How AI helps an airport see why your plane is still at the gate

The aircraft is outside the window. Baggage vehicles are moving, a fuel truck is nearby and people are boarding. For a waiting passenger, the scene offers little explanation of when the flight will leave. Schiphol has spent years trying to make this busy interval legible to the people running it. A 2026 case-study abstract by its aircraft-turnaround process owner reports that its AI system, Deep Turnaround, is operating on 102 stands.[1]

Turning a moving scene into a progress report

The documented setup uses two cameras at each stand, sending snapshots to the cloud every five seconds. A slightly elevated view helps reveal activity around the aircraft, while masking excludes irrelevant areas from processing. Computer vision converts the images into events that staff can follow through a dashboard or receive inside their existing software.[2]

Schiphol’s technical case study describes two connected models. One detects activity in the camera images. A second combines that information with flight data to forecast how the turnaround will develop. Observation and prediction are separate steps: identifying that a process has started supplies evidence for estimating when the wider set of work might finish.[3]

The current product description says it detects more than 70 distinct events across 30 turnaround processes. That provides a more detailed picture than a single message saying the aircraft is running late. It can also support historical analysis of recurring delays. These are the system’s documented capabilities; the figures do not mean every possible event is visible or detected perfectly.[4]

Ready to leave is a different time from leaving

One of the most revealing details is a name. Schiphol calls the system’s predicted time PEGT, short for Predicted End of Ground Handling Time. Its explanation says this makes the meaning clearer: the forecast concerns when the turnaround will be ready, rather than when the aircraft will actually depart.[5]

The Dutch air navigation service’s October 2026 Schiphol guidance makes the distinction explicit. Start-up planning considers readiness alongside factors including local traffic, taxi time, any applicable take-off slot and separation requirements. An earlier forecast of completed baggage loading cannot independently remove those constraints. A plane may become ready sooner without receiving permission to move sooner.[6]

Why the last unfinished task matters

Consider this deliberately simplified example. Assume three remaining jobs can run in parallel and all other departure requirements are fixed. Their forecast finish times are shown below. The earliest possible completion of these three jobs is determined by the latest finish, rather than their average.

The group finishes at 08:50. Making cleaning five minutes faster leaves that time unchanged. Moving baggage completion to 08:45 brings it five minutes earlier. If an external restriction still prevents departure before 09:00, even that improvement need not change the actual departure. It does, however, change how much margin the ground operation has. This is why identifying the task controlling readiness is more useful than simply demanding that every team work faster.

The practical benefit is deciding what to do with that information. In Schiphol’s published interview with Menzies Aviation, staff describe using detailed status to prepare a pushback team when only a few bags remain. They also describe reassigning a crew after seeing boarding delayed during a storm. The intervention follows the observed work still outstanding, rather than a generic late-flight alert.[7]

The information has to reach the people who can act

SunExpress describes integrating Deep Turnaround events into its AviroConnect operational data store and a shared turnaround screen. Its headquarters and ground teams can then see activities such as refuelling and unloading. That matters because an observation buried in a separate dashboard may arrive too late to influence the next decision.[8]

The same interview identifies a less glamorous obstacle: different airports and handlers provide different subsets of events, sometimes entered manually after the work. The airline’s task is to combine camera-derived signals, mobile updates and airport data into a usable account of the turnaround. This is a deployment problem that remains even when an individual vision model works well.[8]

Gate planning is another use. Schiphol describes forecasts helping planners anticipate when stands become available and helping handlers allocate equipment and people. If an outbound aircraft is likely to occupy its stand longer, an incoming flight’s plan can be reconsidered earlier. The forecast provides additional time to make a coordinated decision.[2]

What the reported improvement does—and does not—establish

The April 2025 Menzies interview reports turnaround punctuality increasing from 67% in 2023 to 76% in 2024, alongside adoption of the predictions and changes to working procedures. That is a nine-percentage-point increase in the reported measure. The interview was published by Schiphol, which supplies the system. It provides no flight denominator, precise punctuality definition or controlled comparison, so it cannot isolate how much of the change AI caused.[7]

The useful distinction is between seeing activity, forecasting completion and improving an outcome. Each needs its own evidence. A camera event can be checked against what happened; a prediction can be compared with the eventual time; a claimed reduction in delay needs a comparison that accounts for changing traffic and operating conditions.

For the passenger watching through the window, the benefit begins with a better-informed operation: staff can see which work is unfinished and respond while there is still time. The parked aircraft becomes a set of visible, connected tasks. Whether that knowledge gets this particular flight away earlier depends on which task or constraint was holding it back.

Report an error or suggest a correction