Beyond the Kiosk: An AI Take on EES Congestion
I spent years working at Schiphol. I've built time-series forecasting models for airport operations, watched shift supervisors improvise under arrival surges, and seen what happens when passenger flow management works — and when it doesn't. So when the EU Entry‑Exit System came across my radar, I didn't approach it as a policy story. I approached it as a forecasting problem. And the more I looked, the more I realised that most airports are preparing for the wrong version of it.
This is no longer a hypothetical. EES entered its progressive rollout on 12 October 2025 and is scaling up in defined phases across Schengen borders. Full deployment — the point at which passport stamping ends entirely — is scheduled for 10 April 2026. We are weeks away. The queues are already forming. The question is whether airports will do something intelligent about it, or learn the hard way when summer peaks hit.
What EES Actually Changes at the Border
EES replaces passport stamping with a digital biometric registration process. For a traveller's first visit to the Schengen area, border officers must capture a facial image plus four fingerprints and enrol the person in the EES database. Subsequent visits are faster — a verification against the existing record.
In the long run, this is a genuine improvement in border intelligence. In the short to medium term it creates a structural bottleneck, and we can already quantify it. According to ACI EUROPE, border control processing times have increased by up to 70% during the progressive rollout — from roughly 45 seconds per passenger at baseline to around 76 seconds under EES. A 70% increase in service time translates to roughly a 41% drop in theoretical lane capacity. Apply that to the banked intercontinental arrival waves that large hub airports deal with daily, and queue mathematics turns exponential quickly.
These are not projections. IATA, ACI EUROPE and Airlines for Europe are jointly signalling waits of up to 2 hours at the current 35% registration threshold, rising to 4 hours or more at summer peaks if no structural intervention is made. Roughly 300 million border crossings per year fall under EES scope. This is not a niche operational adjustment.
Airports are responsible for passenger flow and terminal experience — yet the most critical resource in this equation, border officers, sits entirely outside their direct control.
That's the structural trap. Airports can't hire more border officers. They can't speed up biometric capture. What they can control is how intelligently they prepare the system around border control — the forecasting, the staffing signals, the passenger routing upstream. That's where the real leverage is. And it's the part of the conversation that has received the least serious attention.
The Forecasting Problem Nobody Is Talking About
Here's where my background in time-series ML becomes directly relevant, and why I think the industry is missing something important.
Most airport forecasting for border staffing is built around departures. Departures are tractable: passengers check in, drop bags, clear security — the airport controls much of the upstream timing and uncertainty is bounded. Arrival forecasting is a different discipline entirely. Weather, ATC restrictions, en-route diversions, and slot decisions all compound across a flight's journey. By the time an aircraft is on approach, the precision you'd want for staffing decisions simply isn't available from a schedule alone.
This is why arrival forecasting must be probabilistic, not point-based. A single expected arrival time is operationally misleading — it tells you the mean, but not the variance, and it's the variance that creates the 3-hour queue situations ACI EUROPE is already documenting. What a border authority needs is a probability distribution: what does the 90th-percentile arrival surge look like in the next two hours, and how much flexible capacity should be held in reserve to absorb it?
ML models handle this kind of uncertainty far better than human planners working from fixed schedules. Not because human judgment is worthless — experienced ops teams have pattern recognition no model fully replicates — but because probabilistic optimisation across dozens of simultaneous inbound flights is simply beyond what a shift supervisor can compute manually. The model doesn't replace the supervisor. It gives them a tool that makes the right call before the queue tells them they were late.
A Framework for Where to Intervene — and When
What I find most clarifying about EES congestion, viewed through a crowd-flow lens, is that solutions can be organised by when they act relative to the passenger reaching the kiosk. That ordering matters enormously: earlier interventions carry more leverage. Schiphol, with 161 kiosks and upstream kiosk clusters positioned well before its arrival filters, is operationally ahead of most peers — but even that investment is hardware. The forecasting and flow intelligence layer is still largely missing across the industry.
-
ML-driven arrival models feed border authorities a baseline staffing level and a flex pool sized by forecast uncertainty — communicated via a traffic-light model or 90th-percentile scenario. To maintain pre-EES service buffers, effective processing capacity needs to scale by approximately 1.7× — the same factor as the service time increase. Achieving that through staffing intelligence is far cheaper than hardware alone. This is the highest-leverage point, and currently the least invested.
-
The Frontex Travel to Europe app allows travellers to pre-register document data and facial image up to 72 hours before arrival. One important caveat: fingerprints cannot be pre-registered and remain mandatory at the border. Pre-registration reduces, but does not eliminate, the first-time processing burden. Industry bodies explicitly cite very low uptake as a structural problem. Driving adoption through airline check-in channels is one of the highest-ROI moves available. Simultaneously, smoothing departure peaks frees staff capacity that can be redirected toward arrival processing.
-
As updated delay data and gate assignments come in, operations teams can move staff to the right pier before queues form. This requires integrating arrival forecasts, gate systems, and real-time delay feeds into a single operational view — turning live data into deployment decisions automatically, rather than waiting for congestion to become visible on the floor.
-
When queues form, airports can offer passengers a scheduled return window instead of a physical queue — sensors estimate the wait, passengers take a slot, and they spend the interval in retail or F&B rather than a corridor. Brussels Airport's 61 kiosks and dedicated stewards are a step in this direction. The next layer — actively metering the flow rather than just absorbing it — converts congestion from an uncontrolled event into a managed one. This is the last line of defence, not the primary one.
A queue at border control is a symptom. The cause was set in motion hours earlier — in a staffing plan built on the wrong forecast, or a passenger who wasn't prompted to pre-register at check-in.
The Right Intervention Mix Depends on the Airport
There is no universal answer here. A large hub with banked intercontinental arrivals, a high share of first-time EES registrants, and a structurally tense relationship with its national border authority faces a fundamentally different problem than a regional airport with two international gates and a predictable schedule.
What I'm increasingly convinced of — from my forecasting background and from working through this problem carefully — is that selecting the right intervention mix requires an operational diagnosis that most airports haven't yet done rigorously. Which intervention point gives the most leverage for this airport's specific passenger profile? Where does forecast uncertainty actually sit? What is realistic pre-registration uptake given the airline mix? How much of the observed congestion is service time versus staffing variability versus kiosk downtime?
These are answerable questions. But they require someone to ask them systematically — ideally now, not after the first summer peak in June reveals the gaps.
The Industry Needs to Decide What This Is
Here's the provocation I want to leave you with.
EES is being treated, in most airports I'm aware of, as an infrastructure project. Add kiosks — Schiphol has 161, Brussels has 61. Build lanes. Buy sensors. The assumption is that the bottleneck is physical capacity, and that more hardware solves it.
I think that assumption is dangerously incomplete. Airports that have invested heavily in kiosks are still reporting 2- and 3-hour queues. Because hardware without intelligence doesn't solve a forecasting and flow management problem. It gives you expensive equipment standing idle during off-peaks and overwhelmed during surges — because the surges are still unpredicted.
The airports that handle EES well won't necessarily be the ones with the most kiosks. They'll be the ones that built the forecasting and operational intelligence layer first, and let infrastructure decisions follow from that analysis. That requires different expertise, a different kind of partner, and a different framing of the problem entirely.
Full deployment is 10 April 2026. Summer is 10 weeks after that. The window to get this right has nearly closed.
TimeSlotsPro is a consultancy specialising in time-series forecasting and intelligent flow management for airports. We help operations teams diagnose where EES congestion will actually hit and design the right intervention approach for their specific context — whether that's probabilistic staffing models, pre-registration strategy, real-time redeployment systems, or timeslot-based queue management.
Working on EES readiness?
We're talking to airports and industry bodies who want to think through this carefully — before the summer tells you where the gaps are. If you're involved in shaping how Europe's airports respond to EES, we'd like to be part of that conversation.
