Keeping yield management under control May 1998
Yield management is basically a simple idea, but its application can become hideously complicated in practice. In this article Thomas Weyer reviews the development of yield management techniques and comments on how best to measure their success:
The mission of airline yield management is the control of the reservations inventory offered to the flying public. The availability of seats is managed in such a way as to maximise — if possible — the revenue earned at aircraft departure time, given the flight schedule and fare structure.
The ultimate goal of yield management is the improvement of marginal passenger revenue, through an increase of average fares, or an increase in departure load factors, or both. To achieve this, yield management concentrates on three basic functions — overbooking, discount allocation and origin and destination control.
Over–booking is the practice of selling more reservations for a flight than there are seats on the aircraft, attempting to offset the passenger cancellations and no–shows that airlines usually permit without penalty. Thus over–booking attempts to minimise the number of "spoiled" seats — those departing empty at flight time. The practice has its risks: if over–booking levels are too high and more passengers show for a flight than there are seats, penalty costs and ill–will are incurred. Mathematically, yield management attempts to control over–booking to the point at which the benefit of allowing an additional reservation is negated by the marginal cost of an over–sale.
Most airlines offer two, and sometimes three, distinct classes of service: first class, business class and economy. Each class is generally distinguished by identifiable features: seat pitch and size, meal service, check–in privileges, and other amenities for which a passenger would expect to pay a higher price.
In the early/mid–1970s, however, airlines began offering discount fares (within specific classes of service), primarily to stimulate latent demand. The primary difference among degrees of discount tended to be time: the earlier the reservation, the greater the magnitude of discount.
Because the industry began offering a range of seat prices, identified by fare code within the reservation record and on the ticket, control of sales by degree of discount was possible. Today, control of the price at which a seat is sold is exercised by the decision whether or not to reject the next seat sale at a discounted price.
On one hand, if the carrier accepts the passenger request for a discounted seat, the discounted price is the amount the carrier will earn. By rejecting the discount request the carrier runs the risk that the seat will fly empty, without generating even the discount revenue.
On the other hand, rejecting the discount request "reserves" the seat for a potential sale at a higher fare code, hopefully generating greater incremental revenue. The decision whether or not to reject a discount sale rests on the probability that a higher fare ticket will be sold.
This probability varies according to the season, day of the week, time of day, closeness to holidays, origin and destination, and the proximity of the impending flight date. The number of potential influences on the probability that a higher fare ticket can be sold appears almost unlimited, and control of these influences boils down to three basic factors:
- The magnitude of future expected demand;
- The accuracy of the demand forecast, and;
- The probability of sell–up to a higher fare class.
Origin and destination control
The implications of hubbing for yield management are dramatic. By virtue of connection potential at hubs, a single flight carries passengers to multiple destinations. Because of the passenger diversity on a single flight, seat inventory can no longer be optimised by concentrating only on fare code. Within a fare class on a single flight, multiple fare levels represented by multiple destinations and fare combinations compound the complexity of revenue optimisation.
Furthermore, fares are constantly changing as a result of operating cost adjustments, market promotions, and competitive influences. Carried to an extreme, it is possible for any particular flight to have as many different revenue levels as there were sold seats.
To simplify management of the permutations of fares, fare combinations are now grouped into a limited number of "buckets" of similar dollar magnitude. Fare code analysis has therefore given way to fare bucket analysis, the goal of which is the establishment of a reasonable difference between bucket levels. Thus the ability to sell or restrict the sale of seats (by bucket) includes the effect of multiple origin and destination parameters.
Techniques and analysis
Successful implementation of yield management requires an ongoing, periodic forecast of demand based on accurate data. Forecasts are initiated and maintained up to departure time by fare bucket for each flight within the control of the revenue management system.
Traditional yield management systems forecast demand by flight leg (aircraft hop) or by flight segment (passenger haul), as both historical leg and segment data are readily available in the airline reservation system.
Leg class controls
In this case inventory controls are established by flight leg for each fare class, based on a demand forecast by fare class flowing over the flight leg.
The inventory controls or allocations of seats per fare code may be nested or non nested. With multiple serial nesting a lower class is nested into the next higher class. This means that seats in a lower valued class will not be made available for sale when a higher valued class is closed for sale.
With multiple parallel nesting each lower fare class is nested into the highest fare class, but there is no nesting of the lower classes into intermediate classes. Hence, with parallel nested controls a lower class may be open for sale while an intermediate (higher) class is closed.
For these two nesting schemes several variations can be applied. For example, on a single flight both serial and parallel controls can be enacted. The obvious advantages of serial controls may be employed for normal traffic, with parallel controls reserved for wholesalers or special promotions.
Segment class controls
In this case inventory controls are established by segment for each fare class without regard for O&D demand for various passenger types.
With segment class controls total capacity on a given flight leg is partitioned to each of the distinct segments that flow over the leg. For example, consider a flight routing from A to B to C. Capacity is partitioned based on a forecast for A–B, B–C, and A–C. Furthermore, within each segment classes of fares can then be nested either in serial or parallel, whichever the particular case may require.
For airlines operating flights with multiple stops under the same flight number, segment close indicators (SCIs) may prove advantageous in restricting segment classes that are lower valued when demand exists for higher valued segment classes. So, using the previous example, the airline may be justified in closing a certain fare class on the A–B leg, but keeping open that same fare class on the A–C leg. The weakness of SCIs is that the technique must be continuously monitored for cancellations.
Under the technique known as segment limits, an inventory cap is placed on the sale of class leg sales. In this manner the airlines inventory control system is permitted to backfill seats should passenger cancellations occur.
O & D controls
Origin and destination control can be described as "itinerary" control. An itinerary describes a passenger’s one–way city pair, including connect points and time of day. Control of sales at the itinerary level is highly complex with the value of the passenger determining reservation availability. The value of the passenger is based on several factors such as itinerary, departure date, class of service (first, business, economy), fare class within class of service, published fare, "confidential" fare, and point of sale.
Levels of sophistication inherent in seat inventory control are illustrated by the following examples:
1. No controls
When the airlines reservations system establishes no controls, the last seat sold will be sold at the lowest fare offered over a particular leg.
2. Leg-class controls
Under leg–class controls inventory value is based on the relative value of the various fare classes, without regard to origin and destination. In this scenario, the last seat sold on a particular leg will occur at the highest fare class, ostensibly to preserve seat availability for higher value multiple leg journeys.
3. Segment class controls
With segment–class controls inventory is controlled based on the relative value of the various classes by segment. In this scenario, the last seat sold will occur at the highest fare class over the particular segment.
4. Itinerary class controls
Itinerary class controls govern sales based on the relative value of the various itineraries flowing over a specific flight leg. Trade–offs can be made to accept or reject local, through and connecting itineraries. In this scenario, airlines can sell the last seat to the highest valued itinerary class over the entire network that flows over the particular leg.
As stated earlier, the number of permutations and combinations of fare class restrictions under an O&D (itinerary) control method requires simplification before management of the inventory can become effective.
Simplification demanded that the number of fare classes be reduced to a manageable number of buckets, based on customer value. This is accomplished by clustering the various itinerary fare classes into buckets based on the value of the customer to the airline. So a bucket, which consists of several itinerary fare classes, is used to control seat inventory instead of fare class code control.
To an airline then, the value of an itinerary class is the passengers fare, net of the opportunity costs associated with passengers displaced up–line and downline. Thus, passenger itineraries flowing over each flight leg in the network are clustered into a manageable number of buckets.
The clustering process is highly complex, basically consisting of algorithms to minimise the variance of customer values within a bucket, while at the same time maximising the separation between buckets. The buckets are serially nested to make sure that as sales build up in advance of departure time, the lowest fare classes will close automatically.
The problem of performance measurement of revenue management is a continuing source of debate. Above all, airline industry profitability is highly volatile and management reaction time to the vagaries of the marketplace is relatively short. Thus the performance of the current period may not be comparable to the performance of prior similar periods.
But performance measurement statistics do exist. Overall measures of the effectiveness of a yield management system consist of such post–departure statistics as: spoilage, over–sales, load factor, revenue per revenue passenger mile, revenue per available seat mile, flight and fare class closing rate, and spilled revenue. Available pre–departure statistics include advance bookings, group bookings and inventory classes closed.
Month over month and year over year comparisons can be made, with the difficulty however of weeding out the anomalies which render erroneous conclusions. In so far as successful yield management depends on accurate forecasting, errors in demand forecasts, cancellation forecasts and no–show forecasts must be studied by airlines and eliminated to the highest degree possible.
In some instances yield management disciplines, hastily or improperly applied, can lead to detrimental results. Forecast accuracy is critical. Where flights display a wide degree of variability, consideration should be given to exclusion of the flight from the yield management system. Since yield management is an exercise in fine–tuning sold seat inventory, a reasonable degree of stability/ predictability is required.
Flights which should be considered for exclusion from the yield management system include those with a high number of legs, flights with legs exhibiting a high degree of load variability such that overbooking and/or spoilage penalties prove to be the norm rather than the exception, and flights with low load factors.
Performance measurement invariably requires benchmarks against which improvement or deterioration can be ascertained. Establishment of yield management benchmarks, or standards against which performance can be judged, is not straightforward.
Too many uncontrollable variables cloud concrete comparisons. As a result, industry yield managers have tended to rely on complex mathematical modelling (simulations, or revenue opportunity models) to attempt to demonstrate the improvements attributable to yield management.
One such simulation is known as the "Min/Max Index". This index is designed to estimate flight revenue under three assumptions.
First, the "minimum" estimates revenue if no controls are in place; or in other words, capacity available for sale equals aircraft capacity. Minimum revenue equals unmanaged class demand times the average class fares. The simulation commences with the lowest fare class and continues until either there is no more demand or the aircraft is full.
Second, the "maximum" estimates ideal achievable revenue assuming perfect forecasts of both demand and no–shows have been made and perfect bucket authorisations have been established. Maximum revenue equals unconstrained class demand times the average class fare. Unlike the minimum, the calculation begins with the highest fare class and continues to the lowest fare class until there is no more demand, or the aircraft is full.
Third, the "actual" represents the revenue realised on the flight.
For analytical purposes actual revenue is expressed as a percentage of possible revenue within the range of minimum to maximum. Yield management controls are judged by historical increases in the percentage.
Again it should be emphasised that yield management is a fine–tuning discipline. Effort then should be concentrated on critical flights, or those flights that usually depart with high load factors.
It is these flights which run the risk of denied boardings due to over–booking, and yet they possess the demand availability for marginal increases in passenger load factor. Non–critical flights, or those with lower load factors, seldom impinge on over–bookings and thus sales at any fare level rarely close.