SSOM2 Ch12
SSOM2 Ch12
4
Matching Supply and Demand
The management of capacity in service firms frequently presents a more difficult and more expensive problem than in manufacturing. To determine capacity in many manufacturing firms, one usually considers long run average demand. Inventory is frequently used to compensate for annual seasonality or short term fluctuations in demand. In services, however, although the long run average must be considered, the short term is vital as well. Because of the perishable nature of service sector inventory such as airplane seats or hotel rooms, capacity plans must consider demand by day of the week and even time of daya level of detail that manufactur ers usually find irrelevant. Further, capacity planning mistakes are often more costly in services. When caught short on capacity many manufacturers simply back order a customers request. In many services, though, back orders simply cannot existquite literally, the ship has sailed for a cruise operator. Not having what the customer wants when she arrives can mean either a one time lost sale to the competition, or possi bly the defection of the entire stream of future sales of that customer and whoever else she chooses to tell about her experience. Many strategic and tactical decisions must be made concerning services capac ity. A one size fits all strategy will not work. Even though it may be in the strate gic interest of one airline to tightly match capacity to demand by heavily overbooking flights, another airline might be most profitable flying at far below capacity on aver age. Many of the qualitative aspects of these topics were covered in Chapter 2. This part of the book looks at putting those general strategies into action.
CHAPTER
12
Yield Management
LEARNING OBJECTIVES
I
Understand the need for overbooking. Use three different methods to calculate an overbooking level.
C APACITY S TRATEGIES
Capacity planning for many service firms can be far more difficult than for manufac turers. Manufacturers can set capacity by looking at long run average demand. For many service firms, however, long run averages become somewhat meaningless when capacity must react to general seasonality, daily demand variations, and time of day demand fluctuations. If the average manufacturer found out that most end consumers bought its product between 2 P.M. and 3 P.M., this knowledge wouldnt change its capacity strategy at all, but it would be important information for many service firms. Capacity decisions in service firms are not only more complex than in manufac turers, but can be more important as well. Manufacturers deal with short term imbalances in production and demand by either carrying inventory or creating a backorder list for later shipment. In most services, the inventory of capacity is employee time, or a fixed asset not being used, such as a hotel room or an airplane seat, so excess inventory cannot be stored for later use. Backorders quite often can not occur: Imagine a sales clerk at a department store stating that he will be able to speak with a customer by next Tuesday. Consequently, a temporary imbalance in supply and demand can result in either idle employees and resources if demand is smaller than supply, or lost sales to the competition if demand is larger than supply. (Service firms that can use physical inventory are discussed in Chapter 13.) These factors turn simple tactical decisions into strategic ones. Consider this simple example of the basic strategic direction for service capacity. An ice cream par lor experiences the following demand for ice cream cones:
234
CHAPTER 12
Yield Management
235
For the manufacturer supplying the cones, capacity is a simple matter: It calcu lates average weekly demand: 5(200) + 1,000 + 800 = 2,800 cones It makes 2,800/7 = 400 cones every day, and carries a small inventory of extra cones for the busier days. For the service provider who fills cones as customers walk in, however, simple arithmetic no longer applies. A strategic decision must be made. The ice cream parlor manager may use one of the four basic strategies outlined next.1 When considering these strategies, assume that one employee can make 100 cones per day. 1. Provide: Ensure sufficient capacity at all times. To carry out a provide strat egy, one would want to always have enough people to handle the maximum demand, so the firm would have 15 employees working on Saturday, 11 on Sunday, and 3 the rest of the week. It is usually difficult to employ significant numbers of part timers, so this strategy would employ enough full time employees to meet those numbers. This strategy is associated with a high service quality generic strategy, but it is also high cost, and would result in sig nificant idle time for employees. Businesses with these characteristics include high margin sales (e.g., jewelry, luxury automobiles) and those with wealthy individuals as clients (e.g., chauffeuring, private banking). Also, firms that compete on delivery speed (often called time based competitors) should adopt this approach. 2. Match: Change capacity as needed. This strategy would use ten employees on Saturday, eight on Sunday, and two the rest of the week, with the excess Saturday and Sunday employees strictly part timers. This approach balances service quality and costs and is representative of a large number of firms, including most mid and low priced restaurants and telemarketing firms. 3. Influence: Alter demand patterns to fit firm capacity. Here, pricing, market ing, or appointment systems flatten demand peaks to conform to capacity. It is most common in high capital intensive services such as airlines and hotels, but highly paid professionals such as medical doctors and lawyers also com monly use it. 4. Control: Maximize capacity utilization. If only full time employees could be used, five days per week, this strategy would have just two employees whose schedules overlapped on weekends. The generic strategy behind this option is to compete on cost by driving employee idle time to zero. It is often used in the public sector (e.g., drivers license bureaus) and low margin services, as well as situations where high priced employees want to maximize their utilization. Many physicians deliberately schedule patient appointments so tightly that a crowd is always in their waiting room. This strategy is willing to sacrifice sales at busy times to ensure the service functions efficiently all the time. To assist in crafting these strategies, a host of specific tactics can be used to man age supply and demand (an in depth discussion of these issues can be found in Klassen and Rohleder, 2001). Supply management tactics include the following:
1. Crandall and Markland (1996).
236
PA R T 4
Workshift scheduling. The unevenness of customer demand throughout a day means utilizing creative work schedules, such as nonuniform starting times, and workdays that have variable work hours. Work scheduling software is available to help construct flexible solutions within a match strategy. Increasing customer participation. A traditional method for a control strat egy cuts total labor by encouraging customers to participate in serving them selves. For example, many fast food restaurants use a semi control strategy in which customers pour their own fountain drinks and procure their own condiments. Adjustable (surge) capacity. Surge capacity means capacity that can be available for a short period of time. By cross training personnel for different jobs, a company can flexibly shift personnel temporarily to increase the capacity of any one position. Because cross training is expensive to under take, and cross trained personnel are more expensive to retain, it is an appro priate approach within a provide strategy. Sharing capacity. Capacity can often be shared between departments or between firms for personnel or equipment that is needed only occasionally. For example, small business incubators often contract with dozens of businesses to share the same secretarial, accounting, and office management team. Partitioning demand. It is not unusual for some components of demand to be inherently random, while some are fixed. This approach melds the more mal leable demand around the tendencies of the random demand. That is, if it is known that more walk in business generally comes in from 11 A.M. to 1 P.M., then schedule appointments either before or after that time. This approach works primarily for provide and match strategies. Price incentives and promotion of off peak demand. This highly common method works in an influence strategy, which many of us see in our telephone bills. It is also commonly used in restaurants (early bird specials), hotels (both off season and day of week pricing), resorts, and so on. Develop complementary services. The way to avoid the inevitable season ality of many services is to couple countercyclical services together: Heating and air conditioning repair, ski slopes in winter and mountain bike trails in the summer. Unfortunately, this approach remains only a theoretical con struct for most services. Yield management. Yield management combines three techniques: (1) over booking, (2) assigning capacity amounts to different market segments, and (3) differential pricing in different market segments. It is used extensively by many industries and is the subject of the remainder of this chapter.
Y IELD M ANAGEMENT
Consumers encounter examples of what is called yield management constantly. A lit tle knowledge about how these systems work can make life easier, or at least less expensive. Some practical examples of dealing with a yield management system include overbooking at a car rental agency. Even though you confirmed your reservation, it still pays to show up early in the day to get a car; sometimes those who show up late are out of luck. If you are a little more flexible about which days you fly, an airplane ticket may cost several hundred dollars less. The airline flight
CHAPTER 12
Yield Management
237
you are trying to book a seat on may be full today, but be patient and keep trying; tomorrow a seat may be available, even without others cancellations. A hotel that says no room is available for you on Thursday night may suddenly find a room for Thursday if you add that you are staying Friday. These situations occur because of yield management systems. The application of yield management practices often leave customers and employees puzzled. This chapter introduces the reasoning and techniques of yield management. Even if you do not work in an industry where yield management is practiced, this material will at least help you be a better consumer. The term yield management itself is a bit of a misnomer because these techniques are not directly concerned with managing yield but are really concerned with man aging revenue. Consequently, the set of techniques described in this chapter is some times called revenue management or perishable asset revenue management. The purpose of yield management techniques is to sell the right capacity to the right customer at the right price. Not every firm can use these techniques, but many capital intensive services can and do use them heavily. The main business require ment for using the techniques of this chapter is having limited, fixed capacity. Many other business characteristics make yield management more effective:
Ability to segment markets Perishable inventory Advance sales Fluctuating demand Accurate, detailed information systems
These characteristics increase the complexity of a business and the profit poten tial from applying yield management. Industries that currently fully utilize yield management techniques are transportation oriented industries, such as airlines, railroads, car rental agencies, and shipping; vacation oriented industries, such as tour operators, cruise ships, and resorts; and other capacity constrained industries, such as hotels, medicine, storage facilities, and broadcasting (selling commercial time). Many other indus tries can partially use these techniques. Yield management is a relatively young science. Airlines are credited with the invention of most of these techniques, especially Sabre, formerly with American Airlines (see the Service Operations Management Practices: Yield Management Increases Revenue $1 Billion/Year at American Airlines). However, the airlines did not develop most of these systems until a few years after the industry was deregulated in 1978, and the techniques only began to spread to other industries in the 1990s. A yield management system consists of three basic elements: 1. Overbooking (accepting more requests for service than can be provided) 2. Differential pricing to different customer groups 3. Capacity allocation among customer groups Each of these elements will be discussed in turn, then some practical implemen tation issues will be addressed.
OVERBOOKING
The need for overbooking is clear. Customers are fickle and do not always show up, so firms that overbook make far more money than those that dont. American Airlines
238
PA R T 4
estimated that their overbooking system garners them an additional $225 million in profit annually (Smith, Leimkuhler, and Darrow, 1992). If airlines did not overbook, planes that are now full would fly an average of 15% empty. No shows cost the worlds airlines $3 billion annually, even after efforts to minimize the revenue loss by overbooking (Cross, 1997, p. 146). No shows for restaurant reservations average about 10%, with some reporting 40% no shows dur ing the Christmas holidays. It has been reported that rental car no shows in the Florida market reached 70% of reservations. Of course, the alternative to overbook ing is to simply charge the customers whether they show up or not. Unfortunately, that approach failed in restaurants and auto rental businesses, and other businesses discarded it out of hand. Consumer resistance was high: Imagine missing your plane flight due to traffic, only to be told, The seat you paid for is in the sky, the ticket you have is worthless. The next flight out will cost you another $500, even though they have empty seats on that one. So the question for many businesses is not whether to overbook, but rather, how much to overbook.
CHAPTER 12
Yield Management
239
To demonstrate some mathematical methods to help determine the level of over booking, consider the following example.
The Hotel California found that it frequently turned down a customer in the lobby because a room was reserved for a customer who never showed up. The manager, felt that the hotels policy of overbooking should be examined. The average room rate was $50 per night, but the hotel could not collect the room rate from the no show customers. If no overbookings were allowed, each no show would in reality cost the hotel $50. If it overbooked too much and filled up early in the night, customers with reservations who arrived later to find no rooms available would be most unhappy. About 10% of those customers did not cost the hotel any money; they merely muttered menacingly and walked out. Another 10% were satis fied with being walked (or transferred) to another hotel at no cost to the Hotel California. The remaining guests were so upset by this situation that the hotel had to repair broken lobby furniture at a cost of $150. The hotels no show experience is summarized in Table 12.1. What should the overbooking policy be? We will discuss three approaches to answering that question.
240
PA R T 4
One way to put the relevant costs into the picture is to use the spreadsheet shown on Table 12.2 (This spreadsheet is also on the CD included with this text.) This spreadsheet calculates the expected cost for every possible scenario. For example, if no overbooking is done, then the column labeled 0 shows that on the 5% of days when there are zero no shows, theres no cost at all, but on the 10% of days when there is one no show, the cost is $50. The total cost at the bottom sums up 0.05($0) + 0.10($50) + . . . + 0.05($500) = $203. The overbooking level with the lowest expected cost is to overbook two rooms, with an expected cost of $137. The advantages of this method are that it incorporates relevant costs and can be spreadsheet based and fairly easy to figure out. Also, as will be seen shortly, if the costs and revenues are uncertain or not quite as easy to figure out as in the Hotel California example, then this method can be readily adapted. Two disadvan tages of this method, though, are that it requires accurate data and it is a brute force type of technique that does not increase a managers intuition about the problem.
7 $840 $720 $600 $480 $360 $240 $120 $ 0 $ 50 $100 $150 $405
8 $960 $840 $720 $600 $480 $360 $240 $120 $ 0 $ 50 $100 $500
9 $1,080 $ 960 $ 840 $ 720 $ 600 $ 480 $ 360 $ 240 $ 120 $ 0 $ 50 $ 603
10 $1,200 $1,080 $ 960 $ 840 $ 720 $ 600 $ 480 $ 360 $ 240 $ 120 $ 0 $ 714
CHAPTER 12
Yield Management
241
242
PA R T 4
10
15
20
25
Number of Dissatisfied Customers Cost shape assumed by equation (12.1) Actual cost curve shape
Dynamic Overbooking
The overbooking decision is often not a one time, static decision. Rather, it is a decision that is dynamic, which here just means that it changes over time. In a typ ical situation for a firm that takes reservations a long time in advance, the dynamic overbooking curve is shaped like Figure 12.2. When the event is still a long time in the future, the allowed number of overbook ings for it is at a peak. As the event nears, the number of allowed overbookings drops. This practice reaches its logical conclusion at the time the event takes place, when allowed overbookings drop to zero. For example, if three people are running to the plane for the final boarding and only two seats are available, the third person will not be sold a seat that cannot possibly be used.
A LLOCATING C APACITY
A difficult problem that afflicts many firms is allocating capacity among their cus tomer groups. That is, when to say no, were full to one customer while holding open capacity in hopes that a more profitable customer will arrive later. For airlines and hotels, especially, reservation activity follows Figure 12.3: High revenue cus tomers tend to make reservations fairly close to the event, while low revenue cus tomers often make reservations months in advance. In the airline industry, the price conscious vacationers make reservations months in advance, while the high paying business travelers may make reservations close to the event. For hotels, price conscious group business may make reservations a year in advance, while more highly profitable transient business may simply walk in the door that day. For the sake of simplicity, this chapter will focus on segmenting capacity between just two
CHAPTER 12
Yield Management
243
Reservations Start
customer groups. The real problem, however, is far more complex. A modern airline can have 10 different customer segments on the plane, each of which requires a capacity allocation decision (Cross, 1004). The situation depicted in Figure 12.3 is complex because one cannot simply let all reservations be taken first come, first served. Doing so would cause a firm to turn away a substantial portion of their most profitable customers while filling up on the less profitable ones. Also, most firms cannot simply say no to lower revenue busi ness, because they cannot fill their capacity solely with high revenue business. Consequently, the firm must decide ahead of time at what point to shut off the low revenue business in anticipation of the high revenue business booking later.
Number of Reservations
Reservations Start
244
PA R T 4
The general methods used to solve this problem can be classified as follows:
To demonstrate the differences in these methods, lets first describe an example problem.
EXAMPLE 12.2: Chancey Travel
Chancey Travel is offering a cruise to the Antarctic and wants to fill the 100 cabins avail able. Its primary market is Premium customers. Premium customers pay $12,000 for the trip, and the variable costs of serving these customers amounts to $2,000 per trip, as they are pampered endlessly and plied with Godiva chocolate three times a day. According to market analysis, the demand from Premium passengers is uniformly dis tributed between 51 and 100, which means a 2% chance that 51 Premiums will sign up, a 2% chance that 52 will go, and so on up to a 2% chance that the entire boat can be filled with 100 Premiums. This means that if the entire boat were reserved for Premiums an average of 75 Premiums would be on board. Premiums come from the idle rich, who tend to make their decisions at the last minute. Because the probability that the entire boat cannot be filled with Premiums is sub stantial, another market is sought. Market research discovered customer interest in a rough experience to Antarctica, but these consumers were more price sensitive. These Discount customers paid $2,500 for the trip, but the marginal cost of serving them was $0: heat was turned off to their cabins (hot water bottles were available for a proper deposit), and they received no food (they had to catch penguins with their own equipment). Demand was such that the ship could be entirely filled with Discounts, and these customers were willing to book far in advance. The dilemma for Chancey Travel is: How many cabins should be reserved in hopes of Premium customers? Several methods for arriving at an answer are discussed here.
C APACITY A LLOCATION A PPROACH 1: S TATIC M ETHODS : F IXED N UMBER , F IXED T IME RULES
We begin with static or one time decision rules. Basic fixed number and fixed time rules are the easiest to implement. A fixed time rule means simply that a firm will accept discount bookings until a specific date. No limit is set on the number of dis count sales. The motivation for this type of rule is the transparency to the customer and the ease of implementation, but it is clearly not close to being optimal. A fixed number rule for the problem may be to, say, reserve exactly 75 slots for Premium cus tomers and exactly 25 slots for Discount customers. The amounts reserved for each group can be viewed as a quota. Although a step up from the fixed time rule, it too pres ents certain problems. For example, if 75 Premiums and 20 Discounts are currently signed up, and a Premium customer wants to pay, this type of rule states that the Premium customer must be turned away because the Premium slots are already filled.
CHAPTER 12
Yield Management
245
are used. Lets still assume that, because an average of 75 Premium bookings is expected, we want to reserve 75 slots for Premiums. Instead of the remaining 25 slots being given to Discounts, the remaining 25 are sold on a first come, first served basis. In this manner, the 75 slots for Premiums can be thought of as a protection level; that is, at least 75 rooms are protected just for Premiums, but they can get more. In this nested system other groups are not allowed into the nest of 75 pro tected rooms, but the protected group can venture outside that number. Our use of 75 as a hypothetical number for the rooms protected for the Premiums does not offer the best strategy, however. To show how some basic calculations can lead to a better strategy, the expected marginal seat revenue (EMSR) heuristic is introduced here (Belobaba, 1989). The EMSR heuristic provides a basic logic for determining how much protection to give different customer levels. The EMSR heuris tic is no longer widely used in industry. However, the methods that are used currently in industry are both highly complex and proprietary. The EMSR heuristic allocates capacity one unit at a time. For this problem, allo cating the 1st through 51st rooms is fairly simplethey should all go to Premiums, since we are certain that at least 51 Premiums will show up. For deciding whether to protect the 52nd room we compare the expected marginal revenue from each group and assign the room to the group that provides the most expected revenue. For allo cating the 52nd room, we are 98% certain a Premium customer will request it, so the calculations are as follows: Premium: 98%($12,000 $2,000) = $9,800 expected revenue Discount: 100%($2,500) = $2,500 expected revenue Because $9,800 > $2,500, the room is reserved for Premiums. For allocating the 53rd room, the following calculation applies: Premium: 96%($10,000) = $9,600 expected revenue Discount: $2,500 expected revenue Again the room is reserved for Premiums, because $9,600 > $2,500. This process continues until we reach the 88th seat: Premium: 24% ($10,000) = $2,400 expected revenue Discount: $2,500 expected revenue Because $2,400 < $2,500, the protection level for Premiums stops at 87. The EMSR heuristic therefore states that we should allocate 87 rooms Premium, 13 rooms Discount. Note, however, that on an average boat, this allocation would result in 75 Premium passengers, 13 Discount passengers, and 12 empty rooms. This result explains why, when one is allocating capacity as best as one can, it frequently results in unused capacity. This procedure can be summarized in a rule similar to the overbooking formula given in equation (12.1) earlier. Set the protection level of Premiums at the smallest number where: P(Premium demand Protection level) Discount revenue/Premium revenue (12.2) Here, Discount revenue/Premium revenue = $2,500/($12,000 $2,000) = 0.25, and there is a 26% chance that Premium demand is greater than 87. To see how to use this from a different perspective, consider Premium demand for Chancey Travel to be normally distributed, with a mean of 70 and a standard
246
PA R T 4
deviation of 20. Lets make Discount revenue $3,000, instead of $2,500. So, Discount revenue/Premium revenue = 0.3, and we are searching for the 30th per centile of demand. This is a one tailed look up of an area of the standard normal distribution table (located at the back of the book). So one desires to look up 0.5 0.3 or 0.2 in the body of the table, which yields a z score of 0.515. Applying this z score to the prob lem data, mean = 70, std dev = 20, 70+0.51520=80 seats should be protected for the Premium passengers. This approach is not limited to just two customer categories. The protection level for each customer category can be calculated by comparing the revenue associated with each successive customer class.
CHAPTER 12
Yield Management
247
Medium
Low
10
15
20
25
30
35
40
every day. The database for this forecast requires 2 terabytes of data (Cross, 2004). Unfortunately, such data do not exist for many firms. Also, events unknown to any computer model may cause reservation activity to suddenly drop or pick up, such as competitor pricing changes, new competitors coming on line, or such things as sport ing events and conventions being scheduled where they did not exist before. Because of these difficulties, computer models help, but the final decisions on capacity allo cation for many industries is still manual, performed by managers with an intuitive sense of the risks.
P RICING
Thus far we assumed prices are exogenous, or outside of our control, which of course is not the case. This section discusses the difficult topic of setting prices in a yield management system. The general idea in yield management is to break up a market into a number of different customer segments and charge different prices for each segment. Figure 12.5 shows the traditional supply and demand curves seen in every introductory economics class.
248
PA R T 4
The market clearing price creates consumers surplus for customers who would have been willing to pay more and economic rent for suppliers whose costs are lower. The idea of yield management is to segment different customer categories with separate market clearing prices, so that high prices are charged to those who would be willing to pay them and low prices are charged to those who would ordi narily not use the service if the single market clearing price were in place. Figure 12.6 shows a general representation of three markets for airline seats. Customer class 1 (first class passengers) wants premium service and flexibility, and cares little about cost; customer class 2 (business class) values the flexibility of making last minute changes over price; customer class 3 (often vacationing families) makes plans well in advance and are highly sensitive to price. The consumer surplus from each group is the area formed by the triangle of the market price (dotted line), demand curve, and the vertical line extending from the dotted line to the customer class bar. Two important economic effects come into play here: Large portions of con sumer surplus are shifted to the supplier, and bringing on customers who would not ordinarily use the service can expand the market itself. The magnitude of pricing differences can be enormous. Table 12.4 shows a snapshot of some airline fares. Here, premium first class and business class are not consideredjust full coach fare versus discounts on that same coach seatand the cheapest prices are 11% to 13% of the cost of a full coach fare.
FIGURE 12.6: Supply and Demand Equilibrium in Yield Management
CHAPTER 12
Yield Management
249
Several difficulties can arise in creating these customer segments. Foremost, price discrimination, per se, is illegal in the United States. A business cannot simply tell some customers they must pay more for a good or service than others, just because the business knows that a customer will pay. Some types of customer segmentation are illegal in the United States, as well, such as segmenting customers according to race or ethnic origin and charging differential prices. This regulation does not apply in some other countries, however, and ethnic pricing of airfares is not an uncom mon practice (Mitchener, 1997). Consequently, it is up to the imagination of the marketing department to deter mine a legal and enforceable method for segmenting markets. For example, airlines want to charge higher prices to price insensitive business customers and lower prices to price sensitive vacationers. However, these markets can only be attacked indi rectly, by segmenting pricing on how far in advance one makes a reservation. The yield management issues of overbooking and capacity allocation lend them selves well to numerical analysis once pricing is set, but determining the best overall combination of prices, capacity allocation, and overbooking is difficult. In practice, marketing usually sets prices in coordination with company policy and competitive response, and the operations area sets capacity allocation numbers after pricing is determined. To show why pricing is so tricky in the yield management environment, consider the following example.
EXAMPLE 12.3: Pricing and Capacity Allocation
Consider a one time event in which, miraculously, perfect economic information is provided on two groups of customers: Premium and Discount. For the Premium cus tomers, the more that is charged, the fewer will attend, according to Table 12.5. Consider three scenarios: (1) a service facility not facing a capacity constraint; (2) a capacity of only 100 customers, and Discount customers pay $50; and (3) again a capacity of 100 customers, but Discount customers pay $75. In scenario (1), the best price to charge premium customers is $90, resulting in a total of $10,800 in rev enue, based on traditional economic logic. In a situation in which demand responds disproportionately to price cuts, prices should be low (high price elasticity, in eco nomic speak). Capacity restrictions and different customer classes, however, turn the traditional economics of price elasticity on its head. In scenario (2), high price elas ticity still remains with the Premium customers, but one cannot take advantage of it, because the sheer number of potential customers overwhelms capacity. The best price to charge Premiums is $100, resulting in $10,000 revenue from dedicating the facility just to Premium customers and ignoring the Discount market. However, if the Discounts pay a bit more, as in scenario (3), the best price to charge Premiums is $110, resulting in more revenue by adding Discount customers.
250
PA R T 4
$10,000
Scenario 2: Capacity of 100, Discount class unlimited demand at $50* Premium price Premium demand Premium revenue Discount revenue Total revenue $ 100 100 110 80 8,800 1,000 9,800 90 100 9,000 0 9,000
$10,000 $ 0 $10,000
Scenario 3: Capacity of 100, Discount class unlimited demand at $75* Premium price Premium demand Premium revenue Discount revenue Total revenue
*Optimal solution in blue.
100 100
$10,000 $ 0 $10,000
The overall lesson for pricing in yield management environments: Traditional reasoning regarding price elasticity does not apply, and pricing depends on the rel ative demand/capacity relationships that must be judged on an individual case by case basis.
I MPLEMENTATION I SSUES
Although yield management systems enjoy success in a number of different indus tries, some practical problems must be addressed in both customer relations and employee relations.
Alienating Customers
The chapter began with some examples of yield management practices that can alien ate customers. From the customers viewpoint, many of the rules for getting into a spe cific fare class seem ridiculous. The idea that they are told a service is full to capacity when they then find out it is not is unbelievable to many consumers. Further, pricing issues can create significant ill will among customers. The general public seems to accept the idea that one person may pay a different rate for an airline seat than another, although many people still get quite angry when they find out. That feeling, however, has not spread to other industries. If someone overhears the person in front of her in line getting a better room rate at a hotel, she will often demand that rate, even
CHAPTER 12
Yield Management
251
though the rate of her own customer class would have been perfectly fine with her before she knew about the better rate. A key in many industries isto be bluntto keep customers ignorant of the rates available to other customer classes.
Employee Empowerment
Strategic decisions must be made concerning how much power is given to employ ees to make decisions. If a system is seen as reducing employee discretion, employ ees can often make up for this perceived indignity by sabotaging the system. In the hotel industry, for example, a significant degree of responsibility for assigning the proper customer class and room rate resides in the clerk who sits at the front desk. If a system lacks enough structure, clerks may whimsically decide whether a cus tomer gets a higher or lower rate depending on their mood rather than the cus tomers true class. However, a system with too much structure can lead to a loss of bargaining power by clerks, which can result in needlessly empty rooms. Consequently, it is important for anyone with such responsibility to understand the overall managerial goals of such a system.
252
PA R T 4
Summary
Yield management systems are used in a wide variety of industries that have limited capacity. Such systems can potentially make a significant difference in profitability for firms that use them well. Three basic components of a yield management system are overbooking, capac ity allocation, and pricing. For overbooking and capacity allocation, numerical meth ods presented in the chapter can help to solve those problems. However, a practical answer to the pricing problem still remains out of reach. Examples were used to show how the environment of a yield management system, with customer classes and capacity restrictions, can turn normal pricing decisions around. Finally, a yield management system is not just a computer program. It is a sys tem that must be implemented with flesh and blood employees and customers. These human elements of the system must be attended to carefully because poor employee or customer education as to the limits and nature of such a system may lead to its circumvention in a variety of ways.
Review Questions
1. A specific problem of yield management techniques noted in the chapter demonstrated that an optimal solution to a particular problem involving a 100 seat plane gave an average of 13 seats to discounters, 75 seats to pre mium passengers, and an average of 12 seats were unoccupied. Explain in words, not numbers, how it is possible for the best solution to leave seats unoccupied on average. 2. If the average number of no shows are, for example, 12, why isnt the best overbooking number necessarily 12? 3. What are nested capacity allocation methods and why are they used? 4. Why cant the standard formula for overbooking be used in all cases?
Problems
12.1. The Greybeard Busing Company is assessing its overbooking policy for the MiamiFort Myers run. The number of customers who dont show up after reserving a ticket is uniformly distributed from 1 to 10 (10% chance of 1 no show, 10% chance of 2, etc.). Tickets cost $25, and if the particular bus run is full, a passenger with a reservation is given passage on a rival bus line, a cost of $60. What should Greybeards overbooking policy be? 12.2. Its normal for an airline to overbook a flight by 20% or so, but Amtrak will overbook its long distance trains only by 5% to 10%, even though they have even more no shows than the airlines do. What reasons would Amtrak have for overbooking so little? 12.3. The Kaluauluauhala Resort rents weekly condominiums by the Hawaiian shore, nearly all to families from Japan and the U.S. mainland. It is the only resort on the west side of the island. These condos rent for $1,500 per week, and guests typically spend another $500 (in terms of profit for the firm) dur ing their stay. A guest who doesnt show up is subject to a cancellation fee of $750. If no condo is available for a guest with a reservation, he has to be
CHAPTER 12
Yield Management
253
transported either to the other side of the island or to another island to a sim ilar resort and Kaluauluauhala picks up half the cost of the other resort (about the same as Kaluauluauhalas price, so the customer receives a half price vacation) and the average $200 transportation charge. No shows by customers average three each week, evenly distributed between 0 and 6. What should the resorts overbooking policy be? 12.4. Consultant Air focuses business on high paying McKinsey consultants, but to fill planes, it also carries the general public. It is flying a 100 seat jet from Atlanta to San Francisco and the consultant demand is normally distributed with mean 60 and standard deviation 10. Consultants pay $1,200 per ticket, the general public pays $150. Of course, the general public must book a flight several weeks ahead of time, while the consultants book flights at the last minute. How many seats should be opened up for the general public? On average, assuming 0 no shows, how full will the plane be under this plan? 12.5 In addition to the customer classes in the previous problem, Consultant Air added a third class, normal business people, who pay $500 per ticket. Demand from these customers is normally distributed with mean 20 and standard deviation 5. How does this class of customers change the solution to the previous problem? 12.6. Assume the curves shown in Figure 12.7 represent only the highest, lowest, and average demand for full fare flights. Other curves are also possible. You are now four weeks before the flight and have received 30 reservations. Assuming a 150 seat plane, what should your capacity allocation policy be? 12.7. The curves for problem 12.6 really represent averages rather than exact rela tionships. If the final boarding numbers represented by the curve are an aver age result, where the actual boarding passengers are normally distributed about that average with a standard deviation of 10, what should your nested capacity allocation policy be? Assume prices of $1,000 per seat for premium, $400 for discount.
254
PA R T 4
12.8. Rene was adamant. If we have to tell customers we overbooked and theres no seating available, they are going to be extremely angry. Ill bet it costs us $150 per couple in goodwill each time. Tom replied, No way, it doesnt cost us a penny. Were so popular now it doesnt matter if one potential customer doesnt come back. What matters is getting the most profit out of each seat ing. We prep everything ahead of time, so it costs us $50 per couple for food whether they show up or not, and we charge $150 per couple. At that rate of profit I dont want to see a single empty table. The restaurant Chez Rogat was run by two Owen grads. They had two seatings each night, 6:00 P.M. and 8:30 P.M., for their 30 table restaurant. They prepared the same meal for everyone at each seating. Due to their popular ity, they always had more reservations than they could handle and reserva tions were made far in advance. They had few no shows, but did have some. They had about an equal chance of either 0, 1, 2, or 3 couples as no shows at any given seating. Given their costs, calculate an overbooking policy for Chez Rogat. Would you recommend any other capacity management strategy?
Selected Bibliography
Belobaba, P. 1989. Application of a Probabilistic Decision Model to Airline Seat Inventory Control. Operations Research, 37(2), 183197. Cook, T. 1998. SABRE Soars. OR/MS Today (June), 2631. Crandall, R., and R. Markland. 1996. Demand Management: Todays Challenge for Service Industries. Production and Operations Management, 5(2), 106120. Cross, R. 1997. Revenue Management. Broadway Books, NY. Cross, R. 2004. February 10 presentation at Emory University. Mitchener, B. 1997. Ethnic Pricing Means Unfair Air Fares. The Wall Street Journal (December 5), B1. Smith, B., J. Leimkuhler, and R. Darrow. 1992. Yield Management at American Airlines, Interfaces, 22(1), 831.
CHAPTER 12
Yield Management
255
CASE STUDY
39 40
33 40
42 56
37 51
34 38
18 29
17 23
13 15
11 14
10 14
7 12
7 8
5 7
4 7
256
PA R T 4
CASE STUDY
Obs. Boarded 1 87 2 81 3 65 4 57
G G G
39 40
100 105
101 111
104 114
85 93
71 79
57 65
44 50
32 32
21 19
Unfortunately, even with his new pricing scheme he was still not able to attain profitability, but a solution became clear. Since his 747 was never full, he negotiated a change in his lease for a much cheaper and more fuel efficient 100 seat plane that would allow him to become profitable. At this point, however, he needs assistance. He could fill his planes with deep dis count and discount flyers, but it would be unprofitable to do so. Further, he had no experience with overbooking policies and was unsure whether he wanted to pursue one. Consequently, Al needs expert service operations consulting to maximize revenue.
Seating Allocation
Seating allocation must come in a nested form. For example, in Table 12.8 no deep discount, up to 50 discount and deep discount, and up to a combined 120 full coach, discount and deep discount seats are set aside six months prior to departure. Implicitly, for a 100 seat plane this allocation contains 20 overbooked seats. Demand history from 40 flights on the larger planes is contained on Tables 12.6 and 12.7 and is available on the text CD. For example, in the Full Coach Fare History, Observation 1 indicates that 5 reservations were obtained 6 months before the flight, 6 reservations obtained 5 months before the flight, and so on, and of the 53 reserva tions in the system the week before the flight, 34 actually boarded the plane. Demand history for the deep discount tickets is not necessary because 100% of them are taken as soon as they are offered. The demand history given can be used to determine threshold curves for each class of passenger. The purpose of developing threshold curves is to predict the even tual number of reservations just prior to boarding. The costs involved in overbooking depend on the number of people not seated for the flight. The revenue from the ticket sales is subtracted (for simplicity, assume revenue of $1,000 for each person not seated), and Al estimates the per passenger
CHAPTER 12
Yield Management
257
CASE STUDY
TABLE 12.8: Game Tracking Sheet
Flight Group Name: Weeks Away from Takeoff 24 Price Full Discount Deep Discount Price Effects on Reservations Full Discount Policy Full Discount Deep Discount Booking Revenues Full Discount Deep Discount Revenue Overbooking Penalty Profit 1,000 400 100 20 16 12 8 7 6 5 4 3 2 1
penalty cost of leaving passengers stranded as $200 times the number stranded squared. So leaving one person stranded costs $200, two people costs $800, three people costs $1,800, and so on.
Pricing
Currently, ticket prices start at $1,000, $400, and $100 for the three fare classes. Prices can be raised by 10% to $1,100 or $440, lowering demand by 15%, or prices can be lowered by 10% to $900 or $360, increasing demand by 15%. Prices on deep discount tickets cannot be changed.
Assignment
As a consultant, you advise Al Niemi concerning overbooking, seating allocation, and pricing for the three classes of passengers. Your group must indicate a policy prior to receiving any reservations. Once all policies are in, the reservation informa tion for the six month out time frame will be given. At this point, you may alter your policies. Iterations of receiving reservation information and altering policies proceeds until plane departure, as depicted in Table 12.8.
EXAMPLE
120 50 0