A Research Framework for Business Models: What is common

among fast fashion, e-tailing, and ride sharing?

Gérard P. Cachon∗

September 16, 2018

Abstract

Every firm has a business model, which is the collection of strategic decisions that deter-
mine how the firm generates a sustainable enterprise through the creation of enough value (its
supply model) and the extraction of a sufficient portion of that value (its revenue model). In-
novative business models, for example fast-fashion (e.g., Zara), e-tailing (e.g., Amazon), and
ride-sharing (e.g., Uber), are capable of offering new products and services that generate con-
siderable consumer utility and transform industries. This paper develops a research framework
for understanding business models and how business models have evolved over time. Links are
made to the existing literature (primarily in pricing and operations) and simple models are
developed to unify and clarify existing research findings. Through this framework it is possi-
ble (i) to identify the few design decisions that explain the success of these diverse firms with
otherwise seemingly disparate models, and (ii) to speculate on potential future business model
innovations.
Keywords:

1 Introduction

“Business model” and its offspring “business model innovation” are buzzwords that have emerged
to describe the upheaval that has occurred in many markets. Examples of firms with transformative
business models include Zara with apparel (fast fashion), Amazon with e-commerce retailing (e-
tailing), and Uber with transportation (ride sharing), among many others. These changes have
∗
University of Pennsylvania, [email protected]. The author thanks Robert Swinney, Jan van Mieghem,
Marty Lariviere, the participants of the Kellogg Workshop, and the editorial team for their helpful feedback on earlier
versions of this manuscript.

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
led to novel sources of consumer value, remarkable financial opportunities for entrepreneurs, and a
bounty of new phenomena and practices for academics to explore.
The first goal of this paper is to construct a framework for conceptualizing business models in
general, and innovations in business models in particular. Although it might appear that business
models come in a vast array of different “shapes, sizes and colors”, it is argued that all business
models are defined by a relatively small set of high-level strategic decisions, each with its own basic
tradeoffs. The second goal of this paper is to use this business model framework to highlight the
linkages that exist among the growing set of research papers on various facets of business models
- academics (appropriately) focus on narrow issues, but combining the various pieces forms a
larger picture. By understanding the common threads among seemingly disparate business models,
we hopefully are better able to evaluate the potential of a model and possibly even predict the
emergence of successful new models.

2 Business Model Definition and Framework

A business model is the set of strategic decisions that determine how a firm generates a sustainable
and successful enterprise. There are three points worth emphasizing regarding this definition. First,
a business model is not a single decision, but rather a set of decisions which must be considered as
a collection. Second, a business model is based on strategic decisions, i.e., decisions that have broad
implications for the firm, are difficult to change and often are made when the firm is established,
thereby defining the essence of the firm and endowing it with its distinctive character. For example,
the lack of a physical storefront has been a defining feature of Amazon’s business model. Third, the
emphasis with a business model is on how the firm does its business rather than what the firm offers
its customers. The emphasis on “how”, rather than “what”, might seem misplaced because the
success of many firms can be traced to what they offer their customers. For instance, Microsoft’s
Windows long dominated the operating system market for personal computers, and Pfizer’s choles-
terol reducing product, Lipitor, can safely be described as the blockbuster of blockbusters in the
pharmaceutical industry. Firms do indeed thrive if they are lucky enough to have a distinctive
and valuable product, especially if they have some legally sanctioned monopoly protection (e.g.,
patents and trademarks). But a great “what” is not the only way to succeed. Zara is not known for
any one particular product or style. Its substantial and remarkable revenue growth has come more
from how it delivers the products it sells rather than what specific products it offers. Similarly,

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
 Figure 1: The components of a business model

Uber’s innovation is not in what it offers (transportation) but rather how it offers it (independent
contractors and dynamic pricing).
To add another layer of detail, the firm’s set of strategic decisions in its business model can
be divided into two parts, as illustrated in Figure 1 - a revenue model and a supply model. The
revenue model is how the firm earns the revenue used to procure resources (e.g., labor and capital).
The supply model is how the firm manages and utilizes its resources to generate customer value.
Customers value having their needs satisfied and doing so with transactional efficiency (e.g., a
simple, fast, and easy process). For the firm to be sustainable over time the amount of customer
value created must be sufficient to enable enough revenue to be generated to cover the cost of the
resources needed to produce the value.1

2.1 Customer value and resources

Customer value is generated when the firm effectively satisfies a need, with as little hassle as
possible, either through a physical product or a service, or both. (For simplicity, “product” is used
to describe physical goods as well as services.) For example, a customer’s need can be as mundane
as a box of pasta to make a meal, or as specific as transportation to move from one part of a city
1
There are alternative frameworks for business models, such as the Business Model Canvas
(https://en.wikipedia.org/wiki/Business Model Canvas). The Business Model Canvas creates 9 dimensions
that managers can use to describe their business. The emphasis is more on the details of each of these dimensions
rather than on high-level strategic decisions and their interactions. Nevertheless, all frameworks for business models
include some element of how firms earn revenue and how they produce value.

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
to another, or as elaborate as medical care. Needs are usually multi-dimensional, as in tasty gluten
free pasta, or quick and safe transportation, or effective and friendly medical care. The quality of
a product, both in terms of performance quality (an absolute measure) and conformance quality
(consistency in meeting specifications), influences how effectively the need is satisfied.
While it is important to satisfy a need, customers also want their interaction with the firm to
be as simple, convenient, and pleasant as possible. In other words, customers value transactional
efficiency. This includes the time and effort to search, evaluate, select and receive the firm’s
product. For example, online transactions are easier than phone transactions, nearby locations are
more convenient than far away locations, and 1-day shipping is preferred over 5-day shipping. A
need well satisfied, but provided with a transactionally inefficient process, is of little value to a
consumer. As such, transactional efficiency is a critical element in the customer value proposition,
and one that plays an important role in many recent business model innovations.
The actual creation of value is done through the coordination of an ensemble of resources. The
two obviously needed resources are labor and capital - firms generally need employees and some
physical plant and equipment to function. Less obvious is the ownership of these resources. For
example, labor could be freely provided by a firm’s customers. The actual ownership structure is
an important element of the supply model.

2.2 Revenue model

The firm’s “revenue model” describes how it earns revenue. Presuming customers value a firm’s
product, the primary role of the revenue model is to extract some of that value for the firm. For
the firm to be sustainable, the revenue extracted must be sufficient to justify the cost of the firm’s
resources.
In its most basic form a revenue model is quite simple - the firm sets a price for its product
and if a customer decides to accept the take-it-or-leave-it offer, then the customer consumes the
product. This basic model involves a single price chosen by the firm for a single product at a single
moment in time which is contemporaneous with consumption. Each of those dimensions provides
an opportunity for the firm to develop a new and more complex revenue model. To be specific, a
revenue model specifies the transactions terms that occur between the firm and customers. There
are three key three components in the revenue model: pricing mechanism (i.e., how are prices
chosen), payment structure (e.g., a fee per use or fixed fees) and dynamics (how are terms adjusted
over time).

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
 Some mechanism is always used to decide the amount that is exchanged between the customer
and the firm. With a basic mechanism the firm chooses a price to offer customers. But the firm
could let customers decide on the price, such as through an auction mechanism or via their own
take-it-or-leave-it offers to the firm. For instance, consumers could submit bids to a hotel for a
room, which the hotel either accepts or not. A firm posted price is transactionally efficient - it is
simple to communicate and can be executed immediately - but it puts the burden on the firm to
choose the “right” price based on what the firm knows about consumer preferences and its own
supply. More elaborate mechanisms, such as an auction and haggling, are transactionally more
burdensome, but facilitate preference discovery and potentially more revenue extracted.
The payment structure refers to how payments are made in the transaction. For example, a
firm selling a physical product, like a photocopier, could sell the copier with a single transaction
to a customer. Or the firm could lease/rent the photocopier to a customer for its use over an
interval of time. And with that option the customer could pay a fixed fee for the use of the copier
or a per-use fee that is proportional to the number of copies made. Firms selling services have a
similar decision. A music retailer could offer customers the option to purchase individual songs in
a library or customers could be offered access to all of the songs in the library. If access is granted
to the library, then the duration of this access must be established along with possible quantity
restrictions.
Although at a particular moment in time the firm and a customer might agree on the terms
of their transaction, there is no requirement that these offered terms remain constant over time or
that both parties stick with the initial agreement. The dynamics of a revenue model are defined by
the frequency and magnitude of changes in offered transaction terms as well as the extent to which
initial agreements can be modified. For example, does an airline increase or decrease seat prices as
the day of departure approaches? Does an apparel retailer maintain its prices or offer a discount
near the end of the selling season? Does a seller of tickets to a sporting event allow the initial buyer
to resell the ticket to another customer? These are challenging questions to analyze theoretically
and equally challenging to implement effectively in practice for a host of reasons, ranging from
mathematical complexity (i.e., can good solutions to these hard optimization problems be found),
to data quality (i.e., what data can be obtained, and are the data reliable), to consumer behavior
and perceptions (i.e., do consumers strategize, and how are they influenced by their views of fairness
and limited cognitive capacity).

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
2.3 Supply model

Like the revenue model, the supply model comes in many varieties. Nevertheless, there are three
crucial decisions in the supply model regarding how the firm manages its resources (labor and
capital): timing, location and control.
The timing of a supply model refers to when the firm activates its resources relative to customer
demand. At the most basic level, there are two choices here - either the firm activates resources
before customer demand or after customer demand. If before demand, which is often referred to
as “make-to-stock”, then the challenge for the firm is to correctly anticipate demand. If after
demand, which is often referred to as “make-to-order”, then the challenge for the firm is to respond
to demand in a sufficiently timely manner. In both cases the goal is to have high utilization of
resources while also generating customer value.
Location refers to where the firm places its resources relative to its demand. One extreme
is a limited number of locations far from demand, which allows the firm to exploit economies of
scale and low procurement costs (for inputs and resources). However, long distances to customers
require additional transportation costs and shipping time. The alternative is many locations close
to consumers, which reduces transportation costs and reduces shipping times, but also decreases
economies of scale and increases procurement costs.
Decisions on control reflect the degree to which the firm can determine the quantity, quality
and timing of the resources used to serve customers. The highest level of control is achieved when
the firm owns the resource. For example, if the firm owns a factory, then it can use the factory to
produce when it wants. If the firm “rents” a factory, then the firm must wait for when it is available
and pay the prevailing price at that time. A spectrum of control also applies to labor - the firm
has greater control over the actions of employees than contractors. In general, the key tradeoff in
the control decision is between flexibility and cost - the greater the firm’s need to directly choose
when a resource is used, how it is used and the quantity used, the firm is more likely to opt for
greater control, albeit at greater cost (usually because of lower utilization).

2.4 Innovation

While every firm operates with some type of business model (whether they can articulate it or not),
not every firm creates a business model innovation, which is a business model that satisfies a novel
need, or has a novel revenue model or implements a novel supply model. More often than not, a

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
business model innovation involves a combination of multiple novelties in needs satisfied, revenue
generation, or supply processes.
Novelty often comes from challenging implicit assumptions regarding what product is important
for a firm to offer or the process by which the firm offers its products. In doing so, a business model
innovation generally involves what could be referred to as a “smart sacrifice” - the firm explicitly
chooses to perform poorly along one dimension of customer needs (the sacrifice) so that is can
dramatically and substantially outperform the competition along another dimension of customer
need (the smart choice). For example, before Amazon, customers had the need for a wide selection
of books that could be physically inspected before purchase and obtained quickly. Amazon sacrificed
physical inspection and immediacy, but dramatically improved upon the selection of available books.
New business models emerge in part because entrepreneurs try many new ideas, some of which
turn out to be both novel and effective. But the fact that so many new business models have
emerged over the last few decades is not entirely an accident. There are other trends that con-
tribute to the creation of new business models. For instance, reductions in transportation costs
enable a firm to consider a broader set of location options in the supply model, enabling both
the movement of goods from far away and the quick movement of goods nearby. The growth of
cities concentrates demand, which builds economies of scale in operations that yield lower costs.
Increased standardization of physical goods (e.g., through precise machinery and the ability to
communicate detailed specifications electronically) enables expanded options in terms of control
- assets that use to have be owned can now be rented. The internet allows retailers to expand
their product offering and to implement complex pricing dynamics. Mobile devices dramatically
increase the amount of data that can be exchanged and the speed of that exchange. All of these
enhancements to telecommunication greatly improved transactional efficiency and opened up new
sources of customer value. Etcetera. In sum, changes in living patterns and technology lead to
cheaper and faster movement of people, resources and information, which enable new combinations
of supply and revenue models to serve new customer needs.

3 Business Model Research

Business model innovations challenge implicit assumptions, thereby revealing new ways of doing
business that were assumed to not be feasible. In effect, they provide a window to a new world that
was not known to exist. Not surprisingly, successful business models garner substantial attention in

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
industry and academia. This section discusses the academic literature on a wide range of business
model innovations from the past few decades, including sponsored search advertising auctions
(Google), “name your own price” and opaque selling (Priceline.com, Hotwire.com), “selling for
free” (Facebook), bundles and subscriptions (Spotify, Netflix), every day low pricing (WalMart),
limited markdowns (Zara), surge pricing (Uber), reselling (StubHub), make-to-order production
(Dell), e-tailing (Amazon), and room sharing (AirBnB), to name a few.

3.1 Revenue model

There is remarkable variation in how firms earn revenue. As discussed in Section 2, the primary
goal of the revenue model is to extract rents from customers to support the resources needed to
serve them. The high-level decisions associated with the revenue model can be divided into three
components: mechanism (what prices are chosen, and by whom), structure (sell or rent, charge
per-use or fixed fees), and dynamics (how often are the offered transaction terms changed, and
what options are available to modify initial agreements).

3.1.1 Mechanism

The simple posted price is probably the most common mechanism for establishing the price a con-
sumer pays. It is easy to communicate and enables a quick and immediate transaction. However, it
also requires that the firm determines the best price based on what the firm knows about its own sup-
ply and consumer preferences. When such knowledge is limited, the firm might make a significant
error. An alternative to the posted price is some type of auction in which the price is endogenously
determined.2 With a well designed auction the final price can reflect the true preferences, thereby
allowing the firm to extract more value. That said, an auction imposes significant transactional
costs on consumers, such as the delay needed to gather auction participants, the uncertainty of
knowing whether a transaction can be completed, and the time needed to participate and monitor
the auction. Thus, the auction mechanism tends to be preferred over posted prices when consumer
preferences are sufficiently disperse (Wang (1993)). For example, auctions have recently been used
to find the lowest cost supplier throughout the world (Mukhopadhyay and Kekre (2002)) and to
match consumers selling their own goods to other consumers (e.g., eBay, Simonsohn and Ariely
(2008)). Among the greatest successes for auctions is with sponsored search advertising markets,
such as those Google operates (Hosanagar and Abhishek (2013)). Thousands of search queries can
2
Haggling is yet another option, with similar strengths and limitations as auctions (Desai and Purohit (2004)).

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
be submitted per second and advertisers vary considerably in how much they are willing to pay at
any given moment to associate their ad with the query. Auctions can be implemented quickly with
modern technology and are superior at generating revenue in these environments relative to posted
prices.
Given that posted prices and auctions each have their own limitations, one solution is to of-
fer both of them to consumers: Etzion et al. (2006) demonstrate that offering both mechanisms
simultaneously can increase revenue because it enables additional consumer segmentation. Alter-
natively, Priceline.com was the first to develop a hybrid mechanism, called “name-your-own-price”
(NYOP), in which consumers submit bids to a firm that are either immediately accepted or re-
jected (Amaldoss and Jain (2008), Tuo Wang (2009), Fay (2004), Spann and Tellis (2006), Hann
and Terwiesch (2003), Terwiesch et al. (2005)). The firm usually implements a relatively simple
strategy - assign a fixed threshold, accept all bids above the threshold, reject all bids lower than
the threshold. Although this initially seems just like the traditional take-it-or-leave-it single price,
there are important differences. First, because consumers don’t know the firm’s threshold, the firm
earns additional revenue whenever the consumer bids above the threshold. Second, other firms
cannot easily observe the firm’s threshold, which changes competitive dynamics.
A surprising extension of NYOP is “pay-what-you-want”, which is NYOP in which the firm
publicly announces a zero threshold for accepting bids, i.e., the firm accepts any bid (Schmidt et al.
(2015); Gneezy et al. (2012)). While consumers are allowed to bid zero, they might actually bid
more because they want to be (or viewed to be) pro-social individuals. As a result, pay-what-
you-want can (surprisingly) generate revenue, and in some cases even more revenue than other
mechanisms.
In NYOP the firm keeps the price it will accept hidden from consumers, but consumers are told
exactly what product they are bidding on. Opaque selling reverses what is known to consumers:
with opaque selling, which is also called “probabilistic selling”, the consumer is given a posted price
for a virtual product that is a lottery over several real products. For example, a consumer could
be offered a 3-star hotel in center city on a particular evening for $125, and only if the consumer
accepts this offer is the actual hotel revealed. Hence, with opaque selling a consumer runs the risk
of being assigned the least preferred product in the set of possible choices. Fay and Xie (2008)
explain how selling an opaque product can work to a firm’s advantage. Say the firm offers two
products and consumers vary in their preferences across the two products, some really like one
of them but not the other, and some are relatively indifferent between the two. The firm has a

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
pricing dilemma: price each product high to cater to the consumers who like the products while
not selling to the indifferent consumers, or price each product low to try to capture sales from
the indifferents. Neither is ideal. However, suppose the firm can create a virtual third product
with which a consumer receives a random choice between A and B (i.e., a probabilistic product).
The indifferents would rather choose the product than receive a random choice, but the other
consumers really do not like the opaque product. Thus, the firm can set a high price to sell to the
picky consumers and also sell to the indifferents via a lower (but not too low) price assigned to
the opaque product. As a bonus, because an opaque product is not a real product, the firm does
not have to invest in the development of another real product to cater to the indifferents. (See
Appendix 5.1 for a simple numerical illustration.)
In Fay and Xie (2008) there is a single firm that offers the third/probabilistic product. But it
can also be helpful for an intermediary firm (e.g., Hotwire.com) to offer an opaque product that
is a random choice between products from competing providers (Jerath et al. (2010)). To explain,
say an airline’s customers can be divided into two types: loyal customers only fly with that airline
and bargain hunters who shop between different carriers. The airline might want to offer an opaque
product choosing among its flights, but if its loyal customers know that they will fly with the airline
even if they choose the opaque product, then it hard to convince them to pay the higher, regular
fare. If instead the opaque product is offered by a third party, who mixes in flights from other
airlines, then the loyal customer are far less likely to consider that channel. This allows the airline
to separate the segments: a high price to the regular customers and a lower price to cater to the
bargain hunters via the opaque intermediary.

3.1.2 Structure

The structure of the revenue model is concerned with the terms of the transaction between the firm
and the consumer. Two key dimensions of transaction terms are (a) ownership and (b) payment
form.
Many transactions between firms and customers involve the issue of ownership. Ownership
refers to the degree of control over a product or resource. For example, the owner of a product
can use it when wanted and to the extent wanted. The owner is responsible for maintaining or
upgrading the product, and the owner can choose to dispose of the product or to sell it to another
owner. Many products are naturally sold to customers, i.e., ownership is transferred. For example,
a bakery only sells loaves of bread. It makes no sense for the bakery to lease or rent bread to

10

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
customers. This is generally true with any disposable physical product. However, ownership need
not be transferred with durable physical products. Say the firm makes mainframe computers. The
firm could sell it to a customer, say a bank. The bank would then be free to use the computer as
much or as little as it wants. Over time the computer depreciates and it is up to the bank to decide
when it might sell the mainframe on the secondary market or if it will use it to the point at which
it has to be scrapped. Alternatively, the mainframe manufacturer could lease the computer to the
bank. The upfront payment is far smaller, but then the bank makes periodic payments for the use
of the mainframe. At the end of the lease, the mainframe is returned to the manufacturer and
it is the manufacturer’s responsibility for salvaging any remaining value. This “sell versus lease”
decision has been investigated extensively. Leasing can be better than selling for the manufacturer
because it commits the firm to not flood the market in future periods, which would depress the
residual value of the units - the bank will not pay a premium for a mainframe today if it knows that
the price of the mainframe will fall steeply in the future because of additional production (Bulow
(1982), Stokey (1981)). However, there are situations in which firm may wish to both sell and lease
(Desai and Purohit (1998)).
In addition to ownership, the firm also must decide how to charge customers for the amount
consumed. Although it is natural to charge a fee for each unit a customer uses, there is an alternative
to per-use (or variable fee) pricing. A firm could offer a fixed price for a bundle of products from
which the consumer can choose as many or as few as desired. For example, Spotify with a fixed
monthly fee allows users to listen to any of the songs in their library. Given the breadth of genres
in their collection, most consumers have little interest in the vast majority of their titles. Yet,
despite the fact that consumers might be knowingly paying for many things that they never want,
this extreme bundling strategy can be better for the firm than charging for each song (Bakos and
Brynjolfsson (1999)). To provide a simple illustration, say there are 10 products and consumers have
independent uniform [0,1] valuation for each of the products. If each product is sold individually,
the optimal price would be 0.5, a consumer would purchase on average 1/2 of the products and
expected revenue per customer is therefore 0.5 x 1/2 x 10 = 2.5. Alternatively, the firm could offer
the bundle of all 10 products at a price that matches its expected value, which is 5. In that case,
half of the consumers would be willing to purchase the bundle and per-customer revenue is again
2.5 - no improvement. But lowering the bundle’s price to 3.9 has a dramatic effect - about 88.5%
of customers are willing to pay 3.9 for the bundle of 10 products, thereby increasing expected per-
customer revenue to 0.885 x 3.9 = 3.45, which is 138% higher than selling the products individually.

11

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
The firm can extract more revenue from customers via the bundle because consumers can be less
price sensitive regarding the bundle than they are for individual products.
Charging per product and charging for the entire bundle are two extreme strategies. The firm
is of course not required to pick just one of those approaches - the firm can offer a menu of choices.
In the mentioned example, the firm can do even better by giving consumers the option either to
purchase a bundle for 3.9 or to purchase items for 0.88 each. Interestingly, the incremental revenue
increases by a very small amount, from 3.44 to 3.47, which is only about 0.7%. This is not the only
example in which a single, simple, pricing option is not optimal, but nearly optimal relative to the
more complex pricing menu (e.g., Cachon and Zhang (2006)).
There are some limitation to bundling. For one, bundling can run into trouble if there is a non-
trivial marginal cost to serve each customer. Continuing with the 10 product example, bundling and
per use yield the same profit when there is a cost of 0.216 per product served to a customer.3 Higher
marginal costs favor per-use pricing. Furthermore, the negative effect of marginal costs on bundling
becomes more substantial as the number of products increases (i.e., the breakeven marginal cost
decreases). Nevertheless, while bundling requires low marginal costs to be the superior strategy,
it doesn’t require zero marginal costs. Next, bundling cannot perfectly segment across customers
that differ in their breadth of preferences. Instead of the 10 product example, say the number
of products, n, is very large (e.g., 100 < n). Half of the market has broad preferences in the
sense that they have some value for all products. But the other half of the market is composed
of consumers who only value a single product. In either case, when valuations are positive, they
are uniformly distributed [0, 1]. If the firm could perfectly segment these two types of consumers,
then the firm could charge the “broad” customers pn = 1/2n for the entire bundle of n products
and earn on average 1/2 per product. To the “narrow” customers the firm charges p1 = 1/2 per
product and earns on average 1/4 per product. Given that broad customers are half of the market,
the average profit per product is 3/8. But this assumes the consumers can be perfectly segmented.
Unfortunately, if the price is p1 = 1/2 per product and pn = 1/2n for the bundle of n, then the
broad customers do not buy the bundle. Instead, they choose to make their own bundle of half
of the products, paying the per-product fee p1 , and the firm is back to earning 1/4 per product.
Fortunately, there is a better solution for the firm - charge p1 = 2/3 per product and pn = (4/9)n
for the bundle. Raising the per product price enable the firm to set a bundle price that earns some
revenue and induces the broad customers to take it. Total revenue is now 1/3 per product, which
3
The breakeven marginal cost solves 3.45 − 10 × 0.885 × c = 10 (1 − c)2 /4

12

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
is better than 1/4 per product, but less than the ideal of 3/8.
The airline industry illustrates that the choice between bundled and per-use pricing is not always
straightforward. Traditionally, airlines bundled the price of the seat in the aircraft with the service
of carrying a passenger’s luggage. However, around 2007 some airlines began to offer separate fees
for each product, i.e., they started to charge baggage fees. Southwest Airlines resisted that trend
(and still offers bundled pricing) but the majority of the other major airlines have transitioned to
separate fees for each service (Nicolae et al. (2017)).
Bundling can be applied across products and it can be applied across time, in which case it is
usually referred to as subscription pricing. For example, Blockbuster dominated the movie rental
business for a period of time with per-use pricing, but Netflix successfully introduced subscription
pricing - customers could rent as many movie DVDs as they wished during the month for a single
fixed subscription price. The advantage of bundling across time is analogous to the advantage
of bundling across products - the user becomes less price sensitive when considering a series of
consumption opportunities (some of which are taken, some not) rather than considering them
individually. For example, it can be better to sell a membership to use a gym over a period of
time rather than to sell individual visits to the gym. The drawback of selling subscriptions is that
consumers might over consume the service - once a gym membership is paid and there is no fee for
each use, a consumer is likely to use the gym even on occasions when the marginal value created
is less than the cost of providing the service. This is particularly problematic when the offered
product suffers from congestion - each additional user in the gym creates some inconvenience to the
other people trying to use the gym. Even though in these situations it might seem that charging
per-use fees would be better (to reduce excessive use), Cachon and Feldman (2011) demonstrate
that subscription pricing still may be preferred. To explain, congestion can either be controlled via
a per-use fee or through buying more capacity/resources. However, the latter requires additional
revenue, and subscription pricing is capable of extracting more revenue from customers than per-
use pricing. Hence, even in services subject to congestion (e.g., mobile telecommunication), it can
be better to sell subscriptions than to charge per-use because it is better to buy more capacity than
to regulate how consumers use capacity (See Appendix 5.2 for details.)
A surprising pricing strategy involves neither the per-use or fixed fees of bundles or subscriptions.
Instead, the firm charges neither, i.e., it offers it’s product for free. Clearly firms cannot survive
by selling all of their products for free. But when can a firm profit from selling some of its
products for free? Alphabet (i.e., Google) and Facebook have been among the most valuable public

13

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
companies (in terms of market capitalization), and yet the vast majority of their users pay nothing
for their products. This remarkable achievement can be understood via the literature on two-sided
markets (e.g., Rochet and Tirole (2006), Parker and Alstyne (2005)). For example, credit card
companies charge merchants to process each transaction but often earn no revenue at all from some
customers. Even though value is created for both sides of the market, a positive price is charged to
only one side. In some domains this occurs because it is technically infeasible to charge one side -
traditional radio and television stations broadcasting their content could not monitor (and therefore
charge) consumers of their music or shows. The introduction of cable and internet technology for
distribution enabled consumers to be charged, and some companies adopted the new revenue model
(e.g., HBO). However, even when it is feasible to charge both sides of the market, sometimes it
is not done (as with Google and Facebook). The justification for the zero price in one market
comes from the existence of externalities across the markets. A free price for using a credit card
allows the firm to acquire a much larger base of credit card users than it would even if it charged a
relatively modest price. The larger user base is far more valuable to merchants, thereby allowing the
credit card company to charge merchants more than is lost from the absence of customer generated
revenue. Without the merchants willingness to pay marginally more as the user base increases,
there is no reason to sell for free to consumers merely to increase their numbers. That said, why
stop at free? Why not pay consumers to join the user base? Once a platform jumps from “free” to
“we pay you to sign up”, it is likely that the incremental demand would be exclusively interested
in the free money and of little value to the other side of the market.
The software industry has special versions of selling for free, one called “open source” and the
other called “freemium”. With open source, not only does a firm allow some customers to use their
product for free, they allow the users to modify and enhance the product (Casadesus-Masanell
and Llanes (2011)). This strategy is successful when the code that is given away flourishes with
independently developed enhancements, which creates a user base that is willing to pay for expertise
on how to use the code effectively. In effect, like in the classic two-sided markets, one group of
agents generates positive externalities that can be captured via another group of customers. With
“freemium”, the firm allows consumers to use a limited-feature version of the software without time
limit and also offers a premium version of the product for a positive fee (Cheng and Liu (2012)). For
example, DropBox allows consumers free use of their service with a limited amount of storage for
as long as a consumer wants. This is different than a two-sided market: there are few externalities
between the two groups in DropBox (free users and paying users). Instead, this model is justified

14

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
by two assumptions: (i) users will not try the product if they have to pay up front nor will they
be inclined to try a product that has a limited time offer (why learn to use the product if you can
use it for only a limited time) and (ii) the desirability of the premium product may increase over
time as familiarity with the product increases (Kumar (2014)).

3.1.3 Dynamics

Part of a firm’s revenue model involves its long-run policy towards the frequency and magnitude
of adjustments to its transaction terms, which generally means its price. For example, does a firm
develop a reputation for periodically offering temporary sales, or setting high prices that are later
marked down, or offering advance-purchase discounts, or prices that swing up and down throughout
the day? All of these cases have been implemented in practice and researched extensively.
Some retailers are known for periodic deep discounts, which has been called “hi-low” pricing.
Theory suggests that this can increase revenue by segmenting consumers either in their willingness
to wait for a discount or in their willingness to carry inventory (Blattberg et al. (1981), Ho et al.
(1998)).4 To give a simple example, say a retailer sells cans of soup and has two types of customers.
The “high types” are willing to pay $4 per can, want one can per week, and do not want to carry
inventory in their home. The “low types” pay at most $2 per can, want one can per week, and are
willing to buy as many as 5 cans at a time. Say there are 100 of the high types and 200 of the
low types. If the retailer always charges $4 per can, then the retailer earns $400 in revenue per
week selling only to the high types. If the retailer always charges $2 per can, then the retailer’s
revenue is $600 per week selling to all consumers. But if the retailers has a regular price of $4 per
can and a discount price of $2 every 5 weeks, then the retailer earns $400 per week from the high
types when the regular price is offered and $2,200 in the promotions weeks (because the low types
buy five cans to carry them to the next promotion), for an average of $760 per week. The hi-low
strategy earns considerably more (27%) than the best non-dynamic pricing policy.
Segmentation via hi-low pricing can work, but it runs the risk of not being popular with con-
sumers, and it might add to operational costs due to higher demand variability (Lee et al. (1997)).
Some retailers have attempted to adopt a “value pricing” or “everyday-low-pricing” (EDLP) ap-
proach that eliminates promotions or at least reduces their frequency and depth (e.g., Walmart).
However, the empirical evidence tends to lean in favor of hi-low pricing, at least when the focus is
4
Learning about demand can be another motivation for dynamic pricing: Besbes and Zeevi (2015), Yu et al.
(2015)).

15

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
on short term revenue (Hoch et al. (1994)).
With seasonal goods, especially with fashion items that have short-lived preferences due to
shifting tastes, a standard pricing model begins with a high price and then discounts are offered
towards the end of the season if inventory remains that needs to be salvaged for something. Hi-low
pricing in this context is sometimes referred to as a “price skimming” strategy. The declining price
path is used to try to sell to the consumers with a strong preference (if they buy at the high price),
while also ensuring that some sales are made through the lower prices at the end of the season
if it is necessary to sell to consumers with weaker preferences (e.g., Lazear (1986), Besanko and
Winston (1990)). This strategy works very well if consumers myopically purchase whenever they
see a price they are willing to pay. Unfortunately, consumers might be strategic, avoiding the high
initial price because of the expectation that the item can be purchased later on at a lower price.5
In fact, that behavior can be self-reinforcing - if all consumers refuse to purchase at the higher price
because they expect a lower price to follow, then the sparse initial sales indeed forces the firm to
discount later on. Retailers tend to have an opinion on this behavior: A CEO of Best Buy once
described the former (myopic) consumers as “angels” (because they purchase at full price) and the
latter (strategics) as “devils” (because they tend to buy at discounted prices (McWilliams (Nov 8
2004)).
The consequences of strategic consumer behavior and the appropriate firm response has been
debated extensively by both practitioners and academics. Some firms, such as Zara, have found
success by severely curtailing discounting (Hansen (2012)), whereas other firms (e.g., JC Penny)
have attempted to avoid discount pricing but found disappointment (Clifford and Rampell (2013)).
A (somewhat) simple example highlights the potential benefits and pitfalls of these strategies (Ap-
pendix 5.3). One firm buys q units before a selling season and pays $50 for each unit. The product
is sold over two periods. A random number of consumers arrive at the start of the first period.
That demand can be described with a gamma distribution with mean 100 and standard deviation
100. All of those consumers are willing to pay up to $100 for the product. Half of them are myopic
- they buy in the first period in which they see a price less than $100. The other half are strategic
- they purchase in the period that gives them the better deal.6 The strategics know that they are
more likely to be able to find the item available for purchase in period 1, but they also know that
5
Chevalier and Goolsbee (2009) and Li et al. (2014) provide empirical evidence of strategic consumer behavior.
6
This model does not consider the possibility that the firm can take actions to manipulate/distort consumer
perceptions of availability, and therefore the likelihood of future discounts (Yin et al. (2009), Özalp Özer (2016)).
Nor do these consumers care whether others are able to purchase (Tereyagoglu and Veeraraghavan (2012)).

16

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
the price is lower in period 2. In particular, if the firm has units remaining at the start of period 2,
then, because ther is an unlimited number of bargain shoppers in period 2 willing to pay $30, the
firm can lower the price to $30 to ensure that all of the inventory is sold.
The firm has four strategic options. With “hi-low” pricing the firm chooses the highest possible
initial price, p1 = $100, and then discounts if inventory is remaining in period 2, p2 = $30. The
strategics get no value from purchasing in period 1, so this approach focuses on selling to the myopics
and the strategics are relegated to discount shoppers. With “value” pricing the firm chooses some
intermediate price to try to get the strategics to purchase in period 1, i.e., $50 < p1 < $100, and
then discounts in period 2 to p2 = $30 if inventory remains. The strategics can be willing to pay a
bit more in period 1 because waiting involves the risk that the item might be out of stock in period
2. A third approach is “never discount”: p1 = p2 = $100. The strategics buy in period 1 but
left over inventory is totally scrapped.7 Su and Zhang (2008) refers to this as price commitment.
The fourth approach is a “refund” strategy (Lai et al. (2010)): set a high initial price, p1 = $100,
discount to p2 = $30 in period 2 if there is any inventory left over, and promise period 1 customers
that they will receive a refund equal to the full price difference, p1 − p2 = $70, if a discount is taken.
The refund eliminates the risk from buying in period 1. Hence, all of the strategics are willing to
buy in period 1 even though the initial price is very high.8
Table 1 displays the results of the four strategies. Value pricing is clearly inferior - it does
attract the strategics to purchase at the regular price, but because the regular price is less than
what the myopics will pay, the firm’s profit suffers overall. Hi-low and refunds perform similarly
even though they are clearly very different strategies - with hi-low the strategics don’t buy at the
regular price, but with refunds the firm needs to be conservative with its order quantity to avoid
the risk of having to give a discount to everyone. The best approach is the one that doesn’t alter
the price across the season, the never discount strategy - the firm incurs the significant cost of being
unable to salvage any leftover inventory, but it has the benefit of forcing the strategics to buy at
the regular price.
It is dangerous to conclude too much from a single example, but in a broader set of scenarios
(reported in Appendix 5.3) it is found that never discount is most frequently the optimal strategy
7
Never discount and value pricing are actually extreme versions of a single strategy in which the firm randomly
marks down to $30 in period 2 with probability θ if there is inventory left over: never discount is a random markdown
with θ = 0 and value pricing is a random markdown with θ = 1. In this case, the firm’s optimal strategy is the
extreme to never discount. See Moon et al. (2016) for a broader analysis of random markdowns.
8
There is one additional strategy that initially seems different but in fact is equivalent to refunds - the firm
announces before the season starts the number of units purchased, q. Assuming this is feasible, the goal is to signal
to consumers that product availability is limited, so they should purchase in period 1 even though the price is high.

17

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
 Table 1: Pricing strategies for a fashion product with uncertain demand
Strategy q p1 p2 Profit
Hi-Low 62.6 100 30 1247
Value 81.7 75.3 30 893
Never Discount 69.3 100 100 1534
Refund 52.0 100 30 1124

and it is robust (i.e., even when it isn’t optimal, it is nearly optimal). Similarly, value pricing is
never optimal, suggesting that it is ill advised to pander to strategics. (That said, the story on
value-pricing isn’t complete until the supply model is considered.)
The strength of the no-discount strategy raises the question of whether a firm should even use
dynamic pricing - there is nothing dynamic about the no-discount commitment that maintains a
single price. However, this is not the only reason managers may hesitate to use dynamic pricing.
Another reason arises from the general perception that dynamic pricing reduces consumer surplus.
Given that dynamic pricing involves “high” and “low” prices, it seems intuitive that “high” prices
are used to extract rents from consumers, thereby lowering their surplus. This is particularly
salient if it is assumed that the total value in the system is a fixed amount - if the firm gains by
a higher price, surely consumers must lose. In the short term this shouldn’t be a direct concern
for managers. But in the long term it could lead to a permanent loss of customers. And that is a
concern. Interestingly, Chen and Gallego (2016) show that when demand is stationary but costs are
stochastic, consumers are indeed generally better off when the firm implements dynamic pricing.
The questions of whether dynamic pricing should be used and who benefits or is harmed are
central to the recent rise of ride-sharing platforms like Uber. Uber matches independent drivers
to riders seeking local transportation. For Uber, the central challenge is not strategic consumer
behavior - many consumers are unable to strategically time their need for transportation. Rather,
the first-order challenge for Uber is to match supply with demand as it learns information about
demand and supply conditions. Uber has a finite pool of drivers at any moment, but its demand
can vary wildly from day to day and hour to hour, based on local events and weather, among other
sources of variability. Too many drivers on the road is a problem because then they are idle more
than they want (which destroys value). Too few drivers on the road is another problem because
then customers wait too long and eventually become disenchanted with the service (which destroys
value). Uber’s innovation is to implement dynamic pricing, which they refer to as surge pricing,
with its demand (customers) and with its supply (drivers). Relative to a pricing regime with fixed
prices and wages (e.g., a traditional taxi model), the optimal surge pricing strategy is much better

18

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
for Uber. Surprisingly, dynamic pricing can also be much better for the drivers (even though they
get paid less during slow periods) and consumers (even though they have to pay much more during
peak periods) because dynamic pricing enables an overall increase in supply, which increases the
potential value generated in the system (Cachon et al. (2015)).9
In many dynamic pricing settings (groceries, fashion apparel, ride-sharing) transactions and
delivery of the product occur approximately at the same time. However, some firms provide cus-
tomers with the opportunity to commit to purchase their product well in advance of actually using
the product. For example, this is customary, and expected, in the travel industry (reserving hotels,
car rentals, airline flights, etc.) This selling strategy works for several reasons. To begin, some
customers value knowing that they can plan to have access to the product at the designated time,
and a portion of this value can be captured by the firm (Gale and Holmes (1993), Alexandrov and
Lariviere (2012)). Second, advance selling allows for segmentation across customers that differ in
their value for the product based on when they arrive to the market. The classic story is that
price sensitive customers arrive early to the market (e.g., leisure travelers) and price insensitive
customers arrive late (e.g., business travelers). Advance selling allows the firm to sell to these
segments at different prices (Su (2007)). A third reason for advance selling is less intuitive - cus-
tomers may be willing to purchase before they are certain of their value for the product (Xie and
Shugan (2001),Chu and Zhang (2011)). To illustrate with a simple model, suppose a customer can
purchase the product either “in advance”, i.e. well before consumption, or “on the spot”, i.e., just
before consumption. In advance the customer is uncertain regarding her value for the product, but
knows that her value is uniformly distributed between 0 and 1. On the spot, the consumer learns
her value. On the spot the firm sets p = 1/2, which maximizes revenue if spot consumers have
valuations uniformly distributed [0, 1]. The consumer anticipates that she earns a surplus of 1/8 if
she purchases on the spot and the firm earns 1/4.10 But if the firm sets an advance price of 3/8,
then the consumer prefers to purchases in advance (before knowing her value) and the firm earns
3/8, which is 50% higher profit than selling on the spot.
It is remarkable that advance selling can be favorable to the firm even in the absence of avail-
ability issues (i.e., no capacity constraint) or segmentation opportunities (i.e., consumers are ho-
mogeneous). Advance selling works, despite the cost it imposes on consumers, because consumers
9
Research on ride sharing is rapidly expanding: e.g., Bimpikis et al. (2016), Besbes et al. (2018), and Hu and
Zhou (2017)
10
There is a 0.5 probability of having a value greater than p = 1/2, and 3/4 is her expected value conditional on
having a value greater than p. So expected value is (3/4 − p)1/2 = 1/8.

19

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
are more homogenous in the advance period (they all have expected value of 1/2) than in the spot
period (valuations are distributed [0, 1]). It can be better to sell to a population of homogeneously
valued consumers even if their valuation for the product is less than 1/2 of consumers in a heteroge-
nous population. This is essentially the same reason for the effectiveness of bundles across products
(e.g., a single fixed price for access to a library of music) or across time (e.g., subscriptions).
Although advance selling can work in the firm’s favor, because of the gap in time between the
transaction agreement and the transaction delivery/completion, it is possible that one or more
of the parties in the transaction learns new information that motivates an adjustment to the
transaction terms. When this possibility can be anticipated, the firm can explicitly incorporate a
particular recourse into the transaction agreement. For example, the firm can offer partial refunds
to consumers who later learn that their valuation for the product is low. If the return/refund
is completed with enough time before consumption, the firm then has the opportunity to sell
the returned product to another consumer (Xie and Gerstner (2007), Guo (2009) Gallego and
Sahin (2010)). The firm could also try to sell to another customer even if consumers are not
allowed to return the product. In effect, the firm tries to sell its capacity twice, a practice known
as overbooking. If the firm does successfully sell the unit to another customer, then the initial
customer is denied service. In the context of airlines, this is called “bumping” a passenger off of a
flight. Although the initial customer might be disappointed, the possibility of being denied service
and the resulting compensation are known in advance. Finally, rather than being involved in a
possible ex-post transfer of capacity from one customer to another, the firm could allow customers
to resell the product (Cui et al. (2014), Su (2010)). Cachon and Feldman (2018) find that a seller
can improve its revenue considerably with any of the three recourse strategies (refunds, overbooking
or reselling), but reselling is generally the best. Reselling is most effective because it helps to ensure
that the consumer who values the product the most is the one who uses the product. Interestingly,
this additional value is equally shared between the firm and consumers, so both benefit from
the practice, which is consistent with the considerable growth in ticket exchanges like Stubhub.
However, this is predicated on the assumption that the reseller actually earns some value from
reselling. If the reseller doesn’t earn anything, then neither does the firm. For example, “reselling”
in digital music is often called “piracy”. It is intuitive that piracy can harm the firm (Rob and
Waldfogel (2006)), but there is also the possibility that sharing facilitates consumer search, which
can be helpful (Zhang (2016)).

20

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
3.2 Supply model

The main function of the revenue model is to extract value and the main function of the supply
model is to generate value. There are three high-level strategic decisions that the firm must make
with respect to how it designs its supply: timing (when are resources activated relative to the
arrival of demand), location (where are resources positioned relative to demand) and control (to
what extent is the firm able to decide when and how resources are utilized).

3.2.1 Timing

A firm needs to decide if it begins work in anticipation of demand, or if it waits for demand to
arrive before starting service. The firm’s positioning on this timing spectrum determines whether
its resources wait for customers or whether its customers wait for resources. In the early years of
the personal computer industry the key insight was that it was better to make customers wait.
That is not always the case.
Some form of waiting invariability occurs because resources are relatively rigid whereas demand
is volatile. Two examples illustrate this point, one in which demand is volatile in predictable ways
(seasonality) and the other in which randomness is the cause.
Most firms face some form of seasonal demand. Say there are two seasons, a low season and a
high season; the seasons alternate in occurrence; each season lasts for τ time; and demand in season
s ∈ {l, h} occurs at constant rate λs , λl < λh . Let λ = (λl + λh ) /2 be the average demand rate,
and let δ = (λh − λl ) /2 be a measure of the amplitude of seasonal variation. The firm chooses a
constant production rate, µ, and incurs a constant cost rate of cµ to maintain that capacity, where
(w.l.o.g.) let c = 1.
The firm’s capacity should be selected in the range [λ, λ + δ] so that it has enough capacity to
satisfy all demand but not so much that a portion of it is never used. Hence, the firm has sufficient
capacity during the low season to satisfy demand, but does not have enough capacity for the high
season. There are essentially two extreme options for dealing with the high season. First, the firm
can do some work in advance of the high season. For example, a toy manufacturer might build and
inventory completed toys before the fourth quarter or a restaurant might prepare its ingredients
before the evening rush so that dinners can be cooked more quickly once orders are received. Either
approach creates inventory carrying costs: let h be the cost incurred per unit of work per unit of
time the work is completed before its demand arrives. Second, the firm can make some high-season

21

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
demand wait, i.e., create a queue of demand that is satisfied when capacity becomes available
(i.e., during the low season). This too is costly: let w be the cost per unit of work that is in the
queue per unit of time. And as in most cases when presented with extreme options, there exist
an intermediate approach that combines the use of doing work before the season starts (inventory)
with doing work after the season begins (queues).
Volatility is never helpful to the firm: costs increase in both the amplitude of the seasonal
variation, δ, as well as the duration of the seasons, τ. (See Appendix 3.2.1 for details.) To cope
with that variability the firm can either just use inventory, or just make customers wait, or some
blend of the two. Interestingly, one of the two extreme strategies tends to be optimal. To be specific,
the intermediate strategy that blends both inventory and queues is optimal only if 1/3 < w/h < 3,
otherwise, one of the two extreme strategies is preferred. For example, the firm’s best solution is
to only use inventory whenever 3 ≤ w/h: if the cost of waiting is more than 3 times the cost of
holding inventory, the optimal solution is to only build inventory. Analogously, if w/h < 1/3, then
it is best to only use a queue. Neither threshold is particularly extreme, which suggests that in
practice a firm is likely to take one approach or the other, but not a mixture of both. For example,
toy manufacturers build inventory before their high season, whereas aircraft manufacturers make
their customers wait.
Now consider the timing question in an environment with stochastic demand. Say a firm sells
N physical products that can be inventoried. The lead time to produce each product is l (with no
capacity constraint). Total demand occurs at rate λ and the ith product demand rate is λi = λ/N.
Inter-arrival times for demand are exponentially distributed. Again, let h be the cost to hold
inventory per unit of time and w the cost to make a unit of demand wait one unit of time.
Each product is managed with a base stock policy. When demand is sufficiently slow, the
boundary between inventory (make-to-stock) and queue (make-to-order) occurs between a base
stock level of S = 1 (make-to-stock) and S = 0 (make-to-order). Making customers wait (S = 0)
is better when
lλ/N < ln (1 + h/w) (1)

So, make-to-order is more likely to be the better strategy when the lead time is low (small l means
customers don’t have to wait too long), there are many products (large N ), holding costs are high
(large h) and customers are relatively patient (low w).
To get a sense of how likely (1) is to be satisfied, let’s evaluate plausible values. First consider h,

22

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
 Table 2: Maximum demand threshold per product to support make-to-order

w h = 20% h = 60% h = 100%

1% 0.069 0.194 0.305
5% 0.014 0.042 0.069
10% 0.007 0.021 0.035
20% 0.004 0.011 0.018
h = annual holding cost rate; w = daily wait cost; l = 1; 280 days per year.

which includes the opportunity cost of capital, some accounting for obsolescence or spoilage costs,
and storage costs. To estimate the latter, consider 2016 data from Kroger (a large grocer in the
United States). Assume retail space costs about $200 per m2 per year, which is approximately the
cost of commercial real-estate. Their annual cost of goods sold was about $90 billion, their average
inventory was about $8 billion and they operated with about 18 million square meters of retail
space. This implies they have on average $8B/18M m2 =$444 of inventory per m2 of retail space.
Storage costs alone are thus $200 per m2 per year /$444 per m2 = $0.45 per dollar of inventory
per year, or 45% of the value of the inventory on an annual basis. Add in the opportunity cost
of capital and other inventory storages and maintenance costs, and Kroger could easily have an
annual holding cost of about 60% (i.e., to hold $1 in inventory for 1 year costs them $0.60). A
similar evaluation for other retailers demonstrates that an annual holding cost of 60% is reasonable
and often is conservative.11 Now consider the waiting cost. Plausible values for w (cost per day)
could range from 1% to 20% of the product cost: w = 5% implies the daily cost to make a customer
wait equals 5% of the cost of the item, which means that after 1/0.05 = 20 days the consumer’s
waiting cost equals the product’s cost.
Using the plausible ranges for h and w, Table 2 evaluates the maximum demand rate that
justifies make-to-order, i.e., the λ that makes (1) bind with equality. The resulting demand rates
appear small, ranging from 0.004 units per day to 0.305 units per day. But note that those rates
are for the individual products. If a firm sells 100 units per day, and offers 10,000 different variants
of the product, then the demand rate per product is 100 / 10,000 = 0.01, which is comparable to
the thresholds in the table, and lower than all of the thresholds with the two higher holding costs.
One can conclude that make-to-order (i.e., queues of customers waiting for resources) combined
with broad variety can be the best strategy, especially when waiting costs are not extreme. That
insight led Dell Computer to become a dominant player in the personal computer industry.
11
One might argue that space costs shouldn’t be included in h because space is not adjustable in the short term.
True, but when comparing two operating systems, long run costs should be compared and space is surely adjustable
in the long term.

23

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
 Although the issue of timing usually involves the boundary between the firm and the customer
(as in the previous examples), the firm makes strategic decisions regarding timing in other domains.
For example, it has been established that a firm can reduce its investment in inventory, while not
changing the assortment of products available to consumers, by delaying the process that adds
variety to the product as late as possible in the supply chain. This strategy is called “delayed
differentiation” or “postponement” or “localize remotely” (Lee and Tang (1997)).
Timing is also important in the decision of which products to offer - the closer in time product
design decisions are made relative to when products are sold, the more likely for better product
designs to be selected. This is a crucial component to Zara’s business model success - they are able
to take a design concept and deliver a new product in a matter of weeks, rather than months (or
many months), which means they are more likely to offer desirable fashions (Cachon and Swinney
(2016)). But even Zara still produces inventory before demand. The next level is to decide which
products to produce only after demand is realized. The extreme version of this is called mass
customization - customers request products customized to their preferences and the firm then
delivers on this quasi-infinite assortment. Customers receive the product they want, but they need
to wait for it. Furthermore, this product strategy has implications for competitive pricing dynamics
(Mendelson and Parlaktūrk (2008)).

3.2.2 Location

Even in a world filled with “cloud” technology, most resources exist in some physical location.
Resources close to consumers can provide faster service. But moving resources closer to consumers
requires dispersing resources into more locations, with each location likely serving less demand. For
retailers the basic tradeoff is between scale and responsiveness - consumer demand can be more
quickly satisfied with nearby resources, but then more numerous locations leads to a loss of scale,
which reduces operating efficiency (Allon and Gurvich (2009)).
Although a tradeoff exists, firms have been successful at different ends of the location spectrum.
Zipcar changed the car rental industry by moving cars closer to consumers, so close that they could
walk to the car. Aldi operates small grocery stores (e.g., 1,000 m2 ), again, closer to consumers than
most of its competitors. In both of those examples, and others, the more numerous/closer locations
comes with a price - to somewhat mitigate the negative effects of lower scale, each of these firms
reduces the variety of products offered (relative to competitors who operate with fewer locations
serving greater demand).

24

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
 Amazon took a different approach. The initial Amazon model eliminated the nearby book stores
and replaced them with a single location for the entire country. This eliminated the expense of
storing inventory in costly retail locations and paying for store employees. But shipping books to
customers is costly and takes time - while the neighborhood bookstore could provide a book within
hours, Amazon could only do it within days.
To illustrate the potential for the Amazon model, Table 3 displays the differences in costs
for products that have different weekly demand rates in two operating models, e-tailing and the
traditional brick-and-mortar store. In both cases the product has a one-week lead time, the target
in-stock probability is 0.99, and orders are placed weekly. The holding cost for traditional retailing
is 60% of the product’s value (as evaluated in section 3.2.1), while e-tailing’s holding cost is 40%
to reflect the lower cost of warehouse space. However, e-tailing must incur 12% in fulfillment costs.
(Amazon’s fulfillment costs ranged from 10-13% of the product’s value from 2011-2016.) According
to the table, the best model for a product depends considerably on its demand rate. If a product’s
total demand (across the area covered by 500 traditional stores) is fewer than 40 units per week, the
e-tailing model dominates. For example, a product with total demand of 10 units per week would
achieve inventory turns of only 1.1 when that demand is spread across 500 stores. The resulting
cost is 56% of the product’s value, which is more than four times the cost that would be incurred
via the e-tailing model (13.2%). However, when the weekly demand surpasses about 40 units per
week, the traditional model begins to do better because it avoids the expensive cost of fulfillment.
In sum, if a product’s turns are about 4.7 of greater when sold through a traditional store, then
the traditional store is the better model. But if the traditional store is unable to achieve 4.7 turns,
possibly because the demand is too low when spread across many stores, then the e-tailing model is
better, potentially considerably better. This raises the question of whether there are enough slow
selling products for an e-tailer to have a decent business. In fact, Brynjolfsson et al. (2003) show
that a considerable portion of Amazon’s sales indeed come from slow selling items - even though
any one item has relatively small demand, the sum across millions of these items can be significant.
(And recall, a slow selling item for Amazon with its limited locations would be an extremely slow
selling item for a retailer with many locations.)
For manufacturing firms, the scale-responsiveness tradeoff continues to exist, but they also face
a tradeoff between production costs and responsiveness. Manufacturing in fewer, more distant
locations gives the firm more options to find locations or supplier with the lowest production costs.
Nike’s innovation was to move the manufacturing of its shoes from the United States to lower

25

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
Table 3: Operating cost differences (measured as a percentage of the product’s cost) between
e-tailing and traditional brick-and-mortar stores*
1 e-tailing warehouse 500 traditional stores
Weekly demand rate Annual turns Total cost (%) Annual turns Total cost (%) Best model
10 32.5 13.2 1.1 56.0 e-tailing
20 41.6 13.0 2.2 27.2 e-tailing
40 49.5 12.8 4.7 12.9 -
80 59.4 12.7 10.3 5.8 traditional
160 68.2 12.6 24.2 2.5 traditional
*Assumptions: Annual holding cost rate for traditional store is 60%, and for e-tailing it is 40%.
E-tailing incurs a fulfillment cost equal to 12% of the product’s value. In both models: one week
lead time; weekly ordering; 0.99 in-stock probability target.

cost Asian countries. Many other firms have followed a similar pattern as they seek to lower their
costs. However, manufacturing in Asia while selling in North America generally requires shipping
product by sea, which results in long lead times and lower responsiveness. (Air shipping is feasible
only for products that have a high value-to-weight ratio.) Given past energy prices, the cost of
transportation is of second order importance relative to typical labor cost differentials. This could
change in the future, which would motivate a shift in strategies.
Zara’s innovation was to emphasize responsiveness over costs. Zara manufacturers in Europe
and in nearby countries, which increases its manufacturing costs, but also gives it quick response
(QR) capabilities. With QR a firm is able to make a small initial production commitment, which
limits the chance the firm needs to mark down left over inventory. At the same time, QR gives the
firm the ability to respond later in the season with additional inventory if demand is strong - the
firm can use QR to avoid the pitfall of excess inventory while also avoiding the opportunity cost
of failing to respond to high demand. Even with exogenous demand, QR has been shown to be a
very effective strategy (Iyer and Bergen (1997)). But the true genius of the Zara model becomes
apparent when one considers the impact on strategic consumer behavior.
Consider again the dynamic pricing model from Section 3.1.3. The one change is now the
firm has QR capability to place a second order, after observing period 1 demand. This 2nd order
becomes available at the start of period 2 but each unit in that order costs about 25% more. The
advantage of the second order is that it is made with better information regarding demand. The
disadvantage is that it costs substantially more.
Without QR the firm’s best strategy is to never discount, earning a profit of $1534. With
QR the firm’s best strategy is to adopt value pricing (which is never optimal without QR), with
an initial price of $94, and a modest initial production quantity of 49.0 units. If demand turns

26

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
out to be weak, units are discounted to $30, but given the small initial production, consumers
know that a markdown isn’t likely (0.22 probability). However, if demand turns out to be strong,
additional units can be purchased, and sold at the profitable price of $94 without incurring lost
sales. Combining these features yields an expected profit of $3412, which is 222% greater than the
profit with the best non-QR strategy. Cachon and Swinney (2009) find a similar result in a related,
but broader, model. But the real confirmation of these ideas comes from Zara - their decision
to locate manufacturing closer to their primary demand resulted in higher production costs, but
gave the company a very powerful tool to avoid discounts, which led consumers to realize that
there is little to gain from strategic behavior, which enabled Zara to maintain high margins, earn
considerable profit, and grow to be the world’s largest fashion apparel retailer.12
Although we have made some progress to understand the Amazon and Zara business models,
the issue of location is by no means fully explored. For example, as Amazon has grown, and
as customers have developed a taste for fast shipping, Amazon has begun to operate with more
locations, which moves its inventory closer to consumers. Consequently, a customer might order
items that are stored in different fulfillment centers, which requires Amazon to decide between
sending all items in the order to one location so that they can be combined into one shipment to
the customer, or should Amazon send multiple shipments to the customer (called split shipments)
(Acimovic and Graves (2015))?
While Amazon remains largely a pure e-tailer, many traditional retailers now operate both
physical stores as well as an online store (e.g., Warby Parker for eyeglasses, and Bonobos and
Indochino for apparel). This creates a host of coordination issues between the channels. For
example, some retailers now offer the customers the option to “buy-online, pickup-in-store” (BOPS).
One might assume that this additional functionality would increase online sales. But in fact, Gallino
and Moreno (2016) find that BOPS can decrease online sales while increasing sales overall because
consumers use the online channel to do product research while using the store to do final product
inspection. And Gallino et al. (2017) demonstrate that this capability increases sales dispersion
across the product assortment, which can lead to higher inventory investments. Instead of using
the online channel for product information and the store for physical fulfillment (BOPS), a retailer
could reverse those roles: use physical showrooms to provide product information and the online
channel to provide fulfillment (Bell et al. (2016)).
12
While Zara doesn’t rely heavily on discounting, they nevertheless must discount some merchandize. See Gallien
and Caro (2012)for a discussion of how Zara optimizes markdowns.

27

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
 Finally, there is growing interest on how location decisions influence environmental sustainability
(e.g., Cachon (2014), and Kabra et al. (2015)) and pricing (Bimpikis et al. (2016), Besbes et al.
(2018)).

3.2.3 Control

Firms need a collection of resources to generate value. Each resource can be placed on a spectrum
that describes the degree to which the firm has control over the resource. At one extreme the firm
“owns” a resource and therefore can dictate when, where and how often it is used. At the other
extreme a firm “rents” a resource. In that case the firm makes requests to the owner of the resource
for its use, but use can only occur when an agreement is made between the two.
There are several benefits to ownership. To begin, it facilitates the coordination of resources.
For example, when two resources are owned and need to be used concurrently or sequentially,
then owning both allows the firm to guarantee that this can be done. For example, Zara owns its
trucks used to deliver product from its fulfillment center to its stores. Consequently, clothing can be
packaged in the truck that facilitates the store employee process of unloading the clothes in the store
to ready them for display and sale (e.g., the clothes arrive unwrinkled). Further, truck deliveries
can be timed when it is best for the store employees to receive (i.e., at times that are not crowded
with customers), rather than when it is convenient for the third party carrier to deliver. Second,
ownership facilitates the standardization of a process. There are two benefits to standardization.
First, customers may value standardized processes because the delivered quality has little variation
over time or locations. Second, standardization allows a firm to purchase less expensive resources. A
machine that does a single, standardized task, is less expensive than a machine that must be flexible.
Wages for workers to complete routine/standardized tasks are lower than wages for workers that do
varied/unstandardized tasks. This occurs in large part because unstandardized tasks require more
judgment on the part of the employee, which requires an employee with greater skill and training.
The challenge with ownership is utilization. A firm’s need for a resource can be volatile, and
volatile needs lead to poor utilization, which increases the cost of the resource. Zara’s outbound
delivery to its stores are completed with a full truck, but then trucks return mostly empty. A third
party carrier’s trucks are loaded a higher fraction of time, thereby lowering the cost of the asset.
Employees skills may not be needed all of the time, but the firm may be responsible for their salary
throughout the year. The advantage of renting a resource is that the firm pays for the resource
only when needed, assuming reasonable terms can be agreed upon with the renter. Hence, the key

28

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
strategic decision for the firm is to get the correct balance with its resources between the virtues
of control/ownership and the flexibility of renting.
Initial research on control focused on the theory of the firm boundary, i.e., which of a firm’s
resources are included inside the firm and which are outside the firm. This can also be thought
of as the “outsourcing decision”, which is often referred to as the “make-buy” decision. The main
insight is that there are frictions both within the firm (e.g., agency costs, i.e., workers/managers
need to be compensated and may not have incentives that are fully aligned with the owners of the
firm) and between firms (e.g., search and coordination costs due to incomplete contracts). The
equilibrium structure is likely to make the best tradeoff between these costs (e.g., Grossman and
Hart (1986), Grossman and Helpman (2002)). And the balance in this tradeoff tips towards more
outsourcing when technology reduces search costs outside of a firm (Malone (1987), Brynjolfsson
et al. (1994)) and enables more complete contracting between firms (Cachon (2003), Cachon and
Lariviere (2005)).
Once a firm has decided to rent a resource, the firm must manage its acquisition of the resource.
For instance, Van Mieghem (1999) demonstrates that it is not always in a firm’s interest to try to
specify and account for a priori all contingencies in the relationship between two firms - sometimes
it is in their interest to decide their terms after observing information. Belavina and Girotra (2012)
explain why even a large firm (e.g., Walmart) might choose to source through an intermediary (e.g.
Li and Fung) rather than directly with suppliers. They demonstrate that an intermediary is better
able to create long-run relationships with suppliers than the firm itself. Long-run relationships
mitigate some of the frictions of disintermediation (e.g., supplier shirking), so a supply chain with
an intermediary can generate more value than a supply chain without one, which justifies the
existence of the intermediary. Ang et al. (2017) explore not only the boundary between two firms
but rather, the structure of the supply chain. In particular, they investigate how a firm should best
mitigate the risk of production disruptions when managing its suppliers based on how the suppliers
source one level higher in the chain from their own suppliers.
The previous examples generally consider the control boundary between a firm and a few other
firms. The rise of the “sharing economy” creates situations in which the typical “one to few”
relationship has become “one to many”, illustrating an extreme penchant for renting over owner-
ship. For example, Uber provides transportation without owning the vast majority of the vehicles
actually used to provide the service. Nor are Uber drivers Uber employees. Instead, they are inde-
pendent contractors. Consequently, Uber has limited control over when they work and how they

29

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
work. AirBnB is another new firm that eschews owning assets. They provide a platform in which
individuals (called hosts) can offer their property or a portion of their property (e.g., a single room)
for rent to others. AirBnb is not able to specify the design of the property, nor when it is made
available, nor the price the host charges. Hence, AirBnB needs to understand how its hosts operate
and then how it can subtlety influence them. For instance, Cui et al. (2016) study the degree to
which hosts exhibit discriminatory behavior. Li et al. (2016) study how hosts price, and how they
could price better if they had more experience.
A surprising feature of the success of Uber and AirBnB is that their strategies starkly contrasts
those of previously successful companies. For example, McDonalds became a dominant player in
the (quick serve) restaurant industry because it was able to standardized many of its processes -
how its beef was raised, how its potatoes were grown and how its hamburgers were cooked. Zara is
able to provide timely fashions in large part because it has vertically integrated its supply chain to a
larger extent than its competitors, thereby enabling greater coordination of its assets. In contrast,
Uber and AirBnB have taken the approach to reduce control over their resources. This has enabled
them to increase utilization of the resources (which lowers costs) but it sacrifices the benefits of
standardization. AirBnB guests have more uncertainty as to the quality of their stay, and Uber
customers are never sure exactly what type of vehicle they will ride in. What is remarkable about
these companies is not that there are negative consequences to renting but rather, those negative
consequences can either be managed or are small enough relative to the benefits of renting.

4 Conclusion

A firm’s business model defines how it delivers (supply model) and profits from (revenue model)
the customer value it creates through the effective satisfaction of needs. The revenue model in-
volves high level decisions regarding the mechanisms for determining prices (e.g., posted prices vs
auctions), the structure of the transaction terms (e.g., per-use or fixed fees), and the dynamics of
adjustments to transaction terms over time. The supply model requires decisions on the timing of
resource activation, the locations of resources and the level of control over those resources. The
combination of these decisions determines the amount of customer value generated and ultimately
the viability of the firm.
A business model innovation generally involves one or more novelties in the revenue or supply
models (or both). These novelties often challenge implicit assumptions, assumptions that are

30

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
so taken for granted that they would be omitted from a list of key assumptions. For example,
traditional fashion apparel retailing involved outsourcing production to overseas suppliers and a
hi-low pricing model that heavily relies on markdowns to clear excess inventory. Zara’s fast-fashion
business model innovation begins with its supply model decisions on location (local production) and
control (vertical integration). These enable Zara to experience far fewer cases of over production
because (a) design decisions are made closer to the selling season (an innovation in the timing
component of the supply model) and (b) its quick response capabilities allow it to be conservative
with initial production commitments. Consequently, Zara also innovated with its revenue model,
adopting a “value pricing” approach which uses markdowns sparingly. Like Zara, Amazon’s e-
tailing business model is grounded in the location decision of its supply model. But instead of
moving closer to demand, Amazon moved further away, initially choosing to stock inventory in a
single warehouse rather than in thousands of retail stores close to consumers. Turns out that many
customers were willing to give up the ability to physically browse books (the traditional assumption)
so that they could have access to a vastly greater breadth of books. More recently, the ride-sharing
market has challenged our presumptions of urban transportation, which was dominated by taxis.
Uber’s entry to this market involved two key business model innovations. First, it relinquished
control of two key assets, the drivers and their vehicles. This reduced the standardization of their
process in the sense that customers cannot know what type of car they will ride in. Second, they
abandoned the rigid pricing of taxis and replaced it with a revenue model based on substantial and
frequent adjustments to pricing. The combination of these two innovations allows Uber to provide
far more vehicles on the road, especially when demand is high, which generates considerable value
both for Uber and consumers.
The business model examples given suggest that there have been many successful models and
they have taken diverse forms. Nevertheless, in all cases these firms made decisions regarding the
elements of their revenue and supply models. New approaches have emerged when firms discovered
new combinations of these decisions that in net yield better results than the industry status quo.
Often (but not always, e.g., Zara) these new business models are enabled primarily by changes
in technology. This pattern is likely to continue. For example, autonomous vehicles will alter the
economics of transportation, which could not only influence how we move people and goods around,
but could also influence where we choose to live. Technology could lead to radical changes in educa-
tion. One possibility is that education becomes more standardized (i.e., the organization providing
the education exerts more control over the process). Gone may be the days when professors across

31

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
 Table 4: Preferences for four customer types over two products
Customer type Value for product A Value for product B Fraction of market
1 1 0 0.25
2 0.7 0.5 0.25
3 0.5 0.7 0.25
4 0 1 0.25

universities teach essentially customized classes. A more standardized approach to education may
be delivered with less expensive workers (standardized processes requires less skill and judgment),
which would expand the potential scope of education and lower its cost. The extreme opposite on
the standardization spectrum could also happen - technology could be used to customize educa-
tion to each student. And these paths are also conceivable in medicine - more standardized and
lower cost, or customized and higher quality. Beyond computer and communication technology, we
are likely to experience substantial changes in our energy portfolio, which could influence where
production occurs. In each of these cases, a firm may discover that moving its resource locations,
or changing its pricing mechanism, or changing which resources it owns, or some mixture of these
decisions leads to more consumer value and/or lower costs. When that happens, the next Zara,
Amazon, or Uber will be well on its way to dominating its market.

5 Appendix

5.1 Probabilistic selling

Probabilistic selling allows the firm to price discriminate between customers that vary in the
strength of their preferences across the product line. A simple example illustrates why this can
improve revenue. Table 4 displays the preferences of four types of customers for two products, A
and B. The 1st and 4th types have strong preferences for only product A or B. The middle types
are relatively indifferent between the two products. Normalize total demand in the market to equal
1, split evenly across the four types.
If the firm can only sell the two products, then the optimal pricing strategy is to charge 0.7 for
each one, earning a total revenue of 0.7. (Charging 1 for each product only sells to the extreme
types, earning 0.5 in revenue.)
To do better, the firm can create product C, which is a virtual/probabilistic product: if a
customer selects product C then they are given product A with a 0.5 probability and product B
with a 0.5 probability. The optimal prices are 1 for products A and B, and 0.6 for product C. The

32

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
Table 5: Profits and prices with subscriptions, “s”, and per-use pricing, “p”, with two capacity
levels.

Capacity πs πp ps pp
1 1

2
q=1 φ (1 − 2c − φw) − k 2φ 1−c− −k
2 φw φ (1/2 − φw/2) (2 + 2c − φw) /4
1 2
q=2 φ (1 − 2c) − 2k − c) − 2k
2 φ (1 φ/2 (1 + c) /2

extreme types (1 and 4) prefer to purchase their preferred product and the middle types prefer the
probabilistic product C. Total revenue is 0.8, substantially higher than revenue without the virtual
product.
For probabilistic selling to work the expected value of the relatively indifferent types must be
sufficiently high. Say the indifferent types’ values are 0.5 and 0.3 for the two products. Now they
are willing to pay at most 0.40 for the probabilistic product C. But if 0.4 is offered for product C,
the other types (1 and 4) also purchase product C and total revenue drops to 0.4.
Selling virtual products is costly to the system: total surplus is 0.85 when the firm sells only
products A and B, but is 0.80 with the virtual product included. It works in the firm’s favor
because the firm is able to capture a greater share of that surplus: the firm’s share is 0.7 / 0.85 =
82% with just the two products, and 0.8 / 0.8 = 100% with the virtual product. This undesirable
feature of probabilistic selling could work against it the domain of public perception.

5.2 Subscription vs per-use

This section describes a simplified version of the model in Cachon and Feldman (2011). There is
one firm and two potential customers. Each customer values the service with probability φ. If they
value the service, their value is uniformly distributed on the interval [0, 1], otherwise they have no
value for the service. To serve a customer requires one unit of capacity and the firm can choose
capacity q = 1 or q = 2. If q = 2 then both customers can be served immediately. If q = 1 then one
customer can be served immediately and a second customer is served with delay cost w. The firm
can charge a subscription to each customer or charge per use. With a subscription, each customer
pays before observing their value and before observing the queue length. With per use, customers
decide whether to pay after observing their value for the good and the queue length. The cost for
each unit of capacity is k. The marginal cost to serve a customer is c < 1/2.
There are four cases to consider: two types of contracts, two capacity levels. Table 5displays
the profits and prices in each case.
To compare the strategies, first consider when congestion costs become small, i.e., as w → 0. In

33

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
this case subscription with q = 1 dominates because there is no need for a second unit of capacity if
customers don’t care about about congestion (πs (q = 2) < πs (q = 1)) and subscription generates
more revenue than per-use (πp (q = 1) < πs (q = 1)). However, when capacity is limited, q = 1,
subscription profit is more sensitive to congestion costs than per use:

∂πs (q = 1) ∂πp (q = 1)
< <0
∂w ∂w

This disadvantage of subscriptions is intuitive - subscriptions do not regulate usage and therefore
lead to excessive consumption which leads to additional congestion. But even though subscription
is more sensitive to congestion, it starts with a higher profit, and therefore may be superior to
per use even with high congestion costs (but not so high as to render either contract infeasible).
To illustrate this point, the worse case for subscriptions relative to per use pricing occurs when
w = k/φ2 : for smaller w subscription chooses q = 1 and its profit falls faster than per use profit,
whereas for larger w subscription chooses q = 2, its profit is independent of w but per use profit
continues to decrease with w. Even in the worse case for subscription (i.e., w = k/φ2 ), subscription
yields a higher profit than per use for all c ≤ ĉ, where

√ k
ĉ = 2−1− .
2φ

So subscription is surely better than per use when the marginal cost of service is low. But this
doesn’t tell the entire story. If
k √
< 3 − 2 2 ≈ 0.17
φ

then subscription there exists a threshold c0 , c0 < ĉ, such that for all c ∈ [c0 , ĉ] subscription earns a
positive profit but per use does not, i.e., πp < 0 < πs . In other words, it is possible that subscrip-
tions yield higher profits than per use for all feasible values of congestion cost. Counter-intuitively,
when it would seem that controlling congestion is most important (when consumers strongly dislike
congestion), it can be optimal to use the pricing scheme that doesn’t control congestion (subscrip-
tions) because it is more important to generate revenue to enable more investment in capacity than
it is to control consumer behavior with limited capacity.

34

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
5.3 Dynamic pricing

This model extends the one studied in Su and Zhang (2008). One firm sells a single product over a
selling season with two periods. The product is sold for the full price, p, in period 1 and potentially
sold for the discount price, v, in period 2. The firm purchases q units before the selling season at
a cost of c per unit. There is a random amount of demand (i.e., customers) in period 1, which
creates the risk of left over inventory. Let F (·) and f (·) be the distribution and density functions
of demand, which satisfy the reasonable regularity property of an increasing generalized failure rate
(IGFR) (Lariviere (2006)). Let d be realized demand, and µ be expected demand.
There are two types of consumers. A fixed fraction of demand, α, is “myopic”: myopic con-
sumers receive utility u = 1 if they purchase in period 1 and they only consider purchasing in period
1 (e.g., the earn zero utility if they purchase in period 2). The other 1 − α fraction of demand is
“strategic”: strategic consumers choose which period to make a purchase and receive u = 1 utility
for a purchase in either period. Let β be the fraction of demand that attempts to purchase in
period 1. If the number of period 1 customers exceeds the firm’s quantity, q ≤ βd, then units are
randomly allocated. Otherwise, the unsold product from period 1 is sold in period 2 for v per unit,
v < c. Strategic consumers who chose to wait to period 2 to purchase are the first to be served
in period 2. If any inventory remains in period 2 after those consumers are served, the remaining
inventory is sold for v to a large pool of discount shoppers. Thus, all remaining inventory at the
start of period 2 is sold for v.
Strategic consumers choose in which period to purchase to maximize their net utility given their
rational expectations. In particular, they believe there is a φt probability that they receive a unit
if they choose to purchase in period t. In equilibrium, their beliefs are correct. For consumers,
purchasing in period 1 has a higher price, v < p, but also a higher likelihood of product availability,
φ2 < φ1 .
Let Sβ (q) = E [min {βd, q}] be the expected sales function when α is the fraction of demand
that purchases in period 1:

R q/β
Sβ (q) = 0 βxf (x) dx + (1 − F (q/β)) q = βS (q/β)

where S (q) = S1 (q) for notational convenience.

The firm’s optimal period 1 price with the value strategy is the maximum price that induces

35

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
consumers to purchase in period 1 given their beliefs:

p = 1 − (1 − v) φ2 /φ1 (2)

Assuming in equilibrium strategics purchase in period 1, then strategics believe they can purchase
a unit in period 1 with probability φ1 = S1 (q) /µ. (Note, from the consumer’s perspective, df (d)/µ
is the conditional probability of being in a market with d units of demand when all consumers
participate.13 ). In period 2 the correct belief for strategics, is

Z q
xf (x)
φ2 = dx
0 µ

The firm’s profit with price p, β fraction of consumers purchasing in period 1, and surely
discounting in period 2 is :

π (p, q, β) = (p − v) Sβ (q) − (c − v) q (3)

With the high-low price strategy p = 1 and β = α. With value pricing p < 1 and β = 1. With
either strategy the optimal quantity satisfies

p−c
F (q ∗ /β) = p−v
(4)

(Note, q must be consistent with expectations but changing q cannot change consumer expecta-
tions.) Given the optimal price, (2), the quantity equation for the value-price strategy can be
written as
qF (q ∗ )2 c−v
= (5)
S (q ∗ ) 1−v

The left hand side of (5) is decreasing for an IGFR distribution, which implies there exists a unique
optimal q ∗ .
With the never-discount strategy the firm selects p = 1, strategics purchase in period 1, and to
evaluate the firm’s profit it can be assumed that the firm’s period 2 price is v = 0.
With the refund pricing strategy the strategic consumers are indifferent between purchasing in
13
To explain, suppose demand can either be d = 1 or d = 99, each equally likely. A consumer should not assume
that there is a 0.5 probability of being the only consumer in the market (d = 1). Instead, the conditional probability
of being in a market with d consumers is df (d) /µ: the probability a consumer is the only one in the market is
1f (1) /µ = 0.01 and the probability the consumer is in a market with many others is 99f (99)/µ = 0.99.

36

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
period 1 or 2. To explain, their expected period 1 surplus is (1 − v) φ2 because they earn the 1 − v
surplus if there is a period 2 discount. Their period 2 surplus is the same - they earn 1 − v in period
2 only if there is a discount, which occurs with probability φ2 . Thus, with this refund strategy, the
firm’s profit is
π (q) = (1 − v) qF (q) − (c − v) q

The firm’s optimal quantity, q ∗ , uniquely satisfies (because demand is IGFR) the following equation

q ∗ f (q ∗ )
 
∗ c−v
F (q ) 1 − =
F (q ∗ ) 1−v

With the quantity commitment contract, the price, given q, is given by (2). Substitution into
the profit function yields the same profit function as with the refund strategy.
To compare the various strategies for the firm, construct a set of 108 scenarios from all combi-
nations of the following parameters in which demand is modeled with a Gamma distribution:

α ∈ {0.25, 0.5, 0.75} µ = 100 c =∈ {0.25, 0.5, 0.75}

σ/µ =∈ {0.25, 0.5, 1, 1.5} v/c =∈ {0.3, 0.6, 0.9}

The best of the firm’s four strategies always earns less than the firm’s upper bound on profit (which
occurs when all consumers are myopic, α = 1), and earns on average only 72%, 74% and 82% of
that bound when the share of myopic consumers is α = 0.25, α = 0.50 and α = 0.75, respectively.
Trying to woo strategic consumers with a low initial price (the value strategy) is generally not
effective. It can be better than the hi-low strategy only when there are many strategics and
few myopic consumers (α = 0.25). However, the value strategy is never optimal among all of the
strategies. The never-discount strategy is optimal in 80 of the 108 scenarios (74%). Because salvage
revenues are of second order importance relative to full price sales, committing to never discount
only performs poorly in extreme scenarios in which it is very costly to forgo salvage revenues (high
demand uncertainty, high product cost and high salvage prices). Finally, while committing to a
full-refund can be optimal, it rarely is: it is optimal in only 6 of 108 scenarios. (That said, there
may be better designs for refund policies.)

5.4 Timing

See the main text for the description of the model. First consider inventory. During the low phase
the firm can build inventory that will allow it to serve demand during the high period. Average

37

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
inventory is  
1 λh − λl
I= (λh − µ) τ
4 µ − λl

Let µ = φδ + λ where φ ∈ [0, 1] and δ = (λh − λl ) /2. The firm chooses φ. The inventory equation
can now be written as  
1 1−φ
I = τδ =Q
2 1+φ

which is also the average queue length, Q, if the firm chooses to make customers wait rather than
to build inventory.
Optimization. Let π be the firm’s total cost. With the make-to-stock approach
    
1 1−φ 1 1−φ
π = µ + hI = (φδ + λ) + h τ δ = δ φ + τh +λ
2 1+φ 2 1+φ

π is convex in φ. The resulting cost is


λ + 21 δτ h τh < 1





 √
π∗ = λ + δ 2 τ h − 1 − 1 τ h
 
1 ≤ τh ≤ 4 (6)
 2



λ + δ

4 < τh

The firm’s cost per unit of demand (π ∗ /λ) increases with both forms of seasonal variability, the
amplitude(δ) and the duration (τ ). If doing work in advance is relatively inexpensive (h < 1/τ )
then the firm chooses the minimum capacity, uses the entire low season to build inventory to satisfy
high season demand. On the other hand, if doing work in advance is sufficiently costly (4/τ < h)
then the firm chooses the maximum capacity and doesn’t do any work in advance. If the firm just
uses a queue, then the results are identical to (6) with the exception that h is replaced with the
cost of waiting, w.
Now consider the case in which the firm is able to do some of the work in advance and some
that is done only after demand arrives. Say γ fraction of excess demand during the high phase is
completed in advance. The amount of excess high phase work done in advance is γ (λh − µ) τ . The
average inventory of work is
   
1 τI + τ 1 1−φ
I = γ (λh − µ) τ = γδτ (1 + γ + (1 − γ) φ)
2 2τ 4 1+φ

The remaining (1 − γ) (λh − µ) τ of work during the high phase creates a queue that is then drawn

38

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
down during the low-phase. The average queue length is then
 
1 1−φ
Q = δτ (1 − γ) (2 − γ (1 − φ))
4 1+φ

The firm’s cost is

   
π = µ + 14 hγτ δ (1 + γ + (1 − γ) φ) 1−φ
1+φ + 1
4 w (1 − γ) τ δ (2 − γ (1 − φ)) 1−φ
   1+φ
1 1−φ
= δ φ + 4 τ 1+φ (hγ (1 + φ + γ (1 − φ)) + w (1 − γ) (2 − γ (1 − φ))) + λ

Holding φ fixed, the optimal γ (unconstrained) is

(3 − φ) (w/h) − (1 + φ)
γ∗ =
2 (w/h + 1) (1 − φ)

Note that
∂γ ∗ w/h − 1
=
∂φ (w/h + 1) (1 − φ)2
which is positive for 0 < w/h. It is possible to show that the upper bound on γ < 1 is achieved
with φ = 1 and w/h = 3. Thus, if the cost of waiting is more than 3 times the cost of holding
inventory, the optimal solution for all φ is γ = 1, which means to build inventory. Analogously, if
w/h < 1/3, then γ = 0 is optimal for all values of φ : if waiting costs are low, then it is best to
build a queue.

References

Acimovic, Jason, Stephen Graves. 2015. Making better fulfillment decisions on the fly in an online
retail environment. Manufacturing & Service Operations Management 17 34–51.

Alexandrov, Alexei, Martin A. Lariviere. 2012. Are reservations recommended? Manufacturing &
Service Operations Management 14(2) 218–230.

Allon, Gad, Itay Gurvich. 2009. Pricing and dimensioning competiting large-scale service providers.
Manufacturing & Service Operations Management 12 449–469.

Amaldoss, Wilfred, Sanjay Jain. 2008. Joint bidding in the name-your-own-price channel: a strate-
gic analysis. Management Science 54(10) 1685–1699.

39

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
Ang, Erjie, Dan A. Iancu, Robert Swinney. 2017. Disruption risk and optimal sourcing in multitier
supply networks. Management Science 63 2397–2419.

Bakos, Yannis, Erik Brynjolfsson. 1999. Bundling information goods: Pricing, profits, and efficiency.
Management Science 45(12) 1613–1630.

Belavina, E., Karan Girotra. 2012. The relational advantages of intermediation. Management
Science 1614–1631.

Bell, David, Santiago Gallino, Antonio Moreno. 2016. Offline showrooms in omni-channel retail:
Demand and operational benefits. forthcoming Management Science .

Besanko, D., W. L. Winston. 1990. Optimal price skimming by a monopolist facing rational
consumers. Management Science 36(5) 555–567.

Besbes, Omar, Francisco Castro, Ilan Lobel. 2018. Surge pricing and its spatial supply response.

Besbes, Omar, Assaf Zeevi. 2015. On the (surprising) sufficiency of linear models for dynamic
pricing with demand learning. Management Science 61 723–739.

Bimpikis, Kosta, Ozan Candogan, Daniela Saban. 2016. Spatial pricing in ride-sharing networks.
Stanford University working paper .

Blattberg, Robert C., Gary D. Eppen, Joshua Lieberman. 1981. A theoretical and empirical eval-
uation of price deals for consumer nondurables. Journal of Marketing 45(1) 116–129.

Brynjolfsson, Eric, Ye Hu, Michael D. Smith. 2003. Consumer surplus in the digital economy:
Estimating the value of increased product variety at online booksellers. Management Science 49
1580–1596.

Brynjolfsson, Eric, Thomas W. Malone, Vijay Gurbaxani, Ajit Kambil. 1994. Does information
technology lead to smaller firms? Management Science 40 1628–1644.

Bulow, J. 1982. Durable-goods monopolist.

Cachon, Gérard P. 2003. Supply chain coordination with contracts. Handbooks in operations
research and management science 11 227–339.

Cachon, Gérard P. 2014. Retail store density and the cost of greenhouse gas emissions. Management
Science 60 1907–1925.

40

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
Cachon, Gérard P., Kaitlin M. Daniels, Ruben Lobel. 2015. The role of surge pricing on a ser-
vice platform with self-scheduling capacity. forthcoming Manufacturing & Service Operations
Management .

Cachon, Gérard P., Pnina Feldman. 2011. Pricing services subject to congestion: charge per-use
fees or sell subscriptions? Manufacturing & Service Operations Management 13(2) 244–260.

Cachon, Gérard P, Pnina Feldman. 2018. Pricing capacity over time and recourse strategies: facil-
itate reselling, offer refunds/options, or overbook?

Cachon, Gérard P, Martin A Lariviere. 2005. Supply chain coordination with revenue-sharing
contracts: strengths and limitations. Management science 51(1) 30–44.

Cachon, Gérard P., Robert Swinney. 2009. Purchasing, pricing, and quick response in the presence
of strategic consumers. Management Science 55(3) 497–511.

Cachon, Gérard P., Robert Swinney. 2016. The value of fast fashion: Quick response, enhanced
design, and strategic consumer behavior. Management Science 62(2) 410–435.

Cachon, Gérard P., Fuqiang Zhang. 2006. Procuring fast delivery: sole sourcing with information
asymmetry. Management Science 52(6) 881–896.

Casadesus-Masanell, Ramon, Gastón Llanes. 2011. Mixed source. Management Science 57 1212–
1230.

Chen, Ningyuan, Guillermo Gallego. 2016. Do consumers benefit from dynamic pricing? Yale
University working paper .

Cheng, Hsing Kenneth, Yipeng Liu. 2012. Optimal software free trial strategy: The impact of
network externalities and consumer uncertainty. Information Systems Research 23 488–504.

Chevalier, J, A Goolsbee. 2009. Are durable goods consumers forward-looking? evidence from
college textbooks.

Chu, Leon Yang, Hao Zhang. 2011. Optimal preorder strategy with endogenous information control.
Management Science 57 1055–1077.

Clifford, Stephanie, Catherine Rampell. 2013. Sometimes, we want prices to fool us.

41

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
Cui, Ruomeng, Jun Li, Dennis J. Zhang. 2016. Discrimination with incomplete information in the
sharing economy: evidence from field experiments on airbnb. working paper Indiana University
.

Cui, Yao, Izak Duenyas, Özge Şahin. 2014. Should event organizers prevent resale of tickets?
Management Science 60(9) 2160–2179.

Desai, Preyas, Devavrat Purohit. 1998. Leasing and selling: optimal marketing strategies for a
durable goods firm.

Desai, Preyas, Devavrat Purohit. 2004. ”let me talk to my manager”: Haggling in a competitive
environment.

Etzion, Hila, Edieal Pinker, Abraham Seidmann. 2006. Analyzing the simultaneous use of auctions
and posted prices for online selling.

Fay, Scott. 2004. Partial-repeat bidding in the name-your-own-price channel. Marketing Science
23(3) 407–418.

Fay, Scott, Jinhong Xie. 2008. Probabilistic goods: A creative way of selling products and services.
Marketing Science 27(4) 674–690.

Gale, Ian L, Thomas J Holmes. 1993. Advance-purchase discounts and monopoly allocation of
capacity. The American Economic Review 135–146.

Gallego, Guillermo, Ozge Sahin. 2010. Revenue management with partially refundable fares. Op-
erations Research 58(4) 817–833.

Gallien, Jeremie, Felipe Caro. 2012. Clearance pricing optimization for a fast-fashion retailer.
Operations Research 60 1404–1422.

Gallino, Santiago, Anthonio Moreno. 2016. Integration of online and offline channels in retail: The
impact of sharing reliable inventory availability information. forthcoming Management Science .

Gallino, Santiago, Antonio Moreno, Ioannis Stamatopoulis. 2017. Channel integration, sales dis-
persion, and inventory management. Management Science 63 2813–2831.

Gneezy, Ayelet, Alex Imas, Amber Brown, Leif D. Nelson, Michael Norton. 2012. Paying to be
nice: consistency and costly prosocial behavior. Management Science 58(6) 179–187.

42

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
Grossman, Gene M., Elhanan Helpman. 2002. Integration versus outsourcing in industry equilib-
rium. Quarterly Journal of Economics 85–120.

Grossman, Sanford, Oliver Hart. 1986. The costs and benefits of ownership: a theory of vertical
and lateral integration. Journal of Political Economy 94 691–719.

Guo, Liang. 2009. Service cancellation and competitive refund policy. Marketing Science 28(5)
901–917.

Hann, Il-Horn, Christian Terwiesch. 2003. Measuring the frictional costs of online transactions:
The case of a name-your-own-price channel. Management Science 49(11) 1563–1579.

Hansen, Susan. 2012. How zara grew into the world’s largest fashion retailer.

Ho, Teck-Hua, Christopher S. Tang, David R. Bell. 1998. Rational shopping behavior and the
option value of variable pricing. Management Science 44 S145–S160.

Hoch, Steve, Xavier Drèze, Mary E. Purk. 1994. Edlp, hi-lo, and margin arithmetic. Journal of
Marketing 58(4) 16–27.

Hosanagar, Kartik, Vibhanshu Abhishek. 2013. Optimal bidding in multi-item multi-slot sponsored
search auctions.

Hu, Ming, Yun Zhou. 2017. Price, wage and fixed commission in on-demand matching.

Iyer, Ananth, M.E. Bergen. 1997. Quick-response in manufacturer-retailer channels. Management

Science 43 559–570.

Jerath, Kinshuk, Serguei Netessine, Senthil K. Veeraraghavan. 2010. Revenue management with
strategic customers: Last-minute selling and opaque selling. Management Science 56(3) 430–448.

Kabra, A., E. Belavina, K. Girotra. 2015. Bike-share systems: Accessibility and availability. Uni-
versity of Chicago working paper .

Kumar, Vineet. 2014. Making ”freemium” work. Harvard Business Review .

Lai, Guoming, Laurens G. Debo, Katia Sycara. 2010. Buy now and match later: Impact of pos-
terior price matching on profit with strategic consumers. Manufacturing & Service Operations
Management 12(1) 33–55.

43

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
Lariviere, Martin. 2006. A note probability distributions with increasing generalized failure rates.
Operations Research 54(3) 602–604.

Lazear, Edward. 1986. Retail pricing and clearance sales. American Economic Review 76(1) 14–32.

Lee, Hau L., V. Padmanabhan, Seungjin Whang. 1997. Information distortion in a supply chain:
the bullwhip effect. Management Science 43(4) 546–558.

Lee, Hau L., Christopher S. Tang. 1997. Modelling the costs and benefits of delayed product
differentiation. Management Science 43 40–53.

Li, Jun, Nelson Granados, Serguei Netessine. 2014. Are consumers strategic? structural estimation
from the air-travel industry.

Li, Jun, Antonio Moreno, Dennis J. Zhang. 2016. Pros vs. joes: agent pricing behavior in the
sharing economy. working paper University of Michigan .

Malone, Thomas W. 1987. Modeling coordination in organizations and markets. Management

Science 33 1317–1332.

McWilliams, Gary. Nov 8 2004. Analyzing customers, best buy decides not all are welcome.

Mendelson, Haim, Ali K. Parlaktūrk. 2008. Product-line competition: customization vs prolifera-

tion. Management Science 54(2) 2039–2053.

Moon, Ken, Kosta Bimpikis, Haim Mendelson. 2016. Randomized markdowns and online monitor-
ing.

Mukhopadhyay, Tridas, Sunder Kekre. 2002. Strategic and operational benefits of electronic inte-
gration in b2b procurement processes.

Nicolae, Mariana, Mazhar Arikan, Vinayak Deshpande, Mark Ferguson. 2017. Do bags fly free? an
empirical analysis of the operational implications of airline baggage fees.

Özalp Özer, Yanchong Zheng. 2016. Markdown or everyday low price? the role of behavioral
motives. Management Science 62(2) 326–346.

Parker, Geoffrey G., Marshall W. Van Alstyne. 2005. Two-sided network effects: a theory of
information product design. Management Science 51 1494–1504.

44

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
Rob, Rafael, Joel Waldfogel. 2006. Piracy on the high c’s: music downloading, sales displacement,
and social welfare in a sample of college students. Journal of Law and Economics 49 29–62.

Rochet, Jean-Charles, Jean Tirole. 2006. Two-sided markets: a progress report. The RAND Journal
of Economics 37(3) 645–667.

Schmidt, Klaus M., Martin Spann, Robert Zeithammer. 2015. Pay what you want as a marketing
strategy in monopolistic and competitive markets. Management Science 61(6) 1217–1236.

Simonsohn, Uri, Dan Ariely. 2008. When rational sellers face nonrational buyers: evidence of
herding on ebay.

Spann, Martin, Gerard J. Tellis. 2006. Does the internet promote better consumer decisions? the
case of name-your-own-price auctions. Journal of Marketing 70(1) 65–78.

Stokey, N. 1981. Rational expectations and durable goods pricing.

Su, Xuanming. 2007. Intertemporal pricing with strategic consumer behavior. Management Science
53 726–741.

Su, Xuanming. 2010. Optimal pricing with speculators and strategic consumers. Management
Science 56(1) 25–40.

Su, Xuanming, Fuqiang Zhang. 2008. Strategic customer behavior, commitment, and supply chain
performance. Management Science 54(10) 1759–1773.

Tereyagoglu, Necati, Senthil Veeraraghavan. 2012. Selling to conspicuous consumers: pricing,

production, and sourcing decisions.

Terwiesch, Christian, Il-Horn Hann, Sergei Savin. 2005. Online haggling at a name-your-own-price
retailer: theory and application. Management Science 51(3) 339–351.

Tuo Wang, Rabikar Chatterjee, Esther Gal-Or. 2009. The name-your-own-price channel in the
travel industry: An analytical exploration. Management Science 55(6) 968–979.

Van Mieghem, Jan A. 1999. Coordinating investment, production and subcontracting. Management
Science 45 954–971.

Wang, R. 1993. Auctions versus posted-price selling.

45

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441
Xie, J., Steve Shugan. 2001. Electronic tickets, smart cards, and online prepayments: When and
how to advance sell. Marketing Science 20(3) 219–243.

Xie, Jinhong, Eitan Gerstner. 2007. Service escape: Profiting from customer cancellations. Mar-
keting Science 26(1) 18–30.

Yin, Rui, Yossi Aviv, Amit Pazgal, Christopher S. Tang. 2009. Optimal markdown pricing: Impli-
cations of inventory display formats in the presence of strategic customers. Management Science
55(8) 1391–1408.

Yu, Man, Laurens Debo, Roman Kapuscinski. 2015. Strategic waiting for consumer-generated
quality information: dynamic pricing of new experience goods. Management Science 61 723–
739.

Zhang, Laurina. 2016. Intellectual property strategy and the long tail: Evidence from the recorded
music industry. forthcoming Management Science .

46

Electroniccopy
Electronic copyavailable
available at:
at: https://ssrn.com/abstract=3250441
https://ssrn.com/abstract=3250441