Discussion Paper Advances in Airline Pricing, Revenue Management, and Distribution

Transcription

1 Discussion Paper Advances in Airline Pricing, Revenue Management, and Distribution Implications for the Airline Industry Prepared for ATPCO by PODS Research LLC Peter P. Belobaba William G. Brunger Michael D. Wittman October 2017

2 Executive Summary The creation of an airline fare quote for a specific itinerary at any given moment is the result of decades of scientific development in pricing and revenue management (RM), complex information technology, and hard-earned practical experience. In the past several decades, airlines have served as leaders in making new advancements in pricing and RM, as well as developing new types of products to offer to their customers. In 2017, the state of airline pricing, revenue management, and distribution is at a notable inflection point. Airlines are creating increasingly numerous and complicated product offerings, and distributing them on an ever-growing variety of channels. Moreover, technological advancements are giving airlines access to more information about the characteristics of their customers. At the same time, next-generation pricing, revenue management, and distribution mechanisms are currently under development within the airline industry. If implemented, these mechanisms could have wide-ranging implications for airline revenues, competition, internal processes, and external interactions with customers and regulators. However, the mechanisms themselves and their potential effects remain poorly defined and are often misunderstood. In this discussion paper, we explore advances in airline pricing, revenue management, and distribution, beginning with the development of early RM systems following airline deregulation and ending with the development of next-generation pricing capabilities. We do not endeavor to prescribe a single pricing or revenue management approach that is best for all airlines, but rather to present a full spectrum of technologies, implications, and viewpoints to enable each airline to evaluate the extent to which investment in next-generation pricing is appropriate for its unique position and corporate strategy. In this executive summary, we present the high-level findings from our review. Current airline pricing and revenue management practices evolved in response to technological constraints and industry structures Airline pricing and revenue management has a rich history dating back to the deregulation of the U.S. airline industry in With more freedom to vary the fares charged on each flight, airlines began developing scientific techniques to adjust prices based on the remaining capacity on each flight and a forecast of future demand-to-come. Later advancements in revenue management would expand optimization approaches to full networks of flights through O-D control, and to less-restricted fare structures through fare adjustment and hybrid forecasting. In the past four decades, the filed fare class and automated fare quote process has remained central to airline pricing, revenue management, and distribution. As they begin to sell more complex products, airlines have started to run into limitations with current systems, which allow up to 26 reservation booking designators or RBDs in each market. New distribution technology is beginning to emerge to give airlines more control over the information they communicate to the marketplace, which has started to enable the development of next-generation pricing mechanisms. 1

3 The sophistication of airline pricing is best-in-class among travel-related industries, but other industries have started to develop more complex pricing mechanisms In the 1980s and 1990s, airlines were pioneers in developing sophisticated pricing and revenue management methods that were well-suited for an industry with limited, perishable inventories. We call these practices assortment optimization, since they involve selecting prices for a given product from a relatively small and finite assortment of pre-defined price points. Hotels, passenger rail operators, and rental car agencies have all experimented with pricing and RM practices adapted directly from those used in the airline industry. Unlike airlines, many of these industries are not required to publicly file the price structures they use for assortment optimization, and may not be limited by the same technological constraints faced by airlines. In other industries without these constraints, new pricing mechanisms have started to emerge. These include dynamic price adjustment, in which a firm applies increments or decrements in certain situations relative to a pre-defined price point. For example, the online retailer Amazon.com often uses dynamic price adjustment to mark down the prices of books from the manufacturer s list price. Another advanced pricing mechanism is continuous pricing, where firms freely select prices from a continuous range of possible values instead of from a small set of previously-defined options. At the limit, prices could be computed uniquely for each transaction. The ride-hailing app Uber uses transactional continuous pricing, since it computes fares for each ride individually and can change these prices from moment to moment in response to market demand. Next-generation mechanisms are already under development to advance the state of pricing and revenue management in the airline industry The airline industry has begun to develop new mechanisms for pricing and RM to advance the state of the practice from assortment optimization toward transactional continuous pricing. These new mechanisms generally aim to increase the number of price points available in any given market; to increase the velocity at which these price points are updated; and/or to increase the frequency at which prices are changed from transaction to transaction. These next-generation mechanisms include: More frequent updating of fare structures, typically through automation technologies to file fares more rapidly with ATPCO. With these technologies, each airline could create unique fare structures for each market for each departure day; Dynamic availability of fare products, in which the RM availability of fare products could be adjusted for specific customers or in specific situations; Additional RBD capabilities, which could increase the current limit of 26 possible price points available to airlines in each market; Dynamic pricing engines, which apply dynamic price adjustments (increments or discounts) to filed fares in certain situations; Continuous pricing, in which each airline would select prices from a continuous range of possible values instead of from a small number of pre-filed price points; and Dynamic offer generation, which merges the product creation process and the price 2

4 selection process into a single step. An airline would dynamically create and price bundles of itineraries and ancillary services, potentially at a transactional level. Continuous pricing and dynamic offer generation would require significant changes to existing revenue management optimization and forecasting practices, as well as to distribution and IT capabilities. Airlines deciding to implement next-generation pricing will likely proceed incrementally to ensure that these mechanisms are compatible with existing practices and technologies in the short term. Next-generation pricing mechanisms have the potential to increase airline revenues, but also will disrupt the status quo An airline will only choose to practice next-generation pricing if these mechanisms have the potential to increase revenues. Studies of several next-generation pricing mechanisms have suggested revenue gains are possible, either through an increase in yield from offering higher price levels in certain situations, or through stimulation of new demand by offering discounts to certain customers. However, the implementation of next-generation pricing in the airline industry is likely to disrupt the status quo and could increase the incentive for an airline to engage in discounting until it reaches its marginal cost. There may also be implications of next-generation pricing for consumers and regulation. Consumers would likely react negatively if transactional pricing mechanisms resulted in an identified individual receiving a higher price than an anonymous shopper. There is also the possibility that customers would try to game the pricing engines to obtain prices for which they are not eligible. Furthermore, while regulators have to date been relatively permissive of new pricing technologies that better match capacity with demand, some new pricing mechanisms could trigger regulatory reviews with uncertain outcomes. Different viewpoints exist on whether next-generation pricing will be a net benefit for the airline industry Given the wide range of implications next-generation pricing will have on the airline industry, stakeholders have divergent views on the prudence and feasibility of moving beyond existing mechanisms. We present different viewpoints on the potential rewards and risks of nextgeneration pricing, reflecting the wide spectrum of views that appear to be emerging in the industry. These positions do not necessarily reflect the viewpoints of the authors of this study, ATPCO, or any particular airline. One common viewpoint sees next-generation pricing as a chance to break free from existing technological constraints, giving airlines more control over the prices and products they offer their customers. From this perspective, the potential revenue gains from next-generation pricing outweigh the potential for instability. Markets for information will emerge to enable airlines to make rational pricing decisions, and first-mover advantages will exist for airlines that are first to implement next-generation capabilities. An opposing view proposes that the inherent risks of next-generation pricing are too high to justify its development. From this perspective, next-generation mechanisms are seen as an unnecessarily disruptive force to the status quo. New types of pricing could lead to instability in the marketplace and increase incentives for airlines to discount. 3

5 Stakeholders at each airline will have their own views on next-generation pricing that fall somewhere between these two extreme viewpoints. As mentioned previously, development of next-generation pricing mechanisms will likely proceed incrementally, both to avoid disruptions with existing legacy practices and to test the effects of these systems in a limited implementation. The airline industry can take several steps to start preparing for a world of nextgeneration pricing and revenue management There are airlines that will begin (or have already begun) experimenting with next-generation pricing mechanisms in the near future. The airline industry as a whole can take several steps towards managing this transition. These steps include more investigation and research into forecasting and optimization methods centered around customers conditional willingness-to-pay, which will likely be required by many next-generation techniques. Airlines and technology providers will also need to consider how to develop systems and data sources that can function with existing technologies, diverse pricing goals, and practices like interline pricing, corporate contracts, and group sales. Core competencies will also need to be developed among pricing, revenue management, and distribution professionals to use and tune next-generation pricing mechanisms. Next-generation pricing will not only require fluency with new systems and automated technologies, but also an understanding of the complexity of pricing in a new competitive environment. At the same time, the accuracy of the fare calculation process and the ability to meet regulatory requirements will have to be maintained. Conclusions The development of next-generation pricing mechanisms represents an exciting and uneasy time for the airline industry. There still exists significant uncertainty regarding what these new mechanisms will look like, how they will be implemented, and what impacts they will have on the status quo. One certainty is that the landscape of airline pricing, revenue management, and distribution will look much different in five years than it does today. This paper aims to provide some common definitions and concepts to guide this evolution, and to help airlines and technology vendors prepare strategies for transitioning to this new commercial environment. 4

6 Terms and Definitions Assortment Optimization Conditional Willingness-to-Pay Continuous Pricing Customized Pricing Dynamic Availability Dynamic Offer Generation Dynamic Price Adjustment Dynamic Pricing Engine Maximum Willingness-to-Pay Price Selection Mechanism Product Creation Mechanism Revenue Management (RM) Transactional Assortment Optimization Transactional Continuous Pricing A price selection mechanism in which firms first define a finite set of possible price points for a given product, then select prices from among that set to display to customers. The maximum amount a customer would be willing to pay for a product, conditional on the other competing alternatives available for the customer to purchase. A price selection mechanism in which firms select prices from among a continuous range of possible price points. A pricing mechanism in which customers are divided into two or more segments, each of which may receive a unique price/product offering. A next-generation price selection mechanism where airlines make adjustments to the revenue management availability of fare products, depending on characteristics of the booking request or of the customer. A next-generation mechanism in which airlines perform product creation and price selection simultaneously, to offer bundles of itineraries and ancillary services. A price selection mechanism in which firms first select a price from a finite set of possible price points, as in assortment optimization, and then increment or decrement that price in certain situations. A next-generation dynamic price adjustment mechanism in which airlines increment or decrement the filed fare for an available fare product in certain situations. The maximum amount a customer would be willing to pay for a product if there were no other alternatives available for the customer to purchase. The means by which a firm chooses a price to charge for a given product. The means by which a firm creates one or more product offerings. The process of managing inventory of a finite, perishable resource as a function of remaining capacity, the time remaining in the selling period, and a forecast of demand. The selection of prices through assortment optimization based on the characteristics of each individual transaction. The selection of prices through continuous pricing based on the characteristics of each individual transaction. 5

7 Table of Contents Executive Summary... 1 Terms and Definitions... 5 Introduction Evolution of Airline Pricing, Revenue Management and Distribution...10 Economic Fundamentals of Airline Pricing Less Restricted Fare Structures and New RM Developments Willingness to Pay Interpretation and Estimation Fare Quotes, Distribution and Competition A Brief History of Airline Price Competition Recent Trends in Airline Pricing Development of a New Distribution Capability Definitional Framework for Price Selection Mechanisms...29 Assortment Optimization Dynamic Price Adjustment Continuous Pricing Conclusions Summing up the Definitional Framework Pricing and Revenue Management in Other Industries...35 Travel-Related Industries Practicing Assortment Optimization Hotels Passenger Rail Transportation Rental Cars Conclusions: Travel-Related Industries Practicing Assortment Optimization Firms Practicing Dynamic Price Adjustment and Continuous Pricing: Amazon, Airbnb, and Uber 42 Amazon.com Dynamic Price Adjustment in Online Retail Peer-to-Peer Accommodation Services: The Case of Airbnb On-Demand Ride Hailing Services: Uber and Lyft Conclusions Next-Generation Pricing Mechanisms for the Airline Industry...48 More Frequent Updating of Price Points Dynamic Availability of Fare Products Additional RBD Capabilities Dynamic Pricing Engines Continuous Pricing Dynamic Offer Generation

8 Conclusions Implications of Next-Generation Pricing Mechanisms for the Airline Industry...55 Implications for Airline Revenues Implications for Airline Competition Implications for Airline Processes Implications for Airline Consumers Implications for Airline Regulation Conclusions Summary and Conclusions

9 Introduction The development of new technologies for airline distribution, including IATA s New Distribution Capability (NDC), has the potential to transform the way in which airlines price, revenue manage and distribute their products and services to consumers. After decades of relying on legacy systems and dated industry standards, airlines today face the challenge of understanding the implications of impending changes that could not only lead to substantial impacts on pricing and marketing practices, but which are also likely to affect competitive strategies, revenues and profits. It is the objective of this discussion paper to provide the many stakeholders involved in the airline pricing and distribution process with an overview of current pricing and revenue management practices, and how they might be affected by the movement by airlines toward next-generation dynamic pricing and offer generation made possible by NDC. The possible implications of different future mechanisms for airline pricing are described, and different viewpoints as to the benefits and risks of these new developments are presented. We begin in Chapter 1 with a brief review of the economic fundamentals of airline pricing and the evolution of increasingly sophisticated revenue management practices in response to the removal of regulatory controls over airline pricing in many parts of the world. The impacts of the past and current pricing environments on both airline competition and price transparency to consumers are discussed, to provide a foundation for the introduction of emerging new distribution technologies and assessment of their potential impacts. In Chapter 2, we provide a three-part definitional framework to classify the pricing practices used by airlines and in other industries. With assortment optimization, firms select prices from a pre-defined menu of a relatively small number of possible price points. With dynamic price adjustment, firms also begin by selecting from a pre-defined menu of possible price points, and then increment or discount the selected price for specific customers or in specific situations. With continuous pricing, there is no finite menu of possible price points, and firms freely select prices from a continuous range of possible values. These definitions provide a foundation for classifying pricing practices in the remainder of the document. Chapter 3 reviews advanced pricing practices used in a variety of industries, and compares these practices to those used in the airline industry. We find that many travel-related industries, such as hotels, passenger rail, and rental cars, use pricing and revenue management practices that are either inspired by or directly copied from those used in the airline industry. And, some technologically-advanced firms like Airbnb and Uber practice sophisticated continuous pricing techniques where prices can change from transaction to transaction. In Chapter 4, we describe six next-generation pricing mechanisms that are currently under development in the airline industry. Some of these mechanisms, such as dynamic availability and more frequent filing of fares, aim to expand pricing capabilities within the confines of legacy systems and technologies. Other mechanisms, like dual-character RBDs and dynamic pricing engines, would give airlines more pricing flexibility but would also require changes to legacy systems, processes, and messages. The most advanced of these next-generation 8

10 technologies dynamic offer generation would require a significant reworking of most airline commercial processes and would most likely raise compatibility issues with legacy systems. If implemented, the next-generation pricing mechanisms described in Chapter 4 could have significant impacts on the airline industry. Simulation studies have suggested that nextgeneration pricing technologies have the potential to increase airline revenues, but some industry observers worry that the technologies could lead to adverse effects on airline competition, consumers, legacy processes, and regulation. In Chapter 5, we review each of these potential implications and discuss how new pricing methods could change the industry. Finally, in Chapter 6, we summarize our findings by discussing the possible rewards and risks of next-generation pricing mechanisms in the airline industry. While next-generation pricing can be seen as an opportunity to break free of legacy constraints that have constricted the industry s commercial capabilities, the development of new pricing mechanisms also could bring shortterm commercial risks. We close by describing some steps that an airline could take to prepare for a future airline industry landscape in which next-generation pricing is commonplace. 9

11 1. Evolution of Airline Pricing, Revenue Management and Distribution Having made longer-term strategic decisions about their fleet composition and network structure, airlines rely on three tactical processes to maximize profits scheduling, pricing and revenue management. The scheduling process involves interrelated decisions with respect to the frequency of flights, their timetables and the size of aircraft for each flight. Pricing and revenue management decisions with respect to fare structures and seat inventory control, respectively, have the goal of maximizing revenues given the commitment to operate a schedule of flights. An airline s pricing department establishes a fare structure consisting of various fare products, with different prices, service amenities and restrictions, to offer in each origin-destination market. Revenue Management (RM) then determines how many seats to make available at each fare level. The actual fare quote for a travel request is then delivered to consumers via a distribution channel that combines the prices and conditions of the fare structure with the availability of seats calculated by RM. By providing a standardized messaging framework, the emerging capabilities of the New Distribution Capability (NDC) will allow participating airlines to calculate fare quotes for travel requests internally, prior to the distribution step. This new capability has the potential to dramatically change the current airline pricing, RM and fare quote processes. In this chapter, we provide an overview of the development of airline pricing, revenue management and distribution over the past several decades. We begin with a look at the underlying economic fundamentals that have guided airline pricing since deregulation of airline markets. Subsequent sections then describe the evolution of airline fare structures and airline RM systems, including simplified fare structures as well as more recent trends towards both bundling and a la carte pricing. Economic Fundamentals of Airline Pricing Airline fares are typically set for travel in an origin-destination (O-D) market. That is, airline prices are established for travel between origination point A and destination point C, where A-C is the relevant market. Airline prices for travel A-C can depend on the volume and characteristics of the demand between A and C (e.g., trip purpose and price elasticity of demand), as well as the airline supply between A and C (frequency and path quality of flights) and the competitive environment in that market (number and type of airline competitors). Prior to airline deregulation in the U.S. in 1978, airfares were controlled by the Civil Aeronautics Board (CAB). The CAB used a mileage-based formula to set fares based on distance traveled, irrespective of differences in demand volumes, competition or operating costs in different O-D markets. Airlines primarily offered First Class and Economy fare (coach or tourist class) products, both of which were tied to the distance-based fare formula. Similar regulatory regimes were in place in countries around the globe, with governments exerting control not only over airline prices but over route access, as well as frequency and capacity decisions of airlines. International services between countries were also heavily regulated, subject to negotiations of bilateral air service agreements specifying the carriers, routes, frequencies and prices allowed. 10

12 With the US leading the way in deregulation of domestic airline markets, deregulation or at least liberalization of airline competition has spread to virtually every global region. For international services, open skies agreements in many cases have also given airlines increased freedom to make their own scheduling and pricing decisions. However, some international markets are still regulated with respect to traffic rights, capacity and pricing where fares are still subject to government review and approval. Even in deregulated markets, airlines can still be required to file their passenger tariffs with a government agency and other regulations continue to have an impact on airline pricing. For example, U.S. airlines must provide passengers with full refunds for cancellations made within 24 hours of booking or allow them to hold a reservation without payment for 24 hours if the reservation was made seven or more days in advance. More recent regulations require airlines to provide full disclosure of ancillary fees at the time of booking. Nonetheless, with the shift to liberalized airline pricing by most countries, it is now common for different O-D markets to have prices not related to distance traveled or operating costs in each market, as airlines respond to increased competition in an effort to maintain their market presence and share of traffic. Given these new freedoms to set prices, airlines rapidly embraced the concepts of differential pricing, in which different prices are offered not only for different physical products (e.g., First, Business and Economy cabins), but also for identical seats and services within the same cabin. Differential pricing involves the use of both price discrimination and product differentiation. Price discrimination is the practice of charging different prices for the same (or very similar) products that have the same costs of production, based solely on different consumers willingness to pay. 1 On the other hand, product differentiation involves charging different prices for products with different quality of service characteristics and therefore different costs of production. 2 The effective use of differential pricing requires the airline to identify different demand segments. In theory, total revenues would be maximized if the airline were to charge a different price for each passenger based on his or her willingness to pay (WTP). Such a theoretical segmentation is difficult to achieve as airlines cannot determine each individual s WTP for a given trip, nor can they publish different fares available only to specific individuals. Instead, airlines identify market segments with similar characteristics, in terms of trip purpose, price sensitivity and time sensitivity. Business and leisure demand are the most important segments targeted by airlines in their differential pricing efforts. Business travelers are assumed to be willing to pay higher fares in return for more convenience and fewer restrictions on the purchase and use of tickets. Leisure travelers are less willing to pay higher prices, but accept the disutility costs of restrictions on low fare products, longer travel times associated with connecting flights, and perhaps a lower quality of on-board service. 1 Tirole, J The Theory of Industrial Organization. MIT Press, Cambridge, MA. 2 Botimer, T.C. and Belobaba, P.P Airline Pricing and Fare Product Differentiation: A New Theoretical Framework. Journal of the Operational Research Society 50(11):

13 The airline s objective in establishing a differentiated fare structure is to offer a wide enough range of fare product options at different price levels to capture as much potential revenue as possible, while targeting each fare product to specific demand segments with different levels of willingness to pay. At higher price levels, the airline might offer enhanced service amenities that improve the attractiveness of the fare products to travelers who are not price sensitive. And, at low price levels, prices low enough to stimulate new demand for low fare travel are offered to fill empty seats that would otherwise remain empty. The challenge is to find mechanisms to prevent the diversion of consumers with higher WTP (who were expected to buy the higher fare products) to the lower fare products. For airlines, the principal benefits of differential pricing (as compared to the single fares offered prior to deregulation) are increased revenues and higher load factors, with little impact on operating costs. Incremental revenue and higher load factors come from low-fare passengers who otherwise would not fly at a higher single fare. Incremental revenue also comes from the passengers willing to pay fares higher than what the airline would charge under a single-price strategy. These incremental revenues have proven to be critical to profitability, especially for legacy airlines with high costs that would be unable to cover their total operating costs without differential pricing. Consumers have also benefited from the airlines use of differential pricing. Low-fare passengers who otherwise would not fly at the single fare level benefit clearly from the practice. While it is true that the high fare passengers are paying more than they would if the airline offered a single price level, these high fare passengers might actually end up paying less and/or enjoy greater frequency of flights given the presence of low fare passengers. The combination of multiple price levels, fare products with different restrictions and limits on the number of seats made available at lower fares through revenue management controls contributed significantly to the record airline industry profits of the late 1990s. These practices led to higher load factors and increased unit revenues (passenger revenue/ask), as airlines embraced the notion of pricing based on their perception of passengers willingness to pay. Pricing and Revenue Management in Restricted Fare Structures The application of restrictions on lower-priced fare products was traditionally the primary mechanism used by airlines to achieve demand segmentation and prevent diversion. Traditional or restricted fare structures rely on fare restrictions to make lower priced fare products less attractive to consumers with a higher WTP. The lowest fares might have advance purchase and minimum stay requirements, as well as cancellation and change fees. Studies have shown that the Saturday night minimum stay condition is among the most effective in keeping business travelers from purchasing low fares. 3 Figure 1.1 shows a traditional fare structure that relies on fare restrictions to differentiate the fare products offered within the economy class of service. In this hypothetical example, each of the fares lower than the full economy ( Y ) fare has restrictions requiring advance purchase, a Saturday night minimum stay for the traveler, and/or various change fee and/or non-refundability 3 Boeing Company Domestic Fare Survey. Seattle, WA. 12

14 conditions. The restrictions become more severe as the level of discount from the full economy fare increases. Fare Class One Way Fare Advance Purchase Minimum Stay Change Fee Refunds RT Required Y $700 None None None Yes No B $580 3 days None None Yes No M $450 7 days None $150 No Yes H $325 7 days Sat Night $150 No Yes Q $ days Sat Night $150 No Yes V $ days Sat Night $150 No Yes FIGURE 1.1: Traditional Restricted Fare Structure From the first implementations of differential pricing by airlines, it became evident that even the most effective segmentation schemes would not be enough to maximize revenues on their own. Without booking limits on low-fare seats, leisure travelers can displace business passengers on peak demand flights. This is due to the fact that leisure travelers tend to book before business travelers, a phenomenon made worse by advance purchase requirements on the lowest fares. Revenue Management (RM) is the process that determines the number of seats to be made available to each fare class on a flight. The main objective of RM is to protect seats for laterbooking, high-fare passengers. This is accomplished by forecasting the expected future booking demand for higher fare classes and performing mathematical optimization to determine the number of seats that should be protected from (or not sold to) lower fare classes. The protection levels in turn determine the booking limits on each lower fare class. RM booking limits support the objectives of differential pricing, i.e., to make consumers with higher WTP purchase higher fares. On high demand flights, RM systems will set booking limits on low-fare bookings in order to protect seats for later booking high-fare passengers. This can lead to slightly lower average load factors for the airline, but higher yields and increased total revenues. On low demand flights with excess capacity, RM will make more seats available at low fares. This can result in higher average load factors and lower yields for the airline, but higher total flight revenues. The goal of RM systems is to maximize revenues, by achieving a balance between load factor and yield. By the late 1980s, RM systems were developed to perform forecasting and optimization by booking class for each future flight leg and departure date. A typical leg-based RM system 13

15 includes the following capabilities: Collects and maintains historical booking data by flight and booking class. Forecasts future demand by flight departure date and booking class. Makes use of mathematical models to optimize total expected flight revenues, by determining overbooking levels and fare class booking class limits. Provides interactive decision support for RM analysts, allowing them to review, accept or reject the overbooking and booking limit recommendations. The major components of such an RM system are illustrated in Figure 1.2. Historical booking data for the same flight leg and day of week are used to generate a forecast of booking demand by booking class. These forecasts, together with estimates of the revenue value of each booking class, are used by an optimization model to calculate recommended booking limits for the future flight departure. At the same time, an overbooking model makes use of historical information about passenger no-show rates to calculate an optimal overbooking level. Both the booking class limits and overbooking levels are then presented as recommendations to the RM analyst who can approve them or modify them for uploading to the airline s reservations inventory system. FIGURE 1.2: RM System Components RM systems revise their forecasts and booking limits at regular intervals during the flight booking process, as often as daily in some cases. When unexpected booking activity occurs, the RM system re-forecasts demand and re-optimizes its booking limits, potentially resulting in the opening or closure of fare classes and, in turn, a change in the lowest available fare for the flight. A substantial proportion of the revenue gain attributable to fare mix optimization comes from this dynamic revision of booking limits. The vast majority of airlines around the world have implemented RM systems, and the benefits of such systems have been well documented. Effective use of RM algorithms for overbooking 14

16 and fare class booking limits have been estimated to generate revenue increases of as much as 4 to 6 percent. 4,5 Beyond these incremental revenue benefits, the use of RM systems has enabled airlines to better balance demand and supply at a tactical level. Booking limits on lower fare classes applied to high-demand flights can help to channel low-fare demand to empty flights, resulting in more even load factor distributions. RM systems have also played an important role in competitive airline pricing strategies. With RM capabilities, an airline can match or initiate almost any low fare that covers variable passenger carrying costs. Limiting the number of seats available in the lowest fare classes allows the airline to prevent revenue dilution, while maintaining a competitive pricing posture and market share in the face of low-fare competition. In the 1990s, the largest network airlines recognized that optimizing fare class mix on each flight leg independently does not ensure that total network revenues are being maximized. This is especially true for the large hub networks operated by many airlines, in which a substantial proportion of passenger itineraries involve multiple flight legs and a connection at the hub. Network RM (or O-D Control ) gives the airline the additional capability to manage its seat inventory by the revenue contribution of the passenger s origin-destination itinerary to the airline s network, not simply according to the fare class requested on a single flight leg. Central to the implementation of network RM is the concept of the bid price associated with each flight leg in the network on a given future departure date. The bid price represents the expected marginal revenue value of an incremental unit of capacity (a seat) on that leg at a specific point during the booking process. Because the bid price is the opportunity cost (in expected future revenue) of selling each incremental seat, the fare of each accepted booking request must exceed the bid price on the leg (or the sum of the bid prices over multiple legs for a connecting itinerary). In O-D control RM systems, network optimization models are used to calculate leg bid prices, which can change frequently with incremental bookings, revised demand forecasts and/or new fare inputs. O-D control represents a major step beyond the fare class mix capabilities of leg-based RM systems, and has been developed and implemented by the largest and more advanced airlines with hub networks. The benefits of advanced network revenue management methods have been estimated to be about 1-2% in incremental revenue gains, in addition to the 4-6% gains realized from conventional leg-based fare class control. For a large airline with annual revenues of $5 to $10 billion or more, successful implementation of a network RM system can thus lead to total revenue increases of $50 to $100 million dollars per year. 4 Belobaba, P.P The Revenue Enhancement Potential of Airline Revenue Management Systems. ASTAIR Proceedings: Advanced Software Technology for Air Transport, London. 5 Smith, B.C., Leimkuhler, J.F., and Darrow, R.M Yield Management at American Airlines, Interfaces 22, pp

17 Less Restricted Fare Structures and New RM Developments In the early 2000s, the traditional approaches used by airlines for both pricing and RM began to unravel, perhaps in part due to their success in increasing airline revenues in the 1990s. Consumer resistance to increasingly complicated fare structures in which the highest fares could be as much as 10 times the lowest available fare for the same economy class seat began to increase. At the same time, the rapid growth of internet distribution channels, both airline web sites and on-line travel agencies, gave consumers more information about alternative fare and airline options than they ever had before. And, the emergence of low-cost carriers (LCCs) with lower fares and less restricted fare structures increased the number of price, product and itinerary options available to consumers. All of these factors contributed to an unprecedented shift in air travel demand patterns and, in turn, airline pricing practices. The restricted fare structures described in the previous section allowed legacy airlines to effectively segment their demand and to force passengers to pay fares closer to their willingness to pay. With these significant shifts, especially with the growth of low-cost competition that does not need this traditional pricing model to make a profit, a pricing trend toward simplified fare structures began to spread around the world. Fare simplification refers to airline fare structures that have fewer restrictions on lower fares, while still maintaining many different fare levels in any given O-D market. Typically, the restriction that tends to be removed is a required minimum stay at the destination, typically over a Saturday night. Figure 1.3 shows a simplified version of the hypothetical restricted fare structure introduced in Figure 1.1. The Saturday night stay restrictions have been removed, but the advance purchase restrictions remain as a demand segmentation tool. Also, change fees and non-refundability restrictions are maintained on all fares lower than the unrestricted fare. Fare Class One Way Fare Advance Purchase Minimum Stay Change Fee Refunds RT Required Y $700 None None None Yes No B $580 3 days None None Yes No M $450 7 days None $150 No No H $ days None $150 No No Q $ days None $150 No No V $ days None $150 No No FIGURE 1.3: Simplified Less-Restricted Fare Structure 16

18 While consumers benefited from the elimination of the minimum stay restriction on lower fares, its removal also eliminated the most effective way for airlines to segment business and leisure demand. Simulation studies have shown that removal of this powerful segmentation restriction, with all else equal, can lead to airline revenue losses of 10-15% as business travelers are more frequently able to meet the reduced restrictions of the lower fares. 6 Fare simplification was in most cases driven by the entry of low-fare competitors with lower cost structures and therefore less need to maximize their revenues through traditional segmentation. In an effort to remain competitive and protect market share, many legacy carriers responded to new low-fare competitors in terms of both lower price levels and reduced fare restrictions. The negative impacts on legacy airline revenues of matching these simplified fare structures were exacerbated by the fact that airline RM systems were designed for the restricted fare structures of the 1980s and 1990s. The underlying forecasting and optimization models assumed that demands for each fare class are independent, given that restrictions prevent much diversion between fare products. With less restricted fare structures, traditional forecasting methods began to break down, because passengers are less likely to purchase higher fare classes if there are fewer rules or restrictions that differentiate low-fare products from higher-fare products. If RM forecasting methods are not updated to reflect the fact that some passengers are willing to pay more than the lowest available fare, less restricted fare structures can lead to spiraldown. 7 Since fewer passengers will buy-up to higher fare classes in the absence of restrictions, traditional forecasting will assign more demand to lower fare classes. This will cause the revenue management system to protect fewer seats for higher fare classes, and make more seats available at lower fares, ultimately damaging airline yields and revenues. In the 2000s, the significant advances in airline RM science focused mostly on correcting these deficiencies in airline forecasting methods as less restricted fare structures became more widespread. Several new methods for forecasting emerged during this period. For example, a method called Q-Forecasting adjusts the demand for each class to correct for the fact that some customers may have chosen to pay a higher fare if less-expensive classes had been closed, helping to limit the effects of spiral-down. 8 In situations with semi-restricted fare structures in which some fare classes are differentiated through fare restrictions and some are not Hybrid Forecasting can help segment between customers that are more sensitive to restrictions (product-oriented demand) and customers that are more sensitive to price (price-oriented demand). 9 By forecasting the demand for these types 6 Dar, M Modeling the Performance of Revenue Management Systems in Different Competitive Environments. Unpublished Master s Thesis, Massachusetts Institute of Technology, Cambridge, MA. 7 Cooper, W.L. and T. Homem-de-Mello Models of the Spiral Down Effect in Revenue Management. Operations Research 54(5): Belobaba, P.P. and C. Hopperstad Algorithms for Revenue Management in Unrestricted Fare Markets. Proceedings of the INFORMS Section on Revenue Management, Cambridge, MA. 9 Boyd, A. and R. Kallesen The science of revenue management when passengers purchase the lowest available fare. Journal of Revenue and Pricing Management 3(2):

19 of customers separately, airlines can do a better job of protecting seats for higher-class demand even in semi-restricted fare structures. Although these WTP forecasting methods can help airline RM systems to reverse spiral down in less restricted fare structures, modified forecasts alone are not enough to ensure that revenues will be maximized, particularly on flights with more capacity than demand. RM optimizers also needed to be modified to incorporate information about the propensity of passengers to buy down in a given fare structure. The concepts of fare adjustment and marginal revenue optimization 10 allow airlines to modify their existing RM optimization models to account for the risk of buy-down. The marginal revenue transformation recognizes that, in certain cases with less restricted fares, making additional seats available at the lowest fares can be revenue negative for the airline, even if there are empty seats expected on the flight. The marginal revenue transformation has proven to be of practical importance to airlines, as it allows the continued use of the optimization algorithms and control mechanisms of traditional revenue management systems. The transformation and resulting fare adjustment approach can be applied to both leg-based and network RM systems. A combination of hybrid forecasting and fare adjustment is considered state-of-the-art RM technology as of 2017; however, only a handful of airlines currently practice this sophisticated RM technique. Willingness to Pay Interpretation and Estimation These new forecasting and optimization models require estimates of passengers WTP in an O- D market in the form of sell-up probabilities, price elasticities or willingness-to-pay curves. The estimation of these inputs has proven to be a major research challenge. The sparseness and volatility of actual sell-up behavior in historical booking data makes the detailed estimation of specific sell-up probabilities by O-D market, day of week and time of day very difficult if not statistically impossible. Some recent research has led to the development of several estimation methods, but even in a simplified simulation environment, aggregation of sell-up estimates across markets is essential, as demonstrated by Boyer (2010). 11 Willingness-to-pay is an elusive concept that is nevertheless central to forecasting in less restricted fare environments and, as we will see, next-generation dynamic pricing techniques. In an economic sense, a customer s willingness-to-pay for a product is the maximum price that could be charged before a customer decides to forego purchasing the product. For instance, if a customer decides to purchase a product at a price of $150, but decides not to purchase at a price of $151, $150 would be the customer s maximum willingness-to-pay for the product. Some economists refer to willingness-to-pay as a customer s reservation price for the item. Economists often believe that a customer s WTP is an inherent trait of each customer that governs her decision making. However, the idea that a customer has a single, unchanging WTP for any particular product ignores the fact that customers evaluate whether to purchase a 10 Fiig, T., K. Isler, C. Hopperstad, and P. Belobaba Optimization of mixed fare structures: Theory and applications. Journal of Revenue and Pricing Management 9(1-2): Boyer, C Statistical Methods for Forecasting and Estimating Passenger Willingness to Pay in Airline Revenue Management. Unpublished Master s Thesis, Massachusetts Institute of Technology, Cambridge, MA. 18

20 product in the context of other available options. Just because a product is affordable (with a price lower than the customer s maximum willingness-to-pay) does not mean the customer will choose to purchase the product. Particularly, the presence of less-expensive or more-attractive products at the time of purchase may change the passenger s willingness-to-pay. Consider the following example. Suppose a customer is willing to pay up to $500 for a particular itinerary. If she is shown the itinerary at a price of $500 or less, she will choose to purchase it; otherwise, she will decide not to fly. In Figure 1.4, the airline prices the itinerary at $400. Since this price is less than the customer s WTP for this itinerary, she will choose to buy it. FIGURE 1.4: A customer with a maximum WTP of $500 encounters one or two identical itineraries. The customer s WTP changes based on the products they are offered. Now, suppose another airline (with identical service quality) offers an identical itinerary for $300. Since the airlines and itineraries are identical, the customer will simply choose to purchase the least expensive option. Her willingness-to-pay for the first itinerary is no longer $500; in fact, she will not buy the first itinerary at any price above $300. The presence of a lower-priced alternative has changed both her choice behavior and her WTP. This example illustrates the difference between a customer s maximum willingness-to-pay (the amount she is willing to pay for the product given there are no other alternatives presented to her) and her conditional willingness-to-pay (the most she is willing to pay conditional on the other products she observes during the decision-making process). In the examples above, the customer s maximum WTP for the first itinerary would be $500. This maximum WTP does not change depending on the other alternatives she is shown. However, when the second itinerary is displayed for $300, the customer s conditional WTP for the first itinerary changes to $300. The customer s conditional WTP for a flight can never exceed her maximum WTP for the flight, but her conditional WTP will often be less than her maximum WTP, particularly in competitive markets in which customers have many choices. Therefore, the estimates of WTP required by advanced RM forecasting methods are not of the maximum WTP of consumers in an O-D market, but rather their conditional WTP for an airline s fare product. At any point in time prior to departure, this estimate should capture the WTP of total demand (or specific demand segments) conditional on the prices and options being offered by all competitors. Conditional WTP is in fact the only thing that an airline can actually observe. In the example above, Airline 1 would never record a booking at $500, given that Airline 2 has seats available at $300. Estimating conditional rather than maximum WTP for advanced RM forecasting models also 19

21 captures some intuitive characteristics of consumer behavior: Conditional WTP will be lower in markets with competition than in monopoly markets, and lower still when the competitor is a low-fare airline. Conditional WTP will be higher during peak seasonal periods, days of the week, and times of the day, when demand is high relative to available capacity and all competitors are practicing effective RM by closing lower fare classes. Airlines that have implemented new WTP forecasting methods are already estimating conditional WTP based on historical booking patterns. Statistical estimates of conditional buyup rates can be generated by comparing bookings in a fare class when it is the lowest available class with bookings in the next higher class when the lower class was closed at the same point in time prior to departure, as just one example. The estimates can be generated at different levels of disaggregation, and perhaps even conditional on observed historical competitor fare class availability. Keeping track of historical booking patterns under all possible combinations of competitor availability could require a substantial expansion of current airline RM databases. At the much simpler extreme, conditional WTP estimates can be fed into advanced forecasters based on RM analysts judgment about a market s demand characteristics and competitive conditions. Fare Quotes, Distribution and Competition For any given O-D market and fare structure, the generation of a fare quote typically occurs at the time that an itinerary request is made through a distribution channel. That is, the outputs of the pricing and revenue management processes are stored in the airline s inventory system, and fare quotes are calculated given the fares, rules, restrictions and seat availability for each fare class. This fare quote calculation can be performed by a Global Distribution System (GDS) or within the airline s own inventory system for use by traditional travel agencies, Online Travel Agencies (OTAs) and airline websites. The traditional fare quote process is described briefly here, as its characteristics and evolution have implications that include price transparency, consistency, and fare-based competition between airlines. FIGURE 1.5: Schematic of Airline Pricing, RM, and Distribution 20

22 Figure 1.5 shows a schematic of the fare quote generation process as it exists for the vast majority of airlines and bookings made today. As described earlier, the airline pricing function establishes a fare structure for each O-D market, with different price levels and rules associated with a number of different fare (booking) classes. The details of these fare structures are published by the airline, typically through an organization such as the Airline Tariff Publishing Company (ATPCO), which in turn makes these fare structures available to various distribution channels as well as all other airlines. While this publication and dissemination process ensures that all channels have timely and accurate information about an airline s current fare structure, it also means that all competitors can monitor changes made by competing airlines to price levels or restrictions on fare products in each O-D market. For each future flight departure date, the airline s RM system has calculated the seat availability for each fare class, either on a flight leg or O-D basis, as described above. These availabilities reside in the airline s reservation inventory system, but the specific numbers of seats available for each flight, itinerary, and fare class are not as accessible to parties outside the airline (as opposed to the details of the fare structure). When a request is received for a travel itinerary in an O-D market on a particular future date, the process of generating an applicable fare quote typically involves the following steps: Identify which fare products in the airline s published fare structure are feasible for the requested date(s) of travel, given the different restrictions on advance purchase, round-trip travel, etc. Access the airline s inventory to determine which fare classes have the requested number of seats available on flights departing on the dates (and possibly times of day) requested. Calculate the lowest fare for the travel request by combining the feasible fare products and fare class availability information from the above two steps. As an example, consider a booking request for travel in an O-D market with the fare structure of Figure 1.1, with the request involving round-trip travel beginning 4 weeks from today on a Thursday afternoon, with a return the following Wednesday morning. The lowest feasible fare in the fare structure is the V-class product at $149 each way. However, if the airline s RM system has closed V-class for the Thursday afternoon flight due to high forecasted demand, then the Q- class fare of $220 would be the lowest available fare for the outbound flight. Assuming that Q- class is also the lowest available fare class on the Wednesday morning return flight, the resulting fare quote for this request would be a total of $440 round-trip. In restricted fare structures like that of Figure 1.1, different travel requests can receive very different fare quotes. In our example above, another traveler making a request to only fly oneway on Thursday afternoon on the same flight departure would receive a fare quote of $580 one-way. Even though this passenger is booking more than 14 days in advance and Q-class is available for this flight, the lowest feasible fare is the $580 B-class fare because the requested itinerary does not include a Saturday night stay or a return trip. Thus, in restricted and differentiated fare structures, fare quotes are affected both by the availability of seats in each fare class and by the characteristics of the travel request. 21

23 In less restricted fare structures with no minimum stay and no round-trip requirements, fare feasibility becomes less of an issue, although advance purchase rules still play a major role in making lower fare products infeasible. At the extreme, for a fare structure with no differentiation restrictions and no published advance purchase rules, the check for feasibility becomes moot and fare quotes are generated strictly from the inventory availability for each fare class (as determined by the RM system). The sophistication of the fare quote process continued to evolve, starting with the creation of automated airline rule data feeds by ATPCO in the 1980s and 1990s. These data feeds have over 10,000 data elements that can be used to define how a fare may be used for a particular flight leg, O&D and the total passenger journey. These elements reflect the many business rules used to satisfy the tax, regulatory, currency, segmentation, product differentiation, and ultimate fare use requirements. The ability of fare quote systems to automatically load this data and apply these rules to each transaction has allowed for accurate price calculation from the simplest to most complex passenger journeys. Fare structures differ between O-D markets based on the airline s assessment of whether the demand in a market is primarily business, leisure or mixed. Fare structures in a single market can vary over time, with the goal of stimulating demand during low periods with lower fares and perhaps increasing fare levels during peak periods. Promotional or sale fares are more likely to be offered for flights during off-peak periods, whereas some of the lowest fare products might be blacked out during the most popular holiday travel periods. Brand and service considerations can also affect an airline s fare structure in an O-D market. Similarly, an airline that provides frequent non-stop flights in a market that is served only with connecting flights by competitors might choose to charge higher prices for similar fare products. The operating costs of providing service to a specific O-D market have very little to do with the fare structure or overall price levels in that market. Although the airline must charge fares high enough to cover operating costs in the long run if it is going to survive, fare structures in a single O-D market are not determined so as to maximize profit for each O-D market. The factor with the greatest impact on both the type of fare structure and the price levels in an O-D market is competition both its presence and the type of competitor in the market. Typically, the largest impacts on average fares are associated with the entry of a low-cost carrier (LCC) into a market previously served only by legacy airlines. The presence of an LCC competitor is a factor that now affects airline pricing strategies in virtually every global region, as incumbent airlines struggle to respond to dramatically lower (and in many cases less restricted) fares offered by new LCC competitors. Given that competition has been shown to have the greatest impact on the fare structure and price levels in an O-D market, it should not be a surprise that airlines pay a great deal of attention to the fare structures of their competitors. In the vast majority of cases, airlines will match the fare offerings of their competitors, typically their lowest-price competitors. The decision to match a competitor s fare structure can include both the specific price levels associated with each fare product and their specific rules and restrictions. The move to widespread ticket distribution on the Internet has intensified significantly this practice of fare matching. As consumers make use of air fare search engines to find the lowest 22

24 possible fares for a future trip, even price differences of a dollar or two between the fare product offerings of airline competitors can relegate the higher-priced options to lower ranked positions on the web page and lead to significant booking loss. Such competitive price matching activity has a direct impact on customers' conditional WTP, as discussed above. For example, in a market with a low-fare competitor selling tickets at $99, no airline is likely to observe a conditional WTP much greater than that amount. That is, consumer perceptions of reasonable prices for air travel change as price competition intensifies. The decision to match a lower priced competitor is thus typically made in large part based on an airline s fear of losing passengers and market share. However, the decision whether to match a competitor can also be affected by other factors, such as competitive position, as well as service quality and longer-term strategic considerations. An incumbent legacy carrier offering more flights and better service quality than a low-frills new entrant LCC might believe that the immediate threat to its market share and revenues is minimal, and decide to ignore the pricing of its new competitor. For many airlines, however, the risk of not responding to low-fare competition is the potential for the new entrant to expand its presence and ultimately make a much bigger impact on the incumbent s market share, to the point of threatening the viability of serving that O-D market for the incumbent. A Brief History of Airline Price Competition The early years of liberalized pricing were characterized by airline pricing experimentation. Even before deregulation first took effect in the United States, airlines, with the approval of the Civil Aeronautics Board (CAB), had filed a small range of fare levels in each market, many with restrictions to reinforce basic product differentiation and price discrimination. By 1980, U.S. airlines still offered mileage-based First Class and Coach fares, but had filled out the structure with fares for specific target groups like military, children, or groups, and, in most markets, a restricted excursion fare, often with a lower fare for off-peak days of week. New fares, or increases or reductions to existing fares, had to be approved by the CAB in advance, often in conjunction with an economic justification. The CAB stopped reviewing fare changes in 1982 and ceased to exist in 1985, but governmental entities have continued to closely monitor airline pricing with particular focus on anticompetitive behavior and consumer protection, particularly with respect to transparency of fare rules and disclosure or fees and additional costs. In 1980, airlines were given more pricing flexibility, first to adjust only the Coach fare and later to invent new fare products and reset fare levels without regard to the mileage-based formula. The competitive landscape was further complicated by the first wave of new entrant, low fare LCCs. This prompted a period of radical experimentation all airlines, legacy and LCC alike, filed a wide variety of fare products and levels, and invented strategies to counter competitive incursions into markets perceived to be "theirs." Often, as has remained true up to the present, airlines matched competitive fare offerings, but airline pricing departments also developed new strategies and countering tactics. Some strategies were structural, for example American Airlines Ultimate Super Saver fares of 1985, taking the form of broad offerings of extremely low, tightly restricted and easily promoted 23

25 fare levels designed to blunt the low fare message of LCCs to which leisure customers were very responsive. Some were broad marketing programs to change perception and augment market share; for example, Continental Airlines supermarket coupon program of Some were limited market-specific discounts designed to skim large markets and steal market share. In addition to retail fares, airlines developed an array of specialty fares focused on particular market segments, like Bulk Contract fares, which gave discounts to Tour Operators in exchange for volume, and Corporate Travel Discount programs, which were reinforced by the growing Frequent Flyer Programs targeting higher willingness-to-pay business passengers by rewarding them with modest discounts and other perquisites. Airlines also developed a small array of defensive tactics in addition to simple matching. One example was Revenue Recapture pricing. If a competitor with connecting service lowered a fare in a nonstop market, rather than match, the nonstop carrier might file a similar level of fare in one of the originating carrier s nonstop markets in order to recapture the lost revenue without discounting their own market. Often the revenue maximizing decision for the originating carrier was to withdraw the original offer. Most of these experimental tactics and strategies had the effect of lowering fare levels, which expanded demand and resulted in high load factors, but which usually resulted in lower overall revenue. Industry profitability suffered badly. Lowering individual market fares and the overall fare structure was easy; restoring structure or raising fares was much more difficult. To raise fares, airlines often announced fare increases for some future date and only implemented the increase if other competitors also increased fares. One nadir of this period of experimentation was the Department of Justice s investigation of airline pricing practices, culminating in the 1994 Consent Decree in U.S. v. Airline Tariff Publishing Company under which most U.S. airlines agreed to follow new procedures for subsequent fare filings. By the mid-1990 s, after a period of Fare Wars, airline pricing had settled into a new normal. The basic fare structure was less segmented than in the early days of deregulation but fare levels had stabilized, albeit at lower levels. The second wave of LCCs were smaller and more tactical entering large markets with lesser frequency, and often relying on high service quality rather than strictly on low fares. Legacy airlines responded tactically. The new field of battle was airline costs, and the first front was distribution. Even before the radical disruptions caused by the Internet, airlines and travel agencies disputed commission levels, which were substantially reduced in all markets during the 1990s, and by 2002 dropped to zero in most North American markets. Airlines and GDSs disputed reach and fees, where the compromise was reduced fees in exchange for full content agreements under which each airline agreed to provide the GDS clients with fares and availability competitive with any offered by that airline in the marketplace. At the turn of the millennium, major technological disruption in the form of electronic ticketing and Internet distribution exaggerated these trends. E-ticketing eliminated one of the travel agents main functions ticket delivery further weakening their bargaining position. And, perhaps unexpectedly, the Internet encouraged consolidation rather than fragmentation of distribution channels. Internet transparency further commoditized airline products by emphasizing fare levels to the exclusion of other attributes. In a sense, there had always been transparency of fares since the development of shopping screens by the GDSs in the 1980 s, 24

26 but interpretation and communication of that transparency was controlled by travel agents. With the Internet, control passed directly to the customer. With increased transparency, fares went down. But again, the airlines adapted, making use of new capabilities to sell new products by cross-merchandising and unbundling. The Internet also allowed delivery of more complicated messages, making possible the sale of the fare families, branded products and ancillary products, as described below. The implementation of new distribution capabilities is expected to further facilitate product delivery and allow invention of a new crop of fare products. The post-deregulation history of airline pricing is a story of acceleration and volatility. In the early 1980 s, pricing changes arrived at the airlines a few at a time, literally in the form of a small stack of printed tariff pages which arrived in a package every morning via overnight mail. By 2007, pricing changes arrived by electronic transfer 3 to 8 times per day at a volume of 669,000 changes per day (Table 1.6). By 2017, international changes arrived hourly and domestic 4 times per day with a total daily volume of 3.9 million. Distribution Frequency International US/CA Domestic times per day 3 times per day Average Fare Changes per day 669k 3.9M Every hour since times per day since 2011 TABLE 1.6: The increasing pace of distribution and fare filings from 2007 to 2017 (Source: ATPCO) Recent Trends in Airline Pricing Some recent developments in airline pricing suggest a return to the basics of fare product differentiation, including the move to branded fare families. The fare family approach to pricing and demand segmentation involves sets or families of fare products, each defined by significant differences in restrictions and/or amenities. Within a fare family, there are no differences in the characteristics of the fare product, but there can exist multiple price points for that fare product. Airline RM forecasting and optimization models developed for application to fare family pricing structures can determine which price point to offer within each family at a given point in time, based on estimates of conditional WTP and passenger preferences for each fare family. Branded fare families refer to the use by the airline of a marketing name for each of these differentiated products, for example, Tango and Latitude at Air Canada. Fare family pricing can allow the airline to capture higher revenues than traditional fare structures. 12 By differentiating the fare products with a variety of amenities and restrictions, the airline can offer more choices to both business and leisure travelers closer to departure. Whereas the fares targeted at leisure passengers had 21-day advance purchase rules in traditional fare structures, a fare family product with less attractive amenities can be available to leisure passengers later in the booking process, increasing revenues and load factors. 12 Surges, V RM Methods for Airline Fare Family Structures. Unpublished Master s Thesis, Massachusetts Institute of Technology, Cambridge, MA. 25

27 The use of product differentiation among families helps to reduce the revenue risks of diversion. At the same time, business travelers wishing to purchase a fare family product with fewer restrictions and improved amenities early in the booking process can obtain substantially lower fares than the full Y fare of the traditional restricted fare structure, stimulating new business travel demand and contributing to higher revenues and load factors. There is also much evidence that presenting consumers with multiple fare family options to choose from both increases pricing transparency for the consumer and has the potential to increase revenues for the airline. Retailing research has shown that a consumer is less likely to purchase the lowest-priced, least desirable option when presented with 3-4 higher priced alternatives that are more appealing (or have fewer restrictions). This buy-across behavior increases revenues for the airline, while allowing consumers to make a more informed trade-off between fare product characteristics and price. Delta s recent branded fare initiatives increased the airline s revenue by $100 million in second quarter 2017, and senior management said that the carrier is still in the early stages of unlocking the value of segmentation. 13 In contrast to the bundling approach of fare families, charging fees in addition to a basic ticket price is referred to as unbundling or a la carte pricing. Historically, airline ancillary revenues included excess baggage fees and cancellation/change fees as their largest components. More recently, many airlines have begun charging fees for: Passengers checking a first and/or second bag; On-board sales of food, drinks, pillows/blankets, TV/video, and internet; Advance seat assignments, premium seat locations, seats with extra leg room, blocking of a neighboring seat; and Access to priority boarding and expedited security lines. The ancillary revenue generated by this relatively new airline pricing practice is now a major source of revenue and profit for airlines. In 2016, US airlines collected over USD $12 billion in ancillary fees, representing 8% of total revenues. 14 Worldwide, ancillary revenues were estimated to reach USD $67.4 billion in 2016, 15 although this estimate includes a large proportion of revenues tied to sale of loyalty program points. Development of a New Distribution Capability Much of the evolution of airline distribution has been constrained by the capabilities of the earliest computer reservations systems developed more than 50 years ago. Even the most advanced distribution processes are still limited to communicating the availability of each booking class for which a fare has been published by the airline. The price quoted to consumers shopping for flights is determined by the number of seats available in a booking 13 Baker, M.B Branded Fares Boost Delta Q2 Revenues. Business Travel News 13 July United States Department of Transportation Air Carrier Financial Reports (Form 41 Financial Data), Bureau of Transportation Statistics, Washington, DC IdeaWorks Airline ancillary revenue projected to be $67.4 billion worldwide in Retrieved from 26

28 class and whether the booked itinerary meets the rules and restrictions of a fare associated with that booking class. This framework has limited the ability of airlines to distribute either bundled fare products or a la carte ancillary fare options via the traditional GDS distribution framework. These limitations led some airlines to pursue direct connect alternatives on their own, as described earlier, in an effort to by-pass both the costs and product distribution constraints of the GDS channels. Airlines were taking advantage of new technologies and communications capabilities to implement direct connect, but there was no single industry standard for distribution until IATA Resolution 787 launched the New Distribution Capability (NDC) initiative in NDC involves the development of a new XML-based (extensible markup language) data transmission industry standard for airline distribution. 16 The New Distribution Capability (NDC) currently under development as a new global standard for airline distribution enables airlines to communicate all of the features of each fare product offering, not just its price and restrictions. By providing a standardized way of communicating the offers, it could also bring the airlines closer to dynamic pricing and even customized pricing, where different types of requests or even individual customers could receive different price/product offerings based on the airline s assessment of their potential total revenue contribution. 17 The most important change contemplated by NDC is that requests for seat availability and fare quotes can be sent to the airline for real-time evaluation. In essence, NDC will provide for the equivalent of direct connect between the airline s own inventory in its PSS and all distribution channels, both direct and indirect. Indirect channels that involve travel agencies (whether traditional or on-line) as well as GDS providers will receive real-time availability and fare quote information from a new airline pricing and merchandising engine that will reside with the airline. The airline s direct distribution channels, including its own web site and reservations call center, will similarly receive information in real-time. NDC will also give the airline the capability to generate an offer consisting of not simply a published ticket fare, but one that includes additional amenities and services. In contrast to the static dissemination of seat availability by flight and booking class, real-time interactive communications between customers (or their agents) and the airline will enable dynamic packaging and pricing of flight, price and product combinations. The new capabilities will also permit airlines to give their customers the choice of different ancillary product options and build a shopping cart before checking out. The expectation is that NDC will allow airlines to generate offers of fares and products to customers and travel agents without them being prepared and packaged by a third party. Realtime adjustment of seat availability and fare quote responses will become more feasible. Under the NDC framework, if an airline were to choose to do so, it could provide a personalized offer to 16 IATA, New Distribution Capability (NDC): Facilitating Air Retailing, Strategy Paper V1.0, International Air Transport Association, September 2014, Geneva. 17 Westermann, D The Potential Impact of IATA s New Distribution Capability (NDC) on Revenue Management and Pricing. Journal of Revenue and Pricing Management 12(6):

29 each customer if the customer chooses to be recognized 18 for example, by logging in to the airline s web site. Such uses of NDC by airlines will also necessitate changes to the traditional approaches to pricing and revenue management. Airlines will have to develop (or purchase) their own pricing capability to generate a fare quote before sending this information to the distribution channels. The possibilities afforded by the NDC distribution standard are at the same time futuristic and subject to a great deal of uncertainty with respect to its impacts on airlines and consumers. While improvements in the accuracy and consistency of airline flight and fare information are likely to be welcomed by all stakeholders, the impacts on competition, transparency and perceived fairness to consumers of dynamic packaging and individualized price quotes are less clear. The ability (and willingness) of airlines to make what would be massive changes to a variety of legacy systems related to all facets of pricing, distribution and ticketing is uncertain. And, any such changes will need to ensure the continued accuracy of fare calculations for all types of passenger itineraries. What is clear is that NDC has the potential to fundamentally change the traditional distribution process that has been in place for decades. The extent to which this potential is ultimately realized will depend on many constraints organizational, competitive, technological, and regulatory, to name a few as will be discussed in Chapter IATA,

30 2. Definitional Framework for Price Selection Mechanisms Chapter 1 reviewed the evolution of the pricing, revenue management, and distribution practices currently used in the airline industry. Many of these practices were developed in response to the specific architecture of airline IT systems and industry standards. Firms in other industries have also developed and adapted their own pricing and inventory management practices, as a function of their own specific needs, requirements, legacy technologies, and data availability. Before exploring how pricing practices in other industries are related to and different from those used by airlines, it is worthwhile to first define and classify these practices in broader terms. This chapter presents a framework for classifying price selection mechanisms into three categories: assortment optimization, dynamic price adjustment, and continuous pricing. The framework will be used in Chapters 3 and 4 to classify pricing practices in other industries, as well as next-generation pricing mechanisms currently under development in the airline industry. The framework developed in this chapter focuses primarily on the processes that firms use to select which price to charge for a given product. For airlines, the product creation process occurs prior to price selection and could range from pre-constructing branded fare products to dynamically bundling or packaging multiple ancillary services into product offerings. The product creation process itself is not included in our price selection framework. However, in Chapter 4, we discuss a next-generation mechanism called dynamic offer generation that would combine product creation and price selection into a single process. Assortment Optimization One common price selection mechanism is to define a finite set of possible price points and then select which of those fixed price points to display in response to shopping requests. We call this technique assortment optimization. The key feature of assortment optimization is the relatively small and finite menu of possible price points from which the firm selects its prices. FIGURE 2.1: Schematic of Assortment Optimization Figure 2.1 shows a simple schematic of assortment optimization. In the figure, the firm starts with a set of six possible price points for a product, ranging from $299 to $999. In practice, this set of price points can be based on market conditions for instance, based on the number of competitors in the market, or the prices that competitors offer for similar products. The firm then 29

31 selects one of these price points to display to the customer in this case, $375. This price selection process could be managed through business rules that govern when or how particular price points are made available. For instance, a particular price point may only be available if purchased far in advance, if the product cannot be returned or refunded, or if the customer is a senior citizen. Along with simple business rules like these, a firm could also use a mathematical optimization model to determine which price point to display at a given time, as a function of a demand forecast and the amount of inventory remaining. In some industries, firms perform this price selection process infrequently, perhaps on a weekly or monthly basis. Other firms select prices on a daily basis, or multiple times per day. At the limit, prices could be selected from the menu for each transaction individually, depending on the characteristics of the transaction or the customer. We call this real-time selection of prices transactional assortment optimization. Even in transactional assortment optimization, the possible prices are still limited to the price points in the pre-defined menu. Firms practicing assortment optimization will also periodically update the menu of possible price points in response to changes in market conditions. This could be done by adding or subtracting price points from the menu, altering the rules and restrictions associated with each price point, or changing the values of the prices in the menu. However, for each individual transaction, the price shown is always selected from the finite set of price points that is valid at the time. Definition: Assortment Optimization With assortment optimization, firms select one or more prices from a finite menu of possible price points. Each price point may be associated with various rules or restrictions that determine how or when that price can be selected. The menu of price points may be updated periodically, but there is only a limited and discrete set of possible prices that can be selected at any given time. The selection of a price from the menu could be made infrequently or, at the limit, on a transaction-by-transaction basis (transactional assortment optimization). Traditional airline pricing and revenue management, as described in Chapter 1, is clearly an assortment optimization problem. In traditional pricing and RM, airlines first file a set of fare products with a central organization, such as ATPCO. The fare products each have a set of rules and restrictions, as well as a product-specific price. The set of fare products in each market makes up the fare structure, which represents the finite menu of possible price points. Airlines then use RM systems to perform the price selection process: deciding which fare products to make available in response to booking requests. This decision is made not only based on the rules and restrictions associated with each fare product, but also: The number of seats remaining on each flight in each itinerary; The amount of time remaining until the flight departs; A forecast of demand-to-come for each of the flights in each itinerary; and The presence of any special events or holidays that could influence demand. Airlines often adjust the prices and/or rules of fare products to reflect changes in the 30

32 marketplace for instance, if a new competitor enters a market, or if a competitor changes their own set of possible price points. As discussed in Chapter 1, the frequency of changes to airline fare structures has increased in recent years. Yet as long as airlines are selecting prices from a relatively small, discrete set of possible options, they are still practicing assortment optimization. Dynamic Price Adjustment Firms may sometimes wish to offer a price that is not included in their pre-defined sets of possible price points. As discussed above, the firm could do so by updating or adjusting the menu of price points to incorporate this new option, and continuing to practice assortment optimization. However, it may not be necessary, desirable, or feasible to make this type of change to the menu of price points for instance, if the new price point is offered only for a certain segment of customers, or is only valid for a limited time. In this situation, another approach focused on dynamically adjusting the prices from the predefined menu may be warranted. Suppose a firm wishes to give certain customers a discount of 10% off a product s normal price. The firm could do so by first selecting its price using the assortment optimization approach described above, and then dynamically applying the discount for customers that fit the criteria. For example, in Figure 2.2, the firm selects a price of $375 through its assortment optimization process and then applies the 10% discount to transactions that fit the relevant criteria. This allows the firm to offer the price of $337.50, which is not listed in the set of pre-defined price points, to certain customers. Customer requests that do not meet the criteria to receive the discount are offered the original price of $375. FIGURE 2.2: Schematic of Dynamic Price Adjustment We call this approach dynamic price adjustment. Firms using dynamic price adjustment begin with a pre-defined, relatively small set of possible price points. As in assortment optimization, these price points could be associated with various rules and restrictions that govern their applicability. Then, after selecting one of the pre-defined price points using the assortment optimization approaches described in the previous section, the firm can adjust that price in certain situations. This adjustment can either take the form of a discount or an increment relative to the price selected from the menu. The amount of the adjustment could also change from transaction to transaction. Transactions that are not eligible for a discount or increment are shown the unadjusted price. 31

33 Definition: Dynamic Price Adjustment With dynamic price adjustment, firms start by selecting a price from a pre-defined menu of possible price points, as in assortment optimization. Then, for certain customers or in certain situations, this price is adjusted through either a discount or an increment. All adjustments are made in reference to a price from the fixed menu, and some customers are shown an unadjusted price. With dynamic price adjustment, firms are able to offer prices outside of those listed in their predefined sets. A key feature of dynamic price adjustment is that any increment or discount is made relative to one of the pre-defined price points in the menu. Additionally, since dynamic price adjustments may only apply to a subset of customers or transactions that meet certain criteria, some customers will receive unadjusted prices from the menu. As with assortment optimization, the dynamic price adjustments could be computed infrequently (for instance, a flash sale that applies to all customers on a specific day) or, at the limit, at the individual transactional level. In the airline industry, a Dynamic Pricing Engine (DPE) is one mechanism that has been proposed to implement dynamic price adjustments. With a DPE, airlines could choose to markdown pre-filed fares for specific transactions. Since airlines still use a finite, pre-determined set of possible price points (and existing pricing and revenue management techniques) to complete their initial price selection, and since some customers are not given a DPE-adjusted fare, this concept represents an example of dynamic price adjustment. Dynamic Pricing Engines are discussed in more detail in Chapter 4. Continuous Pricing Finally, continuous pricing represents the most flexible practice that firms can use to set prices. With continuous pricing, the firm does not pre-define a relatively small number of possible price points. Rather, prices are selected from a continuous range of possible values. Any price point within the range is feasible for selection for instance, in Figure 2.3, any value between $299 and $999 can be chosen. This gives firms the potential to fine-tune prices to the penny, if necessary, without having to pre-determine a fixed number of possible prices. FIGURE 2.3: Schematic of Continuous Pricing 32

34 Definition: Continuous Pricing With continuous pricing, firms select a price from a continuous range of possible values. There is no underlying finite menu of possible price points, although there may be business rules that determine the range of allowable prices at any moment. At the limit, dynamic prices could be generated individually for each transaction (transactional dynamic pricing). However, prices do not necessarily need to differ from transaction to transaction. Although continuous pricing provides firms with many more possible price points than assortment optimization, firms can use similar methods to select the price that is displayed. As in assortment optimization, firms using continuous pricing can select prices by incorporating business rules, mathematical optimization models, information about remaining inventory or forecast demand, and/or transaction-specific data. Also as in assortment optimization, firms using continuous pricing can complete the price selection process infrequently or, at the limit, on the transactional level. Transactional continuous pricing, in which prices are selected from a continuous range for each individual transaction, represents one of the most complex forms of pricing that can be practiced by firms. This practice requires contextual information about each transaction and sophisticated rules or mathematical algorithms to select prices as a function of this information. As is discussed in Chapter 3, relatively few firms are currently practicing such sophisticated pricing techniques, and those that are practicing transactional continuous pricing are not limited by legacy technological constraints. Conclusions Summing up the Definitional Framework This chapter presented a simple definitional framework for price selection mechanisms that sophisticated firms use to choose which price to charge for a given product. With assortment optimization, firms select prices from a pre-defined menu of a relatively small number of possible price points. Each of the price points could be associated with internal or external rules and restrictions that determine when they can be chosen. Firms can periodically update the menu of price points, but only the prices included in the menu can be selected at any particular time. With dynamic price adjustment, firms begin with a pre-defined menu of possible price points, as in assortment optimization. After selecting one of the pre-defined prices, firms can increment or discount this price for specific customers or in specific situations. These adjustments, which could vary from transaction to transaction, are made relative to the pre-selected price point. This allows firms to offer prices outside of the pre-defined menu, although some customers will receive unadjusted prices. With continuous pricing, firms select prices from a continuous range of possible values. There is no finite menu of possible price points; any price within the range can be chosen. For any of the methods described above, firms have a choice of how often they wish to update their price selection. Prices could be selected infrequently, on a weekly or monthly basis, or more frequently, such as daily or multiple times per day. At the limit, the price selection process 33

35 could occur for each individual transaction. With this type of transactional pricing, prices could change from request to request as a function of information about the customer, the state of forecast demand, or remaining inventory. The frequency of price updates is dependent not only on the degree of technical sophistication of the firm, but also on how much contextual information the firm has about each transaction. While current airline industry pricing and revenue management practices can be best defined as assortment optimization in our framework, other industries practice a wide range of other pricing techniques. In Chapter 3, pricing practices in a variety of industries are reviewed. The definitional framework described in this section is used to classify these practices to allow for comparisons to be made to the airline industry. In Chapter 4, which describes new mechanisms for pricing that are currently under development in the airline industry, the definitional framework introduced in this chapter is used to describe the potential evolution of airline pricing from assortment optimization to continuous pricing. 34

36 3. Pricing and Revenue Management in Other Industries The airline industry is not alone in using the concepts of differential pricing and revenue management. For airlines, high fixed costs and a commitment to operate a schedule of flights on a given date mean that seat inventories are highly perishable once the flight departs, unsold seats have no value. Much of the competitive pricing behavior of airlines described in Chapter 1 is driven by efforts to fill as much of this perishable inventory as possible with incremental revenue, to increase load factors and to protect market share. Although airlines are recognized to be leaders in the development of increasingly sophisticated pricing and revenue management techniques, other industries have followed suit, and in some cases have been able to implement what appear to be even more sophisticated pricing and RM processes. As a general rule, these other industries have not been as constrained as airlines in terms of legacy industry standards, systems and regulations affecting pricing and distribution. These industries also typically sell products at lower prices and with shorter purchasing horizons than airline tickets, giving firms greater flexibility to develop and implement complex pricing practices including, in some cases, transactional continuous pricing. In this chapter, we first review techniques used for pricing and revenue management in several travel-related industries, and describe how these practices compare to those used by airlines. Many of these industries, such as hotels, passenger rail, and rental cars, have adopted some of the same pricing and revenue management practices used in the airline industry, using assortment optimization to set the prices of their products. Unlike airlines, firms in these industries often do not publish price structures publicly. We also examine a small number of technologically-advanced firms, such as Amazon.com, Airbnb and Uber, which are practicing highly-sophisticated, next-generation dynamic price adjustment and continuous pricing techniques. These firms have developed among the most sophisticated pricing practices currently in use worldwide in consumer markets. Table 3.1 summarizes the pricing mechanisms generally used by these industries, as well as whether firms in those industries typically publish their price structures publicly. Travel-Related w/ Asst. Opt. Advanced Pricing Industry Pricing Mechanism Publicly-Available Price Structure? Airlines Assortment optimization Yes Hotels Assortment optimization No Passenger rail Assortment optimization Yes Rental cars Assortment optimization No Online Retail Dynamic price adjustment No Airbnb Continuous pricing No On-demand transport Dynamic price adjustment & transactional continuous pricing Yes/No TABLE 3.1: Summary of pricing mechanisms in various industries 35

37 Travel-Related Industries Practicing Assortment Optimization Many travel-related industries have borrowed liberally from airline pricing and revenue management practices when structuring their own pricing mechanisms. These industries, such as hotels, rental cars, and passenger rail, share many features in common with the airline industry limited and perishable inventories, segmented customer demand, limited selling seasons, and a shared capacity serving many markets that is fixed in the short term. However, a key difference is that these industries typically do not publicly file the price points used for assortment optimization, in contrast to the airline industry. These industries are also not as constrained by the legacy technological systems that have driven much of the development of pricing and RM practices in the airline industry, although many travel-related industries still distribute prices through global distribution systems. While fewer constraints could give these industries more flexibility in pricing their products, in practice the level of pricing and RM sophistication in these industries does not typically exceed that of the airline industry. Hotels Hotels are a global industry whose worldwide revenues totaled over $550 billion in , with an inventory of over 15 million rooms across the globe. 20 The industry is dominated by a number of large brands, including Hilton Worldwide, Marriott International, Starwood Hotels, and the InterContinental Hotel Group (IHG). Although these large hotel companies control the branding and guest experience of their properties, the pricing and revenue management of these properties are typically controlled by individual owner-operators. 21 Hoteliers are concerned with maximizing their total revenue from a fixed number of available rooms. Due to the perishability and limited inventory of rooms, hotels trade off selling rooms in advance for a discount with maintaining an inventory of rooms for late-arriving walk-up customers willing to pay more. This decision is identical to the airline RM problem of computing how many seats to protect for late-arriving passengers. Hotels often construct a rate structure for each room type, consisting of a number of different products designed to segment demand into different price points. Many such rates will exist, and could depend on corporate discount codes, packages and promotions, distribution channels, or customer segmentation. A rate structure may have a high-priced rack rate for last-minute walk-up customers; an advance purchase rate for leisure customers; a discounted rate for members of associations like the American Automobile Association; and a corporate rate negotiated for business customers Hospitality Net The Global Hotel Industry and Trends for Intercontinental Hotels Group Industry Overview. Annual Report and Form 20-F Altin, M A taxonomy of hotel revenue management implementation strategies. Journal of Revenue and Pricing Management 16(3): Goldman, P., R. Freling, K. Pak, and N. Piersma Models and techniques for hotel revenue management using a rolling horizon. Journal of Revenue and Pricing Management 1(3):

38 If the hotel practices revenue management, it uses assortment optimization to decide how many rooms to make available for each of the pre-priced products from the rate structure. As in the airline industry, the rates made available by the hotel s revenue management system depend on the hotel s forecasts of demand for each night. Given that rooms may be booked for multiple nights, some hotels use more sophisticated revenue management techniques that incorporate rolling decision periods to maximize revenue over multiple days simultaneously. 23 This optimization problem is comparable to the use of network RM or O-D controls by airlines. Some hotels also practice total revenue management, which focuses on maximizing not only the room revenue, but also the sum of all ancillary revenues. 24 The casino industry is well known for practicing total revenue management. The ancillary service for a casino is gambling revenue, which in many cases may exceed the revenue for the room. Casinos often track personalized customer data through rewards cards and provided targeted offers and discounts to certain customer segments known to generate high gambling revenues. Table 3.2 provides an example from the casino hotel chain Harrah s, which segments their customers into 10 types based on expected gambling revenue and revenue manages these segments accordingly. 25 TABLE 3.2: Example of casino revenue management based on customer segments. RFB and ROC indicate that room, food, and beverage are provided complementary 26 Overall, the hotel pricing and revenue management environment is in many ways similar to the airline environment. Hotels select a fixed number of pre-priced products to make available in their rate structures. The products are differentiated using characteristics like advance purchase requirements, non-refundability, and customer segmentation. The hotel then selects which products to make available in response to each booking request, and may distribute this information through a Global Distribution System. As a result, hotel pricing is a capacitycontrolled assortment optimization problem, using the taxonomy from Chapter Ibid. 24 Zheng, C. and G. Forgacs The emerging trend of hotel total revenue management. Journal of Revenue and Pricing Management 16(3): Metters et al., Ibid. 37

39 There are some differences between hotel revenue management and airline revenue management. First, the rate of capacity utilization in the hotel industry is typically less than in the airline industry. The utilization rate for U.S. hotel rooms was about 65.5% in 2016, compared to an 85% utilization rate for U.S. airlines in the same year. 27 This excess to some extent limits the control that the hotel revenue management system can impose on prices. Hotels must thus pay more attention to fencing and differentiating their rate structures, since in many cases it will be the restrictions in the available rate, and not a revenue management system, that will increase prices for closer-in customers. Also, hotels traditionally have been more flexible than airlines in providing cancellable or refundable products. As opposed to airline tickets, which are often non-refundable after they are purchased, many hotel room rates (even at low price points) allow cancellation for a full refund up to 24 hours before check in. This trend is starting to change, as hotels are starting to promote non-refundable advance purchase rates on their own websites and on indirect channels. Hotels are also starting to push back their free cancellation windows from 24 hours before check-in to 48 and 72 hours, even on their most flexible rates. Passenger Rail Transportation Globally, passenger railways serviced over 31 billion passenger trips in 2015, according to the International Union of Railways. 28 Many passenger rail operators are government owned, and have historically focused on providing consistent prices to local residents and commuters rather than maximizing revenue from each transaction. 29 In many countries, prices are fixed by distance regardless of when the journey is purchased, or how many seats are remaining. On the other hand, there are many passenger rail companies that have started to practice differential pricing and revenue management in a very similar manner to airlines. Amtrak (US), VIA Rail (Canada), SNCF (France), Deutsche Bahn (Germany) and Eurostar (UK/France) are just a few examples. These operators use revenue management systems that are effectively the same as those used in the airline industry in many cases, technology vendors simply reconfigured existing airline RM systems for use by passenger rail operators. 30 As in airline pricing and RM, rail operators use fare grids to pre-define a number of possible price points. The price points may be associated with various rules and restrictions for example, some price points may only be valid for student travel, or for one-way travel. The fare grids may be published publicly and/or be visible through Global Distribution Systems, but they are updated much less frequently than in the airline industry. The fare grids at SCNF, the French national railway operator, are published every six months, as opposed to the multiple 27 STR, Inc Monthly Hotel Report Total United States December 2016, and MIT Airline Data Project. 28 International Union of Railways Railway Statistics 2015 Synopsis Mitev, N Trains, planes, and computers: From high-speed trains to computerized reservations systems at French Railways. Journal of Transport History 25(2): Abe, I Revenue Management in the Railway Industry in Japan and Portugal: A Stakeholder Approach. Unpublished Master s Thesis, Massachusetts Institute of Technology, Cambridge, MA. 38

40 fare filings that take place in the airline industry each day. 31 The price points are associated with various fare classes or buckets used by the railway s revenue management system for inventory control. Table 3.3 shows an example of a simple bucketed fare structure used for Amtrak s AutoTrain service, which allows passengers to travel with their cars. 32 As with airline revenue management, railway inventory is controlled as a function of capacity remaining, a forecast of remaining demand, and the number of days until departure. TABLE 3.3: Fare structure for Amtrak s AutoTrain Service. (Source: Sibdari et al., 2008) Like the airline industry, passenger rail is also highly network oriented. A single seat on the same train can be used for one or more segments of origin-destination demand. For instance, with 35 station stops, a single seat on a Boston to Newport News train could be sold in a combination of 595 different origin-destination pairs. Railways can thus use O&D control concepts to control inventory across a single train s journey. Product differentiation through fare restrictions and branded fare products is also common in the passenger rail industry. As shown in Figure 3.4, Amtrak offers up to four different fare families for sale on various long-distance itineraries, each of which possesses different fare restrictions. FIGURE 3.4: Differentiated branded fares on Amtrak s Northeast Corridor services With pre-determined fare structures, differentiated fare products, and inventory management systems, pricing and RM in passenger rail is very similar to that of the airline industry. It is not 31 Cote, J.-P. and M. Riss A New Generation of Commercial Optimization Tools for High-Speed Railway Operations in a Competitive Environment. Proceedings of the 2006 World Congress of Rail Research. 32 Sibdari, S., K.Y. Lin, and S. Chellapan Multiproduct revenue management: An empirical study of Auto Train at Amtrak. Journal of Revenue and Pricing Management 7(2):

41 surprising that many of the practices from airline pricing and RM have found their way to passenger rail, since airline systems in many cases formed the basis of the methods that are currently used by passenger rail operators. Rental Cars The rental car industry, which totaled over $27.1 billion in revenues 33 in the U.S. alone in 2015, renting nearly 2.2 million cars 34, also practices differential pricing and revenue management techniques similar to airlines, hotels, and passenger rail. As with hotels, car rental companies offer a diverse array of differentiated inventory for sale for instance, economy cars, SUVs, and convertibles, as shown in Figure 3.5. Each vehicle type can be sold at a variety of rate levels. 35 FIGURE 3.5: Example rate structure for various classes of vehicles across different car rental companies. (Source: Carroll and Grimes, 1995) As with airlines and passenger rail, the rental car RM system selects which rates to make available for each type of vehicle, as a function of forecast demand and available inventory. Rates can also be further differentiated into business and leisure segments with the use of restrictions such as advance purchase requirements, non-refundability, and minimum lengths of rental (MLOR). Since rental car companies select prices from a pre-determined set of rates, their pricing and revenue management practices represent an assortment optimization problem. Rental car companies select the products (i.e., the combinations of vehicle types and pre-priced rates) that are displayed in response to each booking request, as a function of the available inventory and the characteristics of the request. However, the rental car pricing and revenue management problem does possess some unique features relative to the airline industry. Rental cars are typically booked closer to departure than hotels or airline tickets, shortening the booking period, and no-shows are also common. 36 Also, 33 Peltier, D Charts Showing Car Rental Companies Growth in Skift 6 January Ibid 35 Carroll, W.J. and R.C. Grimes Evolutionary Change in Product Management: Experiences in the Car Rental Industry. Interfaces 25: Gordon, R.F Why It Is Difficult to Apply Revenue Management Techniques to the Car Rental Business and What Can Be Done About It. Proceedings of the Northeast Business and Economics Association,

42 since vehicles can be rented for multiple days, and since vehicles can be returned to a different location from where they are rented, the rental car industry has the unique feature of being needing to rebalance inventory between rental locations in short time horizons. 37,38 Some RM approaches to the car rental revenue management problem also incorporate the possibility of complimentary upgrades, similar to the airline multiple-cabin RM optimization problem. 39 One key difference between rental cars and airline industry is that rate structures are not publicly filed by rental car companies, and are therefore not directly visible by other competitors. The lack of publicly-filed fares also allows car rental companies to easily and frequently make changes to the underlying rate structures input into the RM system, with little or no transparency for customers. As a result, although RM in the car rental industry optimizes over a discrete set of pre-priced rates, this set of prices can be easily changed multiple times over the optimization period. This gives car rental companies more flexibility to change rate structures than airlines, Despite the lack of publicly filed fare structures, pricing strategies that simply match the lowest available competitor price are still typical in the car rental industry. This is particularly the case during times of the year where there is an excess inventory of vehicles. Web scraping tools or other business intelligence technologies are used to inform car rental companies when competitors have changed their price points or availability. Aside from the absence of publicly-filed rates and differences in the underlying mathematics, RM in the car rental industry is similarly structured to the airline industry. After optimizing over a discrete set of possible rates, the RM system sends availability information to the reservations system, which can distribute the availability to various channels, record bookings, and pass the booking information to the RM system for use in future forecasts and optimization. Conclusions: Travel-Related Industries Practicing Assortment Optimization In this section, we reviewed pricing practices in three travel-related industries hotels, passenger rail, and rental cars that use similar assortment optimization techniques as the airline industry. The similarity of these underlying pricing problems to the airline industry makes assortment optimization techniques well-suited for use in these industries, and it is no surprise that these industries use similar techniques to optimize prices. However, unlike airlines, these industries are typically not required to publicly file their underlying price structures, and have not been as constrained by legacy systems and standards as the airline industry. Yet despite the relative freedom of pricing in these industries compared to the constraints faced by airlines, the pricing and revenue management practices used in these industries are generally not more complex than those used by airlines. These industries tend to borrow pricing practices from airlines, rather than vice versa. As airlines continue to develop new and innovative pricing mechanisms, it is likely that many of those mechanisms may also find their way to these other travel-related industries as well. 37 Ibid. 38 Fink, A. and T. Reiners Modeling and solving the short-term car rental logistics problem. Transportation Research Part E 42: Guerriero, F. and F. Olivito Revenue Models and Policies for the Car Rental Industry. Journal of Mathematical Modelling and Algorithms in Operations Research 13(3):

43 Firms Practicing Dynamic Price Adjustment and Continuous Pricing Amazon, Airbnb, and Uber Next, we review pricing practices for three technologically-sophisticated firms practicing dynamic price adjustment and continuous pricing. The online retailer Amazon.com, the peer-topeer accommodation service Airbnb, and the on-demand transportation service Uber each use complex machine-learning capabilities to compute prices based on a wide range of market conditions, competitor information, and contextual data about each transaction. These nextgeneration pricing practices have also run into a fair share of controversy as regulators and consumers consider what degree of transactional continuous pricing is acceptable in the marketplace. Amazon.com Dynamic Price Adjustment in Online Retail Amazon.com is the world s largest online retailer, and also serves as a distribution platform for third-party vendors through its Amazon Marketplace. Amazon and its third-party vendors can use complex algorithms to dynamically adjust the prices for the products they sell. Prices for products on Amazon have been observed to change multiple times a day and as frequently as every 20 minutes. 40 In one example, the price of a microwave changed nine times over the course of a day, between $744 and $ FIGURE 3.6: An example of Amazon s dynamic price adjustment mechanism Amazon s pricing mechanism can best be described as dynamic price adjustment. For instance, Amazon typically presents the prices for books in reference to a fixed manufacturer retail price or list price. As shown in Figure 3.6, offered prices are displayed as markdowns in reference to this reference price, and the markdowns offered can change over time in response to market conditions or remaining inventory. This strategy is not without controversy; in July 2017, the Federal Trade Commission began a review of Amazon s reference prices for deceptive pricing practices after a Consumer Watchdog report found that for many products, the reference prices were significantly higher than the prices actually charged in the previous 90 days. 42 Amazon and other online retailers have also used contextual information about customers to alter prices from transaction to transaction. This type of pricing often is the subject of scrutiny if discovered by consumers or the media. There have been a handful of instances where online retailers have changed prices or inventory as a function of customer-specific information: Amazon.com in 2000 apologized after charging different customers different prices for 40 Chen, L., A. Mislove, and C. Wilson An Empirical Analysis of Algorithmic Pricing on Amazon Marketplace. Proceedings of the WWW2016 Conference, Montreal, Quebec, Canada, April Weisstein, F.L., K.B. Monroe, and M. Kukar-Kinney Effects of price framing on consumers perceptions of online dynamic pricing practices. Journal of the Academy of Marketing Science 41: Bartz, D FTC probing allegations of Amazon s deceptive discounting. Reuters. 20 July

44 the same DVDs; 43 Staples.com in 2012 charged different customers different prices for the same products depending on their geographic locations (Staples said that their prices were set according to a variety of factors including the costs of doing business in different places); 44 Orbitz.com in 2012 admitted showing more expensive sets of hotels to users of Macintosh computers than users of Windows PCs; 45 Delta Air Lines in 2012 appeared to show higher prices to customers logged into their frequent flyer accounts (Delta blamed this issue on computer problems ) 46 While dynamic price adjustment has become more common and accepted in many industries, adjusting inventory or prices as a function of individual customer characteristics still remains relatively taboo. Firms appear highly hesitant to indicate that they are pricing based on customer characteristics, perhaps because it leads to a sense of unfairness or discrimination. As a result, while sophisticated pricing techniques could allow Amazon and its sellers to react quickly to changes in market conditions, safeguards are required to prevent automated algorithms from making undesirable choices. Otherwise, the retailer may suffer episodes of poor public relations (if the price is accidentally set too high) or the revenue impacts of firesales (if the price is accidentally set too low), which could wipe out any revenue gains associated with continuous pricing. 47 Peer-to-Peer Accommodation Services: The Case of Airbnb Airbnb is a peer-to-peer accommodation service that started in 2008, and has rapidly expanded to more than three million listings worldwide. 48 Individual property managers, called hosts, list accommodation on the Airbnb website, ranging from shared rooms in the host s living space to full apartments. Airbnb earns income by charging a commission, which is based on the nightly price of the unit, to both the guest and the host. The pricing and revenue management problem for Airbnb is different in several ways from the one faced by hotels and airlines. 49 As opposed to hotels, which have several types of standard rooms for which rate structures can be constructed, each Airbnb listing is unique, which makes pricing with standard rate structures difficult. As with other major hotel chains, Airbnb also does not control the prices that its hosts set for their properties. Since Airbnb makes more money when their hosts charge higher prices, the company has an incentive to recommend to hosts 43 Valentino-DeVries, J., J. Singer-Vine, and A. Soltani Websites Vary Prices, Deals Based on Users' Information. The Wall Street Journal. 24 Dec Ibid. 45 Mattioli, D On Orbitz, Mac Users Steered to Pricier Hotels. The Wall Street Journal. 23 Aug Collin, L Same Flights, 2 Different Prices: Frequent Flyer Discrepancies. WCCO Minnesota. 15 May Useem, J How Online Shopping Makes Suckers of Us All. The Atlantic. May Hill, D The Secret of Airbnb s Pricing Algorithm. IEEE Spectrum. 20 August

45 prices that adapt to the market conditions in each part of the world where Airbnb operates. Airbnb developed a continuous pricing recommendation system called Aerosolve to generate prices to suggest to hosts. Aerosolve takes into account macro-level variables like aggregate market demand, seasonality, and the presence of hotel competition to micro-level details like neighborhood, host reviews, and property features to suggest prices to hosts. 50 It is ultimately up to the host to decide whether to use the Airbnb recommended prices or whether to generate their own prices, though a feature called Smart Pricing allows hosts to let Airbnb handle price adjustments, dynamically adjusting room rates up or down in response to market conditions. FIGURE 3.7: Airbnb Aerosolve Dashboard for hosts (Source: Airbnb) Aerosolve uses machine learning to dynamically adjust the weights its pricing model gives to different attributes of a listing, and to identify similar listings as new units get posted to the site. 51 The dynamic nature of Airbnb s price recommendation engine gives it the potential to be significantly more sophisticated than the simpler fixed-price assortment optimization heuristics currently used by large hotels, although it ultimately remains up to the hosts themselves to decide if they want to accept Airbnb s dynamic recommendations. On-Demand Ride Hailing Services: Uber and Lyft Uber and Lyft, the on-demand ride hailing services accessed through smartphones or other mobile devices, are now ubiquitous in major cities throughout the world. Despite losing $2.8 billion 52 in in 2016 alone, Uber, the world s largest ride hailing service, and its primary competitor Lyft (which lost $600 million 53 in 2016) have revolutionized personal transportation and placed enormous pressure on incumbent taxi operators. Uber and Lyft have also put into 50 Ibid. 51 Ibid. 52 Fiegerman, S Uber Lost $2.8 Billion Last Year. CNN Tech. 14 April Newcomer, E Lyft Loses $600 Million in 2016 as Revenue Surges. Bloomberg. 12 January

46 place perhaps the most sophisticated pricing systems that an average consumer is likely to interact with on a daily basis. Most of the details of Uber and Lyft s pricing algorithms are closely-held corporate secrets. However, some information has started to become public. Uber began with a relatively simple, taxi-inspired fixed tariff schedule, as shown in Figure 3.8. However, the company became the focus of intense media scrutiny as a result of its surge pricing algorithm. Surge pricing multiplies the normal fare by a surge multiplier during periods of high demand, making it an example of a dynamic price adjustment mechanism using the taxonomy developed in Chapter 2. While Uber s normal fare is determined by a fixed pricing schedule, surge pricing allows Uber to adjust prices in real time in response to market conditions. FIGURE 3.8: Uber s tariff schedule for the San Francisco market in 2013 (Source: Uber) In 2016, several notable changes were made to Uber s pricing algorithm and the ways in which prices were displayed to customers. First, Uber introduced upfront fares. Instead of providing a range of prices that would be finalized at the end of the ride according to traffic conditions, customers were presented with a fixed price that would apply regardless of traffic. At the same time, the app made it less obvious when surge pricing was in effect. With upfront pricing, the surge icon was reduced in size to a small badge and a single sentence indicating that surge pricing is in effect. Even with upfront pricing, the underlying pricing schedule was still used to compute the fares that were charged. That is, two riders traveling exactly the same distance in the same traffic conditions would be charged the same fare. However, in 2017, Uber made another, more quiet change to its underlying pricing algorithms in certain cities. With so-called route-based pricing, Uber removed the underlying distance- and time-based pricing structure in some markets and began computing prices for each route purely dynamically. Route-based pricing thus represents a shift from dynamic price adjustment to transactional continuous pricing. The existence of route-based pricing was first revealed by a Bloomberg story in May, 2017 entitled Uber Starts Charging What It Thinks You re Willing to Pay. According to the article, fares under route-based pricing are based not only on time and distance, but also an estimate of passenger willingness-to-pay: Daniel Graf, Uber s head of product, said the company applies machine-learning techniques to estimate how much groups of customers are willing to shell out for a ride. Uber calculates 45

47 riders propensity for paying a higher price for a particular route at a certain time of day. For instance, someone traveling from a wealthy neighborhood to another tony spot might be asked to pay more than another person heading to a poorer part of town, even if demand, traffic and distance are the same. 54 Uber has not released any official details about the factors that drive route-based pricing. In a separate article following the release of the Bloomberg story, Uber Head of Product Daniel Graf denied that the algorithm was personalized, saying it has nothing to do with the individual and that, according to the author of the article, anyone traveling the same route at the same time should get the same price. 55 However, if Uber were to move towards personalized pricing, it would represent one of the most sophisticated advances in price technology in recent decades. More importantly, Uber has begun to condition users to accept constantly changing prices for an identical ride depending on the traffic and demand conditions when they open the app. While Uber has not experimented with personalized discounting, such a feature could also potentially spur demand. However, if personalized pricing was eventually implemented, it could threaten to change customer behavior. A Bloomberg columnist suggested what would likely be a common line of thinking for technologically-sophisticated users: As Uber starts price discriminating, we can try to trick its algorithms into believing that we have low willingness to pay. How? By injecting some false signals into the data stream. For example, we could open the app at random, check the prices on routes we sometimes take, and then close the app without calling for a ride just to make ourselves look price-sensitive. We could further baffle Uber s estimates of our ride preferences by considering (and sometimes taking) rides to unusual locations If Uber s customers start trying to game the pricing algorithms, that could throw off Uber s demand estimates, leading to uncertainty not only in pricing but also in availability exactly what surge pricing was set up to avoid. 56 As described in Chapter 5, any company considering a move towards more dynamic or personalized prices will have to confront possible user behavior as described above. Furthermore, the legality of personalized pricing is likely to be tested in court, particularly if it becomes evident that price discrimination is occurring (either on purpose or by accident) along protected categories, such as age, sex, or ethnicity. Conclusions While the particularities of how firms set prices depend on the context of the industry, there are nevertheless many similarities between the pricing methodologies of the airline industry and those used in other industries: Many industries, including airlines, hotels, passenger rail, pre-select a fixed number of possible price points and then select which of these price points to display for each 54 Newcomer, E Uber Starts Charging What It Thinks You re Willing to Pay. Bloomberg. 19 May Carson, B Uber may charge you more based on where you're going. Business Insider 20 May Kominers, S.D Uber s New Pricing Idea is Good Theory, Bad Business. Bloomberg View. 13 June

48 shopping request. These pricing structures may or may not be publicly available. Firms in these industries then use assortment optimization and revenue management techniques to decide which of the pre-set prices are displayed to a customer at a given time. Prices can move up or down depending on forecast demand-to-come and remaining capacity, but the total number of price points are limited and pre-defined. Of the industries surveyed, the airline industry is uniquely limited by the constraints of legacy distribution technologies and industry standards. It is also relatively unique in that all of its price structures are publicly filed and visible to competitors. Only a small number of companies seem to practice dynamic price adjustment or continuous pricing. The on-demand transportation app Uber now practices route-based pricing in which dynamic prices are generated on the fly for each trip request, and can change from second to second. Some sellers on the Amazon Marketplace have also been observed to practice algorithmic, continuous pricing of their products. From this survey, airline pricing and revenue management remains quite complex and sophisticated as compared to other industries. Of the industries surveyed, only Uber, Amazon, and Airbnb appear to use pricing techniques that are more sophisticated than those currently being used by airlines today. Unlike the airline industry, these industries practicing advanced pricing methods are largely unconstrained by legacy technological systems and distribution standards. In the airline industry, new pricing mechanisms are already under development that could give the airline industry best-in-class pricing capabilities relative to other industries. In Chapter 4, we introduce and discuss some of these new developments in next-generation airline pricing and revenue management and describe how they are related to practices used in these industries. 47

49 4. Next-Generation Pricing Mechanisms for the Airline Industry As discussed in Chapters 1 and 2, pricing and revenue management practices in the airline industry have traditionally involved pre-selecting a list of possible price points and then determining which of those price points to make available in response to shopping requests. While these mechanisms provide airlines with opportunities to segment demand and practice differentiated pricing, they do not allow for the type of transactional, continuous pricing used in some of the industries discussed in Chapter 3. Airlines and technology vendors have continued to invest in and develop new mechanisms for pricing that could move the airline industry closer to transactional continuous pricing, in which prices are computed in real-time on a request-by-request basis. The jump from the status quo of assortment optimization to next-generation continuous pricing is unlikely to happen all at once, considering the money and time airlines have spent in developing, maintaining, and fine-tuning legacy systems. Instead, recent innovations in pricing mechanisms generally aim to accomplish one or more of three incremental goals on the path to continuous pricing: Expanding the set of price point options that can be selected during assortment optimization. Increasing the velocity at which this set of price points is updated in response to market conditions. Increasing the frequency at which prices are selected from the set of possible price points to display to customers. FIGURE 4.1: Main goals of next-generation pricing mechanisms In this chapter, we discuss six new pricing mechanisms that are currently under development in the airline industry and how they serve to accomplish one or more of the incremental goals in Figure 4.1. We focus mostly on describing the mechanisms themselves and how they could change the ways in which airlines price and distribute their products, as opposed to the scientific algorithms that determine which prices are selected at any given moment or for any given transaction. In Chapter 5, we discuss in more detail the potential effects of these new mechanisms on airline revenues, competition, and existing processes. More Frequent Updating of Price Points Historically, airlines updated the price points filed with central organizations such as ATPCO relatively infrequently in response to changes in competition and/or other market conditions. As fare filing technology has improved, and as airlines have become more adept at incorporating market and competitor information into their pricing structures, the speed at which prices are refiled has increased in recent years. As discussed in Chapter 1, the number of fares filed with ATPCO each day increased six-fold from 669,000 to 3.9 million over the last 10 years. 48

50 By updating their fare structures more frequently, airlines can increase the number of price points at which tickets are sold in any given market, while continuing to use existing revenue management architecture to select prices from the fare structure. However, at any given time, the revenue management system still selects prices from a relatively small number of possible price points. There are a number of market conditions that could induce an airline to adjust its set of possible price points. Airlines could choose to adjust their fare structures in order to react to changes in competitor fares, to account for an unexpected increase in bookings or search volumes, or to reflect changing corporate strategies. 57 In response to these triggers, airlines could choose to adjust a small subset of fare levels in a particular market (for instance, only the least-expensive leisure fares), all the fares in the market, or fares across a swath of markets in a particular geography or in which the airline competes with a particular competitor. Frequent price changes can be onerous for airline pricing analysts, who must maintain and file accurate lists of price points and ensure that other downline airline processes are aware of any changes in filed fares. One airline has estimated that these mechanical processes can take up to 80% of a pricing analyst s time, leaving less time available to monitor and respond to market conditions. 58 As a result, some airlines and vendors have started moving towards fully automating the filing of fares. This mechanism allows for prices to be refiled at a more rapid pace, while freeing up pricing analysts to focus on changes in market conditions instead of the mechanics of properly submitting fares for filing. One example of automated rapid fare filing is quantum pricing, which was developed by Etihad Airways. Quantum pricing allows the airline to rapidly adjust its filed fares using automated technology to send those fares to ATPCO for filing. 59 Fares could be changed manually by an analyst setting business rules, or automatically in response to changes in the marketplace. At the limit, a different fare structure could be generated automatically and repeatedly updated for each departure day in each market. It should be noted that mechanisms like quantum pricing do not require significant changes to any of the underlying mechanisms of existing pricing and revenue management systems. A predefined list of price points is still filed with ATPCO for each market (albeit at a more rapid pace), and airlines still make inventory availability decisions among this pre-set list of price points through their revenue management systems. Etihad sees its quantum pricing as a baseline technology can that be used for more advanced pricing mechanisms, and the underlying price recommendations from quantum pricing could be likely adapted if the industry were to begin moving away from pre-filed fares in favor of alternative applications enabled by NDC. Dynamic Availability of Fare Products With more frequent filing of fares, airlines can rapidly update their sets of possible price points in each market, but still use existing revenue management techniques to select which of these 57 Konanki, R., B.R. Guntreddy, O. Oancea, and G. Komirishetty Quantum Pricing: Dynamic Pricing in Action. Proceedings of the 2017 AGIFORS Revenue Management Study Group Meeting, San Francisco, CA. 58 Ibid. 59 Ibid. 49

51 price points to offer. Typically, revenue management systems periodically calculate the number of seats available at each price point; these optimization decisions can be made multiple times per day, but are usually not based on characteristics of individual customers or transactions. Dynamic availability mechanisms work by increasing the frequency at which prices are selected from the list of pre-defined price points. These mechanisms work by overriding the availability decisions made by the airline s revenue management system in response to transaction-specific information or pre-defined business rules. For example, the airline could decide to make a particular fare product available for customers booking with a specific travel agent when the load factor for the flight is less than 50%. In instances in which these conditions are not met, the normal availability as determined by the RM system is distributed. Since RM systems use aggregate data to make their availability decisions, incorporating transaction-specific information to adjust the RM system recommendations could improve revenue outcomes for certain booking requests. Various technology vendors have created mechanisms for an airline to practice rules-based dynamic availability. Each airline may also develop its own internal mechanisms for dynamic availability, using business rules or scientific algorithms to determine when to override the recommendations made by their RM systems. As with more frequent fare filing, dynamic availability mechanisms retain much of the existing pricing and revenue management architecture that is currently used by the airline industry. Airlines can still use existing revenue management systems to determine baseline availability, and internal pricing processes can remain relatively unchanged with dynamic availability. However, with dynamic availability, airlines are still limited to selecting among a relatively small list of pre-defined price points. Additional RBD Capabilities In 2015, ATPCO convened a Dynamic Pricing Working Group among a group of major airlines and technology providers to explore new pricing technologies and define new standards and mechanisms for pricing communication. Unlike frequent fare filing and dynamic availability, the proposed solutions that have emerged out of this Working Group may require some changes to existing pricing, revenue management, and distribution processes and technologies. One of the two solutions investigated by the ATPCO Dynamic Pricing Working Group addresses the limitation of available price points due to the number of RBDs that a carrier has available. As discussed in Chapter 1, airlines typically assign a single alphabetic character to each Reservation Booking Designator (or RBD). Fare class availability is indexed to this letter, meaning that a maximum of 26 possible fare classes can be optimized and made available at any given time in each market. Some RBDs are used for internal airline purposes such as staff travel, further reducing the number of price points available for distribution. As airlines have increased the types of products and services they sell in the marketplace, 26 price points may not be sufficient to ensure that a wide range of prices can be offered for each product. The goal of the Working Group is to find a solution that increases the number of available price points that can be offered and provide airlines more granularity in fare offerings. This can be accomplished through current RBD processing leveraged with additional capabilities that will 50

52 allow airlines to divide a single RBD into multiple products that can be controlled independently. Airlines would continue to select from the pre-defined fare structure when determining which prices to make available in the marketplace. Dynamic Pricing Engines The other new pricing mechanism that has emerged from the ATPCO Dynamic Pricing Working Group is the Dynamic Pricing Engine (DPE). DPEs work by applying dynamic price adjustments to the pre-filed prices that would ordinarily be offered by the airline s RM system. Since the amount of the adjustment could vary from transaction to transaction, the DPE is the first nextgeneration pricing mechanism discussed in this chapter that allows airlines to offer prices that are not included on a list of pre-determined price points. FIGURE 4.2. Schematic of ATPCO Dynamic Pricing Engines (DPEs) (Source: Adapted from ATPCO Dynamic Pricing Working Group) A general schematic for dynamic pricing engines as proposed by the ATPCO Working Group is shown in Figure 4.2. The DPE is housed at the airline, and the business rules that determine the timing and amount of any dynamic adjustments are also proprietary to each airline. During the shopping process, the DPE is queried to determine if the fare that would normally be offered in response to the shopping request should be adjusted up or down, or whether no adjustment should apply. The DPE-adjusted fare (if applicable) is then sent to the customer. During the pricing phase, the DPE is queried again to ensure that the quoted price adjustment is still valid. Each airline using a DPE will have its own rules engine that determines when and how the DPE will adjust the lowest-available fare. This engine could be based on pre-defined business rules designed by analysts or managers for instance, a discount of 5% off the lowest-available price could be offered to transactions that meet certain requirements. DPEs could also use scientific algorithms to compute price adjustments based on transactional information, demand forecasts, or customer segmentation. With either a rules-based or the science-based DPE, the DPE may at times suggest that no adjustment be made to the pre-filed fare. 51

53 Since the DPE makes its adjustments in reference to pre-defined price points, and since some customers may be offered unadjusted prices, a DPE can be classified as a dynamic price adjustment mechanism using the taxonomy introduced in Chapter 2. Depending on the sophistication of the DPE, price adjustments could be made relatively infrequently (e.g., daily), or at the request-by-request level (transactional). To ensure that DPE-adjusted fares can be tracked and verified, the ATPCO Working Group has proposed filing all DPE adjustments as private fares with ATPCO. In this way, these fares could be validated in downline processes such as revenue accounting. It is not yet clear whether DPE-adjusted fares will be visible to other airlines, or if they will remain opaque. As will be discussed in Chapters 5 and 6, the level of opacity of dynamic price adjustments could have significant repercussions for airline competition. DPEs rely on an existing airline revenue management system to determine the lowest-available price point that would ordinarily be offered for each transaction; this in turn requires a relatively small list of pre-selected price points. In other words, existing pricing and revenue management technologies could be used to complete the assortment optimization process as is done today, with a DPE making a last-second adjustment to the price in response to contextual information about the transaction or market conditions. Since many underlying processes and technologies can remain intact, DPEs could provide a solution to airlines that wish to move towards continuous pricing. After two years of discussion in the ATPCO Dynamic Pricing Working Group, development of DPEs has moved to a pilot phase with airlines and vendors working together to determine specifications and data requirements. 60 Continuous Pricing As discussed in Chapter 2, continuous pricing allows firms to select prices from a range of possible values, as opposed to selecting among a relatively small assortment of pre-chosen price points. Continuous pricing could give airlines flexibility to rapidly select and update prices from transaction to transaction. This practice could require significant changes to underlying pricing and RM processes. Mechanically, continuous pricing could coexist with existing fare filing processes if airlines filed a separate fare product for each unit of currency in the range of allowable values. Airlines could then use a continuous pricing engine to select which precise price to make available. This could be a cumbersome process, as the sheer number of possible currency values would require significant improvements to the capabilities of existing RBDs. Alternatively, the development of the New Distribution Capability means that airlines may not necessarily need to retain pre-filed fare structures when practicing continuous pricing. With next-generation distribution technologies, airlines could potentially select prices freely from a range of possible values without a traditional fare filing associated with that value. Airlines implementing continuous pricing may choose to use either a traditional fare product structure or a NDC-style mechanism for implementation. 60 Ratliff, R Industry-standard Specifications for Air Dynamic Pricing Engines: Progress Update. Proceedings of the 2017 AGIFORS Revenue Management Study Group Meeting, San Francisco, CA. 52

54 In either case, continuous pricing would require new revenue management optimization and forecasting approaches. For instance, current revenue management forecasting practices typically center on estimating demand-to-come for each pre-priced fare class. With continuous pricing, a forecasting approach based on customer willingness-to-pay, demand elasticities, or buy-up probabilities would likely be necessary to forecast the demand at each price point. Some airlines and RM system vendors have already incorporated these types of advanced forecasting methods into their RM systems, but additional work would be necessary to move these methods from the world of fare classes to the world of continuous pricing. Existing revenue management optimization practices are also centered around the fare class, as they perform assortment optimization to select which of a small number of fare classes to make available. New optimization methods would be required for continuous pricing to select an optimal price given the forecast of demand to come at various price levels. The Massachusetts Institute of Technology s PODS Revenue Management Research Consortium is currently investigating new revenue management forecasting and optimization methods for so-called classless RM, which could serve as a precursor for continuous pricing. The investigation is still in its early stages, and it is unclear how such an approach could co-exist with filed fare classes and established RBDs, for example, if an airline needs to retain these fares for specific markets, distribution outlets, or types of fare products. Dynamic Offer Generation Each of the pricing mechanisms discussed in this section focuses exclusively on the price selection process, given a choice of product. The most advanced next-generation mechanism we will discuss could join together the product creation process with the price selection process. We call this mechanism dynamic offer generation. FIGURE 4.3: Schematic of Dynamic Offer Generation A schematic of dynamic offer generation is shown in Figure 4.3. Each offer in the offer set consists of an itinerary, a set of zero or more ancillary services, and an offer-specific price. Offers could vary from request to request based on the characteristics of the customer making the request, or on the characteristics of the request itself. The offer-specific price could be 53

55 selected from a list of pre-defined possible price points, or calculated from among a continuous range of possible prices. Note that dynamic offer generation could combine continuous pricing with transactional product selection. Each search request could result in a customized offer with a combination of relevant ancillary products and a price that is selected in real time from a continuous range. It is also possible for airlines to pre-construct a small number of possible offers, and then decide which of these options to display to each customer. These pre-constructed offers could also be dynamically priced. The science that would drive a dynamic offer generation engine has yet to be fully developed, and is likely to be quite complex. As airlines offer more ancillary services, the number of possible combinations of itineraries, services, and prices has continued to expand exponentially. From a technological perspective, dynamic offer generation would also require significant changes to existing pricing, revenue management, and distribution technologies, as well as updates to other airline downstream processes such as revenue accounting and booking and settlement. NDC could provide a framework for disseminating these changes on the distribution level, but internal airline process changes to enable dynamic offer generation will likely require a significant amount of new investment, time and effort and will vary from airline to airline. Conclusions In this chapter, we introduced six next-generation pricing mechanisms currently under development in the airline industry. These mechanisms aim to move airlines incrementally from existing assortment optimization closer to transactional continuous pricing. Each of the new mechanisms aims to increase the number of possible price points, the velocity at which this set of price points is updated, and/or the frequency at which prices are selected from the list of price points. The more complex mechanisms, such as dynamic pricing engines and continuous pricing, also allow for prices outside of the pre-defined list to be offered, either by making adjustments to pre-filed price points or by selecting prices from a continuous range. The mostcomplex mechanism, dynamic offer generation, would allow airlines to combine the product creation and price selection process into a single step. The mechanisms introduced in this section vary in technical complexity. Some mechanisms maintain most legacy systems and technologies and retain pre-filed fare structures for use in existing revenue management systems. Others, such as dynamic offer generation, would require widespread changes to existing processes. Continuous pricing and dynamic offer generation could allow airlines to move away from pre-filed fares, although these mechanisms would still be compatible for use with pre-filed fare structures. The degree to which airlines will implement these next-generation pricing mechanisms depends on each airline s tolerance for risk and disruption along with the possible revenue gains that these mechanisms could provide. The adoption of these mechanisms by one or more airlines could have significant impacts not only on that airline s revenues and costs, but also on the nature of airline competition throughout the industry. The possible effects of new pricing practices remain a controversial topic, and there is no consensus that a move from assortment optimization to continuous pricing will be a net benefit for the industry. 54

56 5. Implications of Next-Generation Pricing Mechanisms for the Airline Industry The next-generation pricing mechanisms discussed in Chapter 4 are new and speculative technologies that have yet to be widely applied in the airline industry. If implemented, these practices could have considerable impacts on many facets of the industry, including airline revenues, competition, downstream processes, regulation, and consumers. Specifically, next-generation pricing practices could disrupt many features of traditional airline pricing and RM that are often taken for granted. These include the accuracy and interoperability of current processes in exchanging information and calculating fares across airlines and channels using different technologies, accounting systems, and payment mechanisms. Current systems are also fairly stable and transparent, and support a variety of pricing tactics such as promotional pricing, group sales, and corporate contracts. Customer interactions with nextgeneration pricing mechanisms may also be different from with current systems, in which customers may expect prices to generally increase over time or with the flexibility of travel. Next-generation pricing mechanisms will result in changes to each of these attributes of the current model. In this chapter, we discuss in detail some of these possible implications of nextgeneration pricing on various facets of the airline industry, customers, and regulators. Airlines deciding whether to implement next-generation pricing will need to consider each impact and assess whether the benefits of such systems outweigh any potential negative effects. Implications for Airline Revenues There are two main reasons why an airline may expect that next-generation pricing methods would lead to an increase in total revenues. First, these methods may offer higher prices than traditional pricing and RM methods in some situations, leading to an overall increase in average revenue per booking. In other words, next-generation pricing strategies could produce revenue gains by closing the gap between prices and customers conditional willingness-to-pay. On the other hand, new pricing methods may offer lower prices relative to traditional methods in certain situations. By offering lower prices, airlines could hope to stimulate new bookings, or to capture customers that otherwise would have booked with a competitor. Increasing or decreasing prices from the base case of traditional pricing and RM could lead to both rewards and risks to airline revenue. To measure the relative effects of these practices, a number of academic studies have used simulations to test various new pricing and RM methods. Depending on the mechanism, the competitive environment, and the amount of information that is available to the airline, studies have shown revenue gains of between 0.5 and 6 percent for next-generation pricing and RM over traditional methods. 61,62,63 In comparison, the 61 Zhang, D. and Z. Lu Assessing the Value of Dynamic Pricing in Network Revenue Management. INFORMS Journal on Computing 25(1): Fiig, T., O. Goyons, R. Adelving, and B. Smith Dynamic pricing The next revolution in RM? Journal of Revenue and Pricing Management 15(5): Wittman, M.D. and P.P. Belobaba Customized dynamic pricing of airline fare products. Journal of Revenue and Pricing Management. Forthcoming. 55

57 revenue gains from moving to a traditional leg-based revenue management system to a more complex network RM system have been estimated through simulation to be between 0.5 and 2 percent. 64 Several of these studies evaluate the performance of new pricing practices using the Passenger Origin-Destination Simulator (PODS). PODS simulates the interactions between passengers, who desire to travel in a particular air transportation market subject to a budgetary constraint and preferences regarding their preferred travel times and fare restrictions, and airlines, which offer networks of flights, generate itineraries, and use revenue management methods to determine the prices to be charged at any given time. By comparing next-generation pricing methods with base cases in which traditional pricing and RM is used, researchers can estimate the revenue performance of the new methods and explore their impacts on simulated airline load factors, yields, and booking mixes. In a series of recent papers, MIT researchers have developed and tested a number of dynamic pricing engine (DPE) heuristics in PODS. 65 These heuristics dynamically adjust the price that would ordinarily be offered by the airline s revenue management system by taking into account demand segmentation (i.e., whether the airline thinks the booking request is coming from a high-budget business passenger or low-budget leisure passenger) and the airline s estimate of passenger willingness-to-pay for each segment. FIGURE 5.1: PODS Simulation of Percent Change in Revenue from Base when One Airline Uses Increment-Only or Discount-Only DPE Heuristics at Various Demand Levels 64 Belobaba, P Optimization models in RM systems: Optimality versus revenue gains. Journal of Revenue and Pricing Management 15(3-4): Note that PODS results are dependent on the network and simulation parameters with which they are run. Furthermore, PODS assumes a steady state in respect to seasonality and competitive schedule, with an underlying fare structure that is shared by all competitors and left unchanged through each simulation run. Changes to these underlying assumptions could result in different results. 56

58 Figure 5.1 shows the revenue impacts of several DPE heuristics when used by a single airline (AL1) in a competitive, four-airline network in various demand scenarios. 66 On the left side of the figure, booking requests identified as belonging to the high-wtp business segment are eligible for increments; on the right, booking requests identified in the low-wtp leisure segment are eligible for discounts. An algorithm is used to determine the amount of any dynamic price adjustment, as well as when no adjustment should be applied. Depending on the simulation environment, these two heuristics result in revenue gains for AL1 of between 0.4 and 3.7 percent over a base case in which it uses traditional pricing and RM methods (and assuming all competitors continue to use traditional methods). The revenue gains from using DPE heuristics in these simulations come from several sources. For the incrementing heuristic, revenue gains come from charging higher prices in certain situations to high-wtp customers. Some customers still choose to book at these higher prices, which increases the airline s revenues and yield. However, some customers that face incremented prices will instead choose to book with a competitor, or decide not to travel. As a result, the airline that is periodically incrementing prices loses overall bookings and sees its load factor decrease. This decline in load factor is not surprising in response to a rise in overall price levels. The fact that AL1 s revenue increases despite a reduction in load factor suggests that the increase in passenger yield outweighs the impacts of lost bookings. Airlines that are not incrementing also see their revenues increase in this situation, since they are able to capture some passengers who face higher prices from AL1. For the discounting algorithm, the airline sees revenue gains since it is able to both capture some passengers from other airlines and stimulate new bookings from customers who otherwise would have decided not to travel. These new passengers increase the airline s load factor. It is possible that the heuristic will at times provide discounts to customers that would have booked at the normal price. Yet despite the fact that only discounts are provided, the airline also sees an increase in its passenger yield, or revenue per mile flown. This is because this heuristic typically provides discounts to lower-wtp customers only in situations when prices are relatively high. By generating new bookings at relatively high price points, the airline s revenue management system will begin protecting more seats for laterarriving, high-paying customers, which increases yields. Other studies using the PODS revenue management simulator have tested the effects of adding additional price points to the airline s fare structure while continuing to use traditional RM practices to control availability. These new prices are inserted in the gaps between existing price points: for instance, if the airline s fare structure contained price points of $100 and $200, a new price point of $150 could be introduced to increase the number of possible prices. This method is analogous to the additional RBD pricing mechanism discussed in Chapter In this simulation, Low Demand is associated with an average load factor of about 77%, Medium Demand is associated with an average load factor of about 83%, and High Demand is associated with an average load factor of about 87%. 57

59 Figure 5.2 shows the incremental gains in revenue as airlines using advanced forecasting methods both move from six price points to 11, 16, and 21 price points in a competitive, twoairline network. In this scenario, adding new price points does marginally increase revenue, but these revenue gains show diminishing marginal returns to scale. This is because adding new price points leads to two opposing effects when used in conjunction with an advanced revenue management system. FIGURE 5.2: PODS Simulations of Incremental Revenue Gains of Additional Fare Classes (6 Fare Classes to 21 Fare Classes) First, new price points can lead to buy-up for some customers that would have otherwise booked at a lower price point. For instance, if a price point of $150 is added between existing price points of $100 and $200, some passengers who would have ordinarily paid $100 will still book at a price of $150, leading to a $50 increase in revenue. Additionally, the new price points can stimulate new bookings from customers who would have otherwise chosen to no-go. Conversely, new price points can also damage revenues as a result of buy-down. The presence of the $150 price point means that some customers who would have paid $200 in the base case will be able to book at a price that is $50 lower. Also, if the presence of the $150 price point means that the $100 price point is offered less often by the RM system, the airline may lose some customers who would choose to travel at $100 but not $150. In this simulation, combining all of these effects leads to an increase in passenger yield and a decrease in load factor as the airline adds more possible price points. This suggests that the buy-up effects may dominate the buy-down effects in this particular simulation. However, we do not see the increases in both load factor and yield that were present in the tests of the discounting DPE heuristic. The results of these simulations suggest that there may be a point at which adding additional price points does not lead to significant increases in revenue. In this particular simulation, 58

60 offering either 16 or 21 price points leads to approximately the same revenue. It should be noted that the price points in this example are still fixed throughout the course of the simulation and controlled using traditional revenue management practices, unlike a dynamic pricing environment. Implications for Airline Competition The revenue results described in the previous section assume that the next-generation pricing methods are used by a single airline. As airlines become more technologically advanced and as vendors develop more sophisticated technologies, it is reasonable to expect that these techniques will be adopted by many airlines across the industry. A shift in underlying pricing technology could lead to significant changes to the ways in which airlines compete in the marketplace. Some of these effects can be estimated through simulation, while others depend on the economic and psychological reactions of airlines to the actions of their competitors. FIGURE 5.3: PODS Simulation of Percent Change in Revenue from Base when All Airlines in the Simulation Use an Increment-Only or Discount-Only DPE Heuristic We will briefly consider the use of the DPE heuristics discussed in the earlier section by multiple airlines. Figure 5.3 shows the revenue results if all four airlines in the simulated network 67 use the incrementing or discounting heuristics discussed in the previous section. Note that if all of the airlines in the simulation use either DPE heuristic, all airlines see increases in revenue over the base. These increases come from the same sources as in the previous section the incrementing heuristic causes an increase in yield, and the discounting heuristic results in demand stimulation, particularly in higher fare classes. Yet these simulation results do not consider the possible psychology of competitive interactions 67 The simulation network is a generic situation that is not intended to represent any particular airline or competitive environment. Any results from PODS are valid for that combination of simulation parameters only. 59