Award seat cost
What does an award seat really cost the airline? Almost nothing on an empty flight, and a paying passenger's fare on a full one.
An award seat costs the airline the larger of two numbers: the marginal cost of one more passenger, and the displacement cost, the fare the seat would otherwise have sold for, which is the bid price revenue management uses. Compare it with what the loyalty programme pays for the seat to see when releasing award seats helps both sides.
Loading the flight…
1. The cost of an award seat, from empty to overbooked
The same flight at every level of expected demand, from half full to 10% more passengers than seats. Where the black cost curve sits below the green payment line, the programme pays more than the seat costs: releasing it helps the member, the programme and the airline.
Cost of the first award seat
CA$ per seat, one wayWith these numbers
Bid prices and why they rise as a flight fills: RM 101, chapter 04. What an award costs the airline and how Loyalty and RM share the seat: Loyalty 101, chapter 10.
The flight
Five fare classes on one cabin, low fares booking first. Demand in each class is uncertain; the standard deviation includes both booking-by-booking randomness and forecast error.
2. How many award seats to release on this flight
Each extra award seat takes one more seat away from paying passengers, so it costs a little more than the one before. Releasing by cost means opening award seats while the next one still costs less than the programme pays. A fixed allocation opens the same number everywhere.
Cost of each additional award seat
CA$ per seatRelease by cost vs fixed allocation
Seats released but not booked stay on sale to paying passengers, so only booked award seats cost anything. The cost of each booked seat is the one in the chart above.
3. Three flights, one rule
The three example flights at their own expected demand and award requests, with the costs, payment and fixed allocation set above.
4. Why saver awards disappear on peak dates
Members read an empty award calendar at Christmas as the programme hiding seats. From the airline's side it is the cost curve at work, for three reasons that stack up:
- The flight is expected to fill. Past about 90% expected demand, an award seat is very likely to take the place of a paying passenger, so its cost jumps from the marginal cost to a fare.
- The fares it displaces are higher. Peak dates carry peak fares and more late, high-fare bookings, so the seat the award takes is worth more.
- The forecast is more certain. Holiday demand is predictable, so revenue management is confident the seat will sell and has little reason to take a chance on it.
A saver award pays the airline the same amount on every date, so on peak dates its line sits under the cost curve and no saver seats open. That is why programmes add a second, higher price: a standard award, or dynamic award pricing tied to the cash fare, which pays enough to cover the displacement. Saver seats can also reappear close to departure: if bookings come in below the forecast, the bid price falls and the RM system releases more award seats.
More on this trade-off: Loyalty 101, chapter 10: Where Loyalty Meets RM, and on what members get for their points: the award value comparator.
How it works
Background
When a member books a seat with points, the airline carries one more passenger and, if the flight would have filled, loses the sale of that seat. So the cost of an award seat is the larger of two numbers: the marginal cost of carrying one more person (fuel, catering, fees) and the displacement cost, the revenue the seat would otherwise have earned. That second number is the bid price of the last seat, the same opportunity cost that revenue management uses to accept or reject any booking.
The flight and its demand
An invented 3½-hour domestic flight with 150 economy seats and five fare classes. Each example flight has its own fare level (85%, 100% and 130% of a base fare ladder of CA$620, 450, 340, 250 and 180), class mix and demand variability. Expected demand in class j is μj = L · 150 · sj, where L is the expected demand as a share of seats and sj the class share. Demand is a negative binomial draw (a Poisson whose mean has a gamma-distributed error with coefficient of variation cv), so its variance is μ + (cv · μ)². The inputs are in assets/data/awardseatcost.json.
Displacement cost: the bid price of the last seat
Let V(x) be the expected revenue from paying passengers with x seats, when classes book from the lowest fare to the highest and the airline protects seats for higher fares optimally (the static single-leg model of Brumelle and McGill, 1993, solved by dynamic programming as in Talluri and van Ryzin, 2004, section 2.2). For class j with fare fj, the seats protected for the higher classes are the largest y with Vj−1(y) − Vj−1(y − 1) > fj, and
Vj(x) = E[ fj · min(Dj, x − y) + Vj−1(x − min(Dj, x − y)) ].
The k-th award seat leaves 150 − k seats for sale, so its displacement cost is Δk = V(150 − k + 1) − V(150 − k). For the first award seat this is the expected marginal seat revenue of the last seat, which EMSR-b (Belobaba, 1989) approximates and Littlewood's rule (1972) gives for two classes.
Cost, release and net value
- Marginal cost m = extra fuel + catering + fees. The defaults assume about 100 kg for a passenger and bag, 3–4% of that extra weight burned as fuel per flight hour (a common rule of thumb, not a measured figure), and a jet fuel price near US$185 a barrel.
- Cost of the k-th award seat: ck = max(m, Δk). Δk never falls as k grows, so neither does ck.
- Release by cost: open award seats while ck < p, the programme's payment per seat. Booked awards = min(seats released, members asking). Airline net = Σ (p − ck) over the booked seats. The fixed allocation uses the same formula with its own number of seats released.
- The cost curve in section 1 repeats the calculation of c1 for every L from 50% to 110%, holding the fares, class mix and variability of the chosen flight. The break-even is where it crosses p, interpolated between whole percentages.
Sources
- Littlewood, K. (1972), "Forecasting and control of passenger bookings", AGIFORS Symposium Proceedings 12; reprinted in Journal of Revenue and Pricing Management 4(2), 2005.
- Belobaba, P. (1989), "Application of a probabilistic decision model to airline seat inventory control", Operations Research 37(2).
- Brumelle, S. and McGill, J. (1993), "Airline seat allocation with multiple nested fare classes", Operations Research 41(1).
- Talluri, K. and van Ryzin, G. (2004), The Theory and Practice of Revenue Management, Springer, chapter 2 (single-resource capacity control).
- IATA and S&P Global Platts, Jet Fuel Price Monitor, global average of about US$185 a barrel, read on 30 September 2026. Used only for the order of magnitude of the fuel default.
- Charts with D3 7.9.0, loaded from jsDelivr. Everything else runs in the page.
Limitations
- The maximum of the two costs is a simplification. Strictly, the expected cost is m · P(seat would fly empty) + Δk: if the seat would have sold, the airline carries a passenger either way. The two differ by less than m, and they agree at both ends of the curve.
- Demand in each class is independent of what is open: nobody buys up when a cheap class closes (RM 101, chapter 05). Once the flight is expected to be full, the curve levels off near the lowest fare still on sale; with buy-up it would keep rising.
- Award seats are decided once, at the opening of sales. Real systems re-optimise as bookings come in, and a connecting award displaces seats on several flights at once (network bid prices).
- The programme's payment is one fixed number per seat. Real arrangements vary: some pay by cabin, route or date, and dynamic award pricing ties the payment to the cash fare. The value of availability for member engagement, which is why programmes accept some awards that don't pay their way, isn't modelled.
- Taxes and carrier charges the award passenger pays, cancellations, no-shows and overbooking are left out.
This is a personal project and isn't affiliated with any airline or loyalty programme.