How airline revenue management works

The problem revenue management solves

A seat on a departing flight is perishable: unsold, it is worth nothing. Selling every seat early at a low fare fills the aircraft but turns away the late business traveller who would have paid several times more; holding seats for that traveller risks flying them empty. Revenue management balances the two — in MIT's framing, a balance of yield (average fare) and load factor.

The idea was formalised early. Ken Littlewood's 1972 paper for the AGIFORS symposium, reprinted in the Journal of Revenue and Pricing Management in 2005, framed the objective as maximising the revenue from each flight rather than the number of passengers. Two decades later, American Airlines' operations researchers estimated the quantifiable benefit of their yield-management system at 1.4 billion US dollars over three years, in a 1992 paper in Interfaces.

Fare classes and the fences between them

An airline sells one cabin through several fare products, each in a booking class, each with restrictions that make it unattractive to the travellers the airline wants to charge more. The example below is the illustrative fare structure used in MIT lecture material.

The fences are imperfect by design: some travellers who could pay more meet the restrictions anyway. The revenue-management system's job is to decide how many seats each class may sell on each departure. For the vocabulary, see fare class.

Nesting and protection levels

Classes share one inventory of seats, and they are serially nested: seats protected for the top class cannot be sold to lower classes, seats protected for the top two cannot be sold below them, and so on. A request in a higher class is accepted as long as any seat is available to a lower one.

Protection levels come from expected marginal revenue. In the EMSRb heuristic taught at MIT, the airline protects seats for a higher class as long as the probability of selling that extra seat at the higher fare, times the higher fare, exceeds the lower fare. In the lecture's example, Y demand has a mean of 10 and a standard deviation of 3, Y pays 1,000 and B pays 700; the eighth protected Y seat still has a probability above 0.70 of selling, the ninth does not, so eight seats are protected from B. With two classes this is Littlewood's rule.

Forecasting, overbooking and constant re-optimisation

• Data. A typical third-generation system keeps historical bookings by flight and fare class for each past departure.
• Forecast. It forecasts remaining demand and no-show rates by departure date and class.
• Overbooking. It sets an authorised capacity above the physical seats, chosen to minimise the combined cost of denied boardings and seats that fly empty.
• Booking limits. It optimises limits for each class, and analysts can accept or override the recommendation — for example around an unforecast event.
• Re-optimisation. Forecasts and limits are revised at checkpoints as bookings arrive, which the MIT material notes can change fare-class availability substantially from one day to the next on the same departure.

That last point is why the cheapest fare on a route can disappear between two searches; how often flight prices change looks at it from the data side, and flight availability at how to observe it.

Origin-and-destination control

Leg-based control treats each flight on its own. A hub airline, though, sells seats on one leg to local passengers and to connecting passengers whose journey uses two or three legs. A connecting fare can be higher than a local one and still be worth less to the network, because the passenger also takes a seat on the next leg that a local passenger might have paid for.

Origin-destination (O&D) control values each request by its network contribution: the total fare minus the displacement cost on the other legs it uses. A common mechanism is a bid price per leg — the minimum acceptable revenue for the next seat; a journey is sold if its fare exceeds the sum of bid prices on its legs. Under high demand, connecting passengers get fewer seats and local passengers more. The MIT material puts the additional gain from effective O&D control at 1–2% of network revenue, rising with load factor, and notes that it is harder to implement than leg-based control.

For a buyer of fare data, the consequence is that the price of A–C via B is not the sum of A–B and B–C, and availability on A–B can differ by the journey it is sold as part of. Airline route intelligence works with exactly that difference.

From 26 booking classes to continuous pricing

IATA's Dynamic Offers factsheet describes the limits of the traditional model: filed fares distributed through one-letter booking classes, with pricing control constrained to a maximum of 26 classes. Dynamic pricing in its sense adjusts the price in real time from contextual information at the time of shopping — length of stay, time before travel, day of departure and arrival, competition, remaining capacity — without filing fares with a third-party system and without necessarily using personal information.

Continuous pricing goes one step further and allows indefinite price points, so a fare is no longer forced onto the nearest class step. Dynamic bundling adds ancillaries to the offer. These rely on airline direct and NDC channels, and IATA states that anonymous shopping remains possible.