AI in Airport Retail: Why More Passengers Aren't Buying, and How Flight-Aware AI Agents Fix It
Airports will handle a record 10.2 billion passengers in 2026, yet a smaller share of them shop. AI in airport retail closes that gap, because airport demand runs on the flight schedule, not the calendar. Here's how travel retailers and airport commercial teams use that to win back penetration, conversion and spend per passenger.
AI in airport retail uses flight schedules, real-time flight status, security queue times, point-of-sale and inventory data to forecast demand for each store by the hour. It then replenishes stock, rosters staff and adapts offers to the passengers actually departing. Agentic AI carries out those steps automatically, with people approving price and staffing changes. It closes the gap between growing passenger numbers and falling shop penetration.
Key takeaways
- The passenger paradox is real. Traffic is at record levels, but a shrinking share of passengers walks into a store or buys.[2]
- Airport demand follows flights, not the calendar. Forecasting tools built for high-street retail miss delays, gate changes and shifts in which routes are departing.
- Time is the currency. A 10% rise in dwell time is associated with about 6% more retail revenue and 8% more F&B revenue in most terminal designs.[4]
- Data coordination is the blocker. Airport, airline and retailer data still sit in separate systems. Fix that first, then add agents.[5]
Why are more airport passengers buying less?
On paper, airport retail should be booming. ACI World expects 10.2 billion passengers in 2026 and 18.8 billion by 2045.[1] One level down, the picture flips: m1nd-set's 2026 research shows fewer passengers entering stores and fewer visitors buying,[2] while non-aeronautical revenue per passenger is still well below 2019.[3]
Some of this is economic: spend per passenger is about 15% below pre-pandemic levels as travellers grow more price-sensitive.[6] But much of it is operational, and that part retailers control. Shelves are empty during the 07:00 departure wave. Tills are understaffed when three long-haul flights board at once. Promotions don't match the people at the gate. And when a flight is delayed by 90 minutes, nobody acts on the extra time those passengers suddenly have.
Airport retail doesn't have a demand problem. It has a timing problem, and timing is something AI is very good at.
Why does high-street retail AI fail in airports?
Most retail forecasting systems learn from calendar patterns: weekday, season, holidays, weather and promotions. In an airport, the departure board matters far more. A calendar model smooths the day into one gentle curve; real airside demand arrives in sharp waves.
| Factor | High-street retail | Airport retail |
|---|---|---|
| Main demand driver | Calendar, weather, promotions | Flight schedule, load factors, destinations, delays |
| Shopper | Repeat local customers | Mostly one-time visitors with a boarding deadline; about 70% of purchases are impulse-led[2] |
| Mix | Stable local demographics | Changes by route: a Gulf departure and a domestic shuttle buy different things |
| Replenishment | Overnight trucks, stockroom buffer | Security-screened airside deliveries, tiny stockrooms, fixed delivery windows |
| Staffing | Shift patterns | Waves tied to departure banks |
We covered the general retail inventory problem in The $1.77 trillion inventory crisis. Airports add a tougher constraint. You can't fix a stock-out with an afternoon delivery when every item must pass security and your stockroom holds a few hours of sales.
What data does flight-aware airport retail AI need?
The good news: most of the data already exists. It is scattered across the airport operator, airlines, ground handlers and the retailer. SITA's 2025 IT Insights report found 89% of airports rank data-driven decision-making as a strategic priority. Its CEO summed up the catch: "Data coordination is not a future priority."[5]
You don't need every feed on day one. Flight schedule plus real-time status, joined to hourly POS and stock for a handful of stores, already explains much of the variation a calendar-based forecast misses. Add queue data and route mix in the next phase.
Want to know which of these feeds your organisation can actually use today?
Get a free AI Readiness ScoreWhat are the highest-ROI AI use cases in airport retail?
These six use cases share one design principle: every decision is tied to a flight, a store and an hour. Where public evidence exists, we cite it. Treat it as directional, because results depend heavily on data quality and execution.
Flight-aware demand forecasting
Forecasts sales per store, per category and per hour from departing flights, aircraft size, loads, destinations and expected dwell time, not just last year's sales. This forecast is the foundation for the other five use cases.
Airside replenishment and stock-out prevention
Plans security-screened deliveries around departure banks, so best-sellers are full before the wave, not after. Flags at-risk items hours ahead and suggests transfers between nearby stores.
Rostering to flight banks
Matches staff to forecast footfall in 15- or 30-minute slots, and recommends breaks during troughs. Fewer abandoned queues at the till means more conversion.
Assortment and offers by departing mix
Adapts featured ranges, bundles, digital screens and staff prompts to the routes and nationalities at nearby gates in the next hour. Avolta's CEO named real-time personalisation through data as a key 2026 opportunity.[7]
Disruption response
When a flight is delayed or a gate changes, the agent recalculates demand for nearby stores. It alerts managers, shifts staff and pushes relevant offers to the right screens and apps within minutes.
Concession and tender intelligence
Generative AI drafts tender responses, benchmarks sales per square metre, and models minimum-guarantee scenarios from your own data. Lagardère Travel Retail's Lumina tool reportedly raised tender-response productivity by 30–40%.[6]
What does a day with a flight-aware AI agent look like?
Here is an illustrative weekday in one terminal with five airside stores. It shows where agents act on their own and where a person decides.
The forecast agent pulls the day's schedule, aircraft types and load estimates, then produces an hourly forecast for every store and category.
AutomaticThe replenishment agent sees fragrance and confectionery short at Store 3 ahead of three Gulf and Asia departures. It adds items to the 05:45 airside delivery.
Automatic, within limitsQueue time jumps to 14 minutes, so passengers will reach stores later and with less time. The rostering agent suggests moving one colleague from replenishment to tills at 07:20.
Manager approvesAbout 300 passengers gain unexpected time near Gate B20. The disruption agent raises the forecast for the nearest store and F&B outlet, and prompts a bundle offer on nearby screens.
Offer approved from a pre-set playbookManagers receive penetration, conversion, availability during peaks and forecast accuracy by store, plus three recommended actions for tomorrow.
AutomaticThe architecture behind this is the same multi-agent pattern we describe in Agentic AI architecture. What differs is the input: the departure board becomes the most important signal in the system.
How should airports and retailers share data safely?
Flight-aware AI only works when the airport operator and its concessionaires share operational data. That makes governance a commercial question as much as a technical one.
| Risk | Control |
|---|---|
| Passenger privacy (GDPR and similar laws) | Use aggregated, flight-level data: counts, route and nationality mix, queue times. Use personal data such as loyalty or boarding passes only with a lawful basis and consent. |
| Price trust and the duty-free price promise | Agents recommend; people approve price changes. Keep pricing transparent and comparable, a point Avolta's CEO made about restoring trust.[7] |
| Unclear data ownership between airport and retailer | Write data-sharing terms, refresh frequency and permitted uses into the concession agreement or a side letter. |
| Cybersecurity | Read-only access to airport systems, least-privilege credentials, and full logs. 71% of airports rank cybersecurity as their top IT priority.[5] |
Is your airport retail operation ready for AI agents?
Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls.[8] In airport retail, most failures trace back to data that was never joined up. Check these six points before funding a build. (Background: Why enterprise AI agents fail before they launch.)
- Flights: Can you get the day's flight schedule and live status for your terminal in a machine-readable feed?
- Sales and stock: Are POS sales and on-hand inventory available by store and by hour, including stock in transit airside?
- Flow: Do you have security queue times or footfall counts near your stores?
- History: Do you have at least 12 months of hourly sales you can line up against historical flights?
- Agreement: Does your concession contract allow the data sharing you need?
- Owner: Has an operations leader agreed to own decisions, thresholds and adoption in stores?
What does a 90-day airport retail AI pilot look like?
Keep the scope tight: one terminal, three to five stores, two use cases (usually forecasting plus replenishment or rostering). Prove value, then expand.
Assess
- Score data readiness across flight, POS, stock and roster data
- Set the baseline: penetration, conversion, availability, labour
- Pick stores and two use cases
- Agree data-sharing with the airport
Prove
- Build the flight-aware hourly forecast
- Run agents in shadow mode next to current planning
- Compare forecast error and recommended versus actual actions
- Tune with store managers
Go live
- Switch on replenishment or roster recommendations
- Managers approve via mobile alerts
- Track KPIs daily against baseline
- Build the business case for the full terminal
Which KPIs prove ROI for AI in airport retail?
Measure per store and per hour, not per month. The value hides inside the peaks.
| KPI | Definition | Why it matters |
|---|---|---|
| Penetration | Store visitors ÷ departing passengers | The industry's weakest number today[2] |
| Conversion | Transactions ÷ store visitors | Shows whether staffing, availability and offers work |
| Spend per passenger (SPP) | Sales ÷ departing passengers | The metric airports and retailers share |
| Peak availability | % of top SKUs in stock during departure banks | Stock-outs at peak are the most expensive kind |
| Hourly forecast error | Weighted absolute error by store and hour | The leading indicator for everything else |
| Sales per labour hour | Sales ÷ staffed hours | Shows whether rostering follows demand |
How Innoflexion helps travel retailers and airports
Innoflexion builds agentic AI on DeepRoot, our platform for data readiness, multi-agent orchestration and governed deployment inside your own cloud. We connect to the systems you already run, including POS, ERP, workforce tools and airport feeds, so there's no rip-and-replace. Our clients include global brands across aviation and retail, among them Singapore Airlines, Japan Airlines and WHSmith.
Results from Innoflexion client deployments in retail, consumer products and aviation; client names for individual results are withheld under NDA. Reference calls available during evaluation.
Find the hours where your stores are losing sales
In 30 minutes we'll look at one terminal's flight pattern against your sales data and show where availability, staffing or offers are costing conversion.
- Complimentary AI Readiness Assessment and DRI score
- The three highest-ROI use cases for your stores
- A pilot plan you can launch in 4–8 weeks
Frequently asked questions
How is AI used in airport retail?
AI uses flight schedules, passenger mix, security queue times, POS and inventory data to forecast demand by store and by hour. It then replenishes stock before departure peaks, rosters staff to flight banks, tailors offers to the routes departing, and responds to delays. Agentic AI takes or recommends those actions automatically, with people approving price and staffing changes.
Why does standard retail forecasting fail in airports?
High-street forecasting relies on calendar patterns, weather and repeat customers. Airport demand depends on which flights depart, when, to where, how full they are and how long passengers have airside. A delay or gate change can swing demand at one store within an hour, so forecasts need flight-level data.
Does longer dwell time really increase airport retail sales?
In most terminal designs, yes. A 2024 study of 89 US airports in the Journal of Air Transport Management found that a 10% increase in dwell time was associated with about 6% more retail revenue and 8% more F&B revenue. The effect was not significant at concourse-style airports.
How long does an AI pilot for airport retail take?
A focused pilot covering three to five stores in one terminal usually shows measurable results in 8 to 12 weeks if flight, POS and inventory data are accessible. A working proof of concept typically takes four to eight weeks and production three to six months.
Is sharing passenger data between airports and retailers compliant with privacy laws?
It can be, if you use aggregated or anonymised data such as flight-level counts, route mix and queue times. Loyalty or boarding-pass data needs a lawful basis and consent under laws such as GDPR, and the terms belong in the concession or data-sharing agreement.
Sources
- ACI World, ACI World releases global airport traffic forecasts, January 28, 2026.
- TRBusiness, TR Consumer Forum: unlocking the potential to generate up to $113bn by 2031 (m1nd-set research), 2026.
- The Moodie Davitt Report, Non-aeronautical revenues continue recovery but remain behind pre-pandemic levels, says ACI World, February 11, 2026.
- Wu, Y., Morlotti, C., & Mantin, B. (2024). Shopping or dining? On passenger dwell time and non-aeronautical revenues. Journal of Air Transport Management, 118, 102620.
- TravelDailyNews, SITA report highlights data coordination as key to aviation technology investment (SITA 2025 Air Transport IT Insights), April 27, 2026.
- Artefact, Rethinking travel retail: how leaders are scaling with pragmatic data-driven solutions, 2025.
- TRBusiness, Avolta: 2026 will be a "year of acceleration" for the travel retail industry, January 2026.
- Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, June 25, 2025.
Last updated October 13, 2026. The day timeline is an illustrative scenario, not a client case. Third-party results are as reported by the cited sources.