomous cargo-tractor trial. The vendor said some forecasting modules were not yet in production; the camera report lacks evidence of scope; and the cargo vehicle remains a supervised test.
Autonomous electric cargo-trailer tow tractor trial at Brussels Airport · Pilot
Brussels Airport is trialling an autonomous electric tow tractor that pulls cargo trailers between warehouses and aprons in its cargo zone, the airport announced on 24 August 2026. The vehicle comes from Charlatte Autonom, a partnership between Charlatte Manutention and Navya Mobility, and the trial runs with cargo handler WFS.
Satellite navigation and laser sensors let the tractor perceive its surroundings and spot obstacles, and Navya's driving software keeps it on its trajectory, the airport says. In this phase it runs on four predefined routes at up to 12 km/h in autonomous mode, can tow up to four trailers and operates off-peak, with a trained operator always on board.
Brussels calls it the first real-life test of autonomous cargo transport at a Belgian airport, following a self-driving staff shuttle trial in 2024. The work sits in the EU-backed Stargate programme, which Brussels leads with 22 partners, and no results or decision on wider use have been published.
International Airport Review reported in November 2025 that Brussels Airport's other projects included the use of cameras to timestamp aircraft turnaround processes.
Superlinear, a Belgian data-science firm, says it won the public tender for data science on Brussels Airport's AOP 2.0 and has been building machine-learning forecasts for five use cases: check-in, security screening, passengers with reduced mobility, baggage and border control. The aim is to tell operations staff where and when bottlenecks will form so they can act in advance.
Superlinear reports a mean absolute percentage error of about 30% per 30-minute interval for its check-in forecast, 14% at day level for the number of passengers needing assistance, and about 28% per flight for bag counts. These are the vendor's figures, and the baggage work was described as still ongoing.
The case study says work on the use cases began in May 2022 and that the models "will soon be put into production". Superlinear now marks the page as reflecting earlier work, and Brussels Airport has not confirmed which of the five modules are live.
Evidence supports a reported claim; it does not automatically establish a measured operating benefit.
Useful intelligence has to change a decision
Brussels presents a practical contrast. Airport planners need forecasts across connected processes; arriving passengers want a credible answer about a single bag.
Both depend on accuracy and timeliness. Those measures, rather than the number of models or sensors installed, should shape the next assessment of results.
Live data from the Aerostack register. Status reflects published evidence only. See Our approach for how records are graded.
Aerostack register · 5 tracked deployments
Aerostack tracks AI deployments across airports and airlines, with the published evidence behind each record. Photography is real aviation imagery. Record thumbnails may show the named airport or airline; other photographs do not.
