01:20
Late-night low-cost, domestic
High load factor, short dwell, price-sensitive, travelling light. Duty-free is close to irrelevant. Hot food, water and convenience carry the bank — and most of the estate is closed.
AeroYield — early deployment
AeroYield turns a flight schedule into a commercial plan.
An AI commercial operating system for airports. It forecasts passenger demand at the flight and cohort level, then orchestrates the commercial estate — retail, food and beverage, duty-free, parking and advertising — against that forecast.
The commercial problem
Two revenue streams, one of them unmanaged
An airport earns from two streams. Aeronautical revenue — landing fees, aircraft parking, passenger service charges — is regulated, and its margin is capped by the regulator. Non-aeronautical revenue — retail, food and beverage, duty-free, car parking, advertising — is unregulated, carries a far higher margin, and is where the profit actually comes from.
The second stream is not driven by how many passengers pass through the terminal. It is driven by which passengers, at which hour, in which state. A late-night low-cost domestic bank and an evening wide-body long-haul bank move through the same square metres and produce almost nothing in common: different dwell, different categories, different willingness to spend, different staffing requirement. Managed as a daily average, both are served badly.
The information needed to tell them apart already exists. It sits in the flight schedule, in historical load factors, in day-of-operations changes and in concessionaire point-of-sale. AeroYield joins those sources into one demand picture, resolves it down to 30-minute windows, and gives the commercial estate something to act on.
The system
Five engines
One forecast, four ways to act on it, and a governed data layer underneath.
Flight-schedule demand forecasting
AeroYield ingests flight schedules, historical load factors, day-of-operations changes and concessionaire point-of-sale, and builds a demand profile per route and per cohort. The output is a 30-minute-window prediction, 7 to 90 days ahead.
- Zone footfall, by 30-minute window
- Category demand — food and beverage, retail, duty-free, convenience
- Staffing requirement per zone and shift
- Re-forecast on day-of-operations changes, not on a nightly batch
Pre-order commerce
Shop before you fly. The passenger orders in the airline or airport app, the order is assembled in back-of-house rather than at a counter, and it is collected at the gate or in the lounge. Food preparation is triggered by the boarding-pass scan, so it arrives hot.
- Ordering in the surfaces passengers already use
- Assembly in back-of-house, off the queue
- Preparation triggered by the boarding-pass scan
- Collection at gate or lounge, timed to the departure
Dynamic pricing and space yield
Idle capacity is used without touching shelf-price integrity. Off-peak promotions run against forecast troughs, parking is pre-bookable and priced against the departure curve, and a revenue-per-square-foot-per-hour heatmap shows which square metres earn, and when.
- Off-peak promotion aimed at forecast troughs, not at the shelf price
- Pre-bookable parking priced against the departure curve
- Revenue per square foot per hour, from anonymised footfall analytics
- Zone-level yield, so under-earning space is visible before the lease renews
Programmatic advertising
Airport screens become a real-time exchange. Advertisers bid against the passenger cohort actually arriving at a screen, rather than buying a static slot and hoping the right audience walks past it.
- Cohort-addressable inventory instead of fixed slots
- Real-time bidding against the forecast, per screen, per window
- Screen inventory priced by who is in front of it
- Cohorts are aggregated groups, never identified individuals
Data clean room
Aggregated, consent-governed and compliant with the Digital Personal Data Protection Act. Behavioural models learned at one airport transfer as a baseline to the next — the model moves, the personal data does not.
- Aggregated and anonymised; never passenger-identified
- Consent-governed under the DPDP Act
- Models transfer between airports; raw data does not
- Reporting is aggregate by construction, not by policy
Worked example
Same terminal. Same square metres. Two different businesses.
Two departure banks in one terminal, roughly nineteen hours apart. The estate around them is identical. What it should be doing is not.
21:10
Evening long-haul, wide-body
Long dwell, early arrival at the terminal, high duty-free intent, meaningful lounge use, and a parking profile measured in days rather than hours. The same floor plate, asked to do something entirely different.
Illustrative model showing relative category demand on a 0–100 index. Not measured data, and not a forecast for any airport.
Deployment
What integration actually involves
AeroYield reads before it writes. Every write path is granted per mechanism and per zone, and every one of them has a rollback.
- 01
- Schedule and operations
- AODB, flight schedule feeds and day-of-operations changes, over the interfaces you already run. No new operational system is introduced into the critical path.
- 02
- Commercial systems
- Concessionaire point-of-sale at aggregate level, parking and pre-booking systems, lounge and food-and-beverage ordering.
- 03
- Estate and screens
- Digital signage and screen networks, wayfinding, and the ordering surfaces in the airline or airport app.
- 04
- Environment and rollback
- Read-only throughout the pilot. Write access is granted per mechanism and per zone, with a rollback path and an audit trail on every action AeroYield takes.
- 05
- Standards
- PLACEHOLDER: list the specific schedule, point-of-sale and signage interface standards supported, with versions.
Governance
Compliance and data governance
The commercial case does not survive a privacy failure, so the constraints are designed in rather than bolted on.
Legal basis
DPDP Act
Processing is designed to the Digital Personal Data Protection Act: purpose limitation, consent governance, and defined retention. Data fiduciary and processor roles are set out in the contract before deployment.
Method
Anonymised analytics
Footfall and dwell analytics are aggregated at the zone and cohort level. AeroYield does not need to know who a passenger is in order to forecast what a cohort will do.
Explicitly excluded
No facial recognition
AeroYield does not use facial recognition and does not build biometric passenger profiles. This is a design constraint, not a configuration option.
Output
Aggregated reporting
Reports resolve to cohorts, zones and windows. Advertisers and concessionaires receive cohort demand, never passenger records.
Questions
The objections worth raising first
Airports build operational systems well, and the constraint here is not engineering capacity. It is the forecasting layer: flight-and-cohort demand models have to be evaluated against actuals continuously, and that needs a team whose full-time job it is. The commercial structure differs too — an in-house build is paid for whether or not it produces uplift. AeroYield is not.
Nothing is taken from them. Concessionaires keep their leases, their pricing authority and their shelf-price integrity. What they gain is a forward demand signal for their zone and category — how many passengers, of which cohort, in which 30-minute window — so they can staff and stock against it. Pre-order and off-peak mechanisms route additional transactions through them. Where a concession agreement sets a revenue share, more turnover through that concession is more revenue to the airport.
Before anything is deployed, we agree a baseline of non-aeronautical revenue per passenger with your finance team: the period it covers, its scope, the seasonality adjustment and the treatment of one-off items, all written down. Uplift is then calculated from your systems of record rather than ours, and the calculation is open to your auditors. The baseline is fixed for the term and does not move because a good quarter arrived.
Least privilege, and read-only first: flight schedules and day-of-operations changes, historical load factors, concessionaire point-of-sale at aggregate level, parking systems and, where the advertising engine is in scope, screen inventory. We do not need passenger identity, we do not use facial recognition, and everything behavioural is aggregated and consent-governed under the DPDP Act. Write access is granted later, per mechanism and per zone, with a rollback path.
Forecast accuracy is measurable inside the pilot, before any commercial action is taken — that is the first thing you will be able to judge us on. Revenue uplift follows the first commercial mechanisms going live and is measured against the audited baseline from that point. AeroYield is in early deployment, so we will not quote a timeline as though it were established practice; we set expectations against your own baseline during the audit, in writing.