MS&E 238 Final Paper

Online Marketplaces

MS&E 238 final paper

Andrew Adams and Lance Martin

Structure of marketplaces -

The natural place to begin a discussion of marketplaces is with defining what we mean when we use the term; as well, it is important to disambiguate what we don’t mean when we use the term. When we refer to marketplaces, we mean entities that facilitate transactions between buyers and sellers; importantly, they act as a third party to the principal transaction at hand. The marketplace is thus not the seller (nor a buyer), and as such it does not own the inventory or services transacted on the platform. If they did, the entities wouldn’t be marketplaces, but rather traditional retailers. Accordingly, it would be incorrect to state that recognized digital marketplaces like Uber, Etsy, and AirBnb sell transportation, handcrafted goods, or lodging [1]. Rather, they facilitate transactions between drivers and riders; artisans and buyers; and hosts and lodgers.

So how do marketplaces facilitate such transactions; what do they actually sell? Marketplaces sell a reduction in transaction costs. As Stanford professor and marketplace expert Ramesh Johari defines, these transaction costs can come in the following types:

1. Search and Information Costs:

Search captures the cost of prospective buyers and sellers finding each other and information captures the cost of surfacing information relevant to either party making an informed decision as to whether they should execute the transaction.

2. Bargaining and Negotiation Costs

Bargaining and Negotiation costs are the cost of determining what is being sold and for how much.

3. Policing and Enforcement Costs [2]

Policing and Enforcement Costs are the costs of ensuring transactions are prosecuted fairly and favorably for both parties.

A marketplace need not mitigate costs from all three dimensions, but we cannot imagine a marketplace that does not mitigate costs along at least one. For example, Craigslist (the web’s original marketplace) mitigates Search and Information Costs by serving as a central repository for buyers and sellers to connect. (In this Search and Information Costs dimension, they provide rather limited infrastructure, as Craigslist does not provide any information beyond unverified information posted by owners). Craigslist does not mitigate costs along the other two dimensions, as they provide no infrastructure to minimize bargaining and negotiation costs or policing and enforcement costs.

Compare that with Uber, a mobile transportation marketplace that provides a reduction in search and information costs (as Uber automatically makes matches between riders on the demand side and drivers on the supply side), automatically set fares (bringing bargaining and negotiation costs to zero), and provides significant liability insurance to drivers as well as reimbursements to riders if drivers take inefficient routes (both of these measures being ones minimizing policing and enforcement costs).

History of online marketplaces -

In ancient Greece, the agora marketplace was the center of athletic, artistic, spiritual and political life, and this spatial and cultural centrality enabled this marketplace to reduce transaction costs. New forms of media have enabled virtual marketplaces, which connected buyers and sellers independent of physical location. Newspaper classified ads were one of the first of such virtual marketplaces and accounted for sizable fraction (~40% in 2000) of newspaper ad industry revenue [3]. Rapid internet growth in the mid-1990s (2,300 annual growth rate in 1994 [4]) established, for the first time, a global network of producers and consumers that could, in principle, be connected by new marketplaces.

Figure 1: Rapid growth in the size of the internet creates a new virtual marketplace.

E-commerce first emerged in response to the demand aggregation afforded by internet connectivity. In 1994, Americans bought $19 billion worth of books, Barnes & Noble and the Borders Group had by then captured a quarter of the market, and profit margins were notoriously thin [8]. Yet, Jeff Bezos recognized that there are more items in the book category than there are items in any other, resulting in a “long tail” of SKUs that are necessarily left out of the ~130,000 titles carried within the average Barnes & Noble. The rise of Amazon has shown that that the long tail can be profitable: a quarter of Amazonʼs book sales come from outside its top 130,000 titles [9]. This suggested that online marketplaces overcome the tyranny of physical space, enabling aggregation of demand across dispersed audiences and a centralized supply chain for fulfillment.

Around the same time as Amazon was started, online peer-to-peer marketplaces emerged. After his wife wanted a way to trade Pez dispensers, Pierre Omidyar recognized that the internet could break down and remove the physical limitations of traditional auctions such as geography, presence, time, space, and a small target audience. Based upon this potential to reduce transaction costs, eBay was born and grew at a staggering rate, from ~ 0.25M users in 1997 to ~ 10M users and $2.8 billion annual revenue in 1999 [5]. In an iconic venture deal, Benchmark Capital invested $6.7M in 1997 (22% ownership), a stake worth $5 billion by April 1999 [6]. Network effects allowed eBay to get stronger as it got bigger, a unique feature of internet technology marketplace businesses vis-a-vis old economy business [7].

Figure 2: Network effects allows eBay to get strong as it got bigger.

Over the past two decades since the inception of eBay, growth in internet connectivity and online marketplaces have dramatically affected revenuers captured by newspapers. By 2000, newspaper classified advertising was $4.6 billion, dropping ~77% in a decade [3] and demonstrating the speed with which marketplaces can evolve.

Technology driving new marketplaces -

As mentioned, by altering transaction costs, IT has opened up the feasibility of new marketplace models. But how, exactly? As Venture Capitalist Fred Wilson explains, historically organizations—whether they be corporations, militaries, or governments—have been organized as top-down bureaucracies. This has been the case because of the prohibitive magnitude of communication and transactional costs. Though slow-moving, big bureaucracies can aggregate transactions and communications such that these per-transaction and per-communication costs are minimized. The web, however, has drastically reduced such transaction costs, and thus these sorts of top-down bureaucracies are being replaced by distributed networks. Wilson provides as an example Twitter, which as a distributed network of content producers is replacing the newspaper newsroom, which has typically been organized in a slow-moving hierarchical structure (writers/reporters reporting to editors, editor reporting to editor in chief, and editor in chief reporting to publisher, etc).

In commerce generally, one can apply Wilson’s framework to the concept of the traditional retailer, which serves as an intermediary between consumers and suppliers. Rather than individual agents from the supply side and demand side connecting on an individual transaction basis, to minimize the transactional costs, retailers purchase in limited ordering cycles; retailers place orders to suppliers on behalf of the consumer on a quarterly or yearly basis, rather than every time an individual consumers seeks out an individual supplier. Marketplaces, which facilitate such unitary transactions between single supplier and single buyer, become feasible when these costs are sufficiently low.

Indeed, the internet, which at its core is fundamentally a communication medium, has greatly diminished these transaction costs. Mobile extends the scope of the internet, allowing people to be constantly online, and further diminishing these costs. There are around 1.6B PCs in use today and an equal number of smart phones. Over the next few years, the majority of the mobile base will convert to smart phones, resulting in as many as ~8B by 2020 [12]. In parallel, the internet of things may grow to ~ 26B devices by 2020, collectively meaning that the scope of the internet - both the number of people and devices connected - will grow by an order of magnitude in the coming years. When everyone and everything is digitally connected, the communication costs required to prosecute a prospective transaction between multiple parties is greatly reduced. This has opened the opportunity for new marketplace businesses.

In parallel with the emergent marketplace opportunity, the cost to build a tech business more generally has fallen, partially through the considerable investment made by Web 1.0 firms. Amazon now offers primitive infrastructure building blocks as services, freeing developers from buying and maintaining costly infrastructure. The Amazon open APIs / IaaS drive more network traffic than its websites [10] and serve as foundational technology for the next generation of internet businesses. Companies can now be far smaller, leading to flatter organizations. As a testament to this trend, WhatsApp managed a messaging system as large as global SMS with only 32 engineers [11].

Types of new online marketplace -

In addition to the framework of transaction costs provided by Johari, we find it useful to further establish our marketplace taxonomy. Consider the following taxonomic distinctions:

Horizontal vs Vertical

Horizontal vs Vertical is perhaps the most intuitive characterization of emerging marketplaces: a Horizontal marketplace serves multiple industries, whereas a vertical marketplace serves only a particular industry. For instance, eBay and Craigslist are the prototypical horizontal marketplaces of the internet. As a horizontal marketplace for goods, eBay facilitates transactions for automotive, fashion, electronics, and collectible goods; Caigslist is even “more horizontal”, because it facilitates transactions for both goods and services (agnostic of industry). Vertical marketplaces [14] focus on a specific industry, such as Uber with transportation or Airbnb for short-term travel lodging.

Supply-Side-Driven vs Demand-Side-Driven

The distinction between a supply-side-driven vs demand-side-driven marketplace can be subtle; in fact, the terminology may seem somewhat counterintuitive as marketplaces by definition are two-sided. Some marketplaces nonetheless cater to minimizing costs of one side of the marketplace more than another. As such, some can be said to be supply-side-driven whereas other could be said to be demand-side-driven. Consider the following two pairs of marketplaces: ‘oDesk’ and ‘fiver’ versus ‘Hatch.co’ and ‘Custommade’. These pairs are horizontal marketplaces in the professional services outsourcing space and the custom goods space, respectively. However, each pair contains one marketplace that is supply-side-driven and one that is demand-side-driven.

Odesk and Fiverr are marketplaces for outsourcing professional services on a contract basis. Odesk is a demand-side-driven marketplace whereas Fiverr is supply-side-driven. On Fiverr, professionals post the services they are capable of performing and the prices they are willing to perform them for, and those on the demand side sort through offerings for what they want and send in offers. On Odesk, those on the demand side seeking contractual labor post the tasks they want performed, and those on the supply side sort through postings and send in offers. Fiverr, thus, is supply-side-driven, as search and information costs are being mitigated for the supply side more than the demand side. The inverse is true for Odesk, where those on the supply side must spend time and effort to find job requests they could perform. A similar dynamic is evident in the comparison between Hatch relative to Custommade, marketplaces for custom goods. On hatch.co, users can search through goods that supply-side craftsmen are offering to customize for individuals; on Custommade, the demand side dictates the jobs, as craftsmen search through jobs they would like to perform and go through the effort of submitting a proposal to the demand-side agent. Again, the search and information costs are being disproportionately mitigated for different parties for these two different models.

Figure 3: Supply-side driven (fiverr, on left) versus demand-side driven (oDesk, on right).

Full-Stack vs Thin-Slice

Full-stack online marketplaces have historically built supply chain infrastructure to achieve economies of scale. Comparing Webvan to Instacart, a recent marketplace for local grocery store delivery, highlights the distinction. Instacart leverages a “virtual supply chain,” viewing local grocery stores as distributed inventory and building a “thin” software and logistics layer on top of this to enable fulfillment. This was far harder to do before the advent of smart phone, as Instacart must manage a distributed local workforce of drivers and shoppers. This model may have benefits relative to the centralized full-stack marketplace (e.g., lower cost to scale and giving consumers local brands) [15].

Features of successful online marketplace -

Successful online marketplaces reduce at least one of the following costs.

1. Search and Information Costs (“meeting”)

Definition: Search cost is the cost of prospective buyers and sellers finding each other. Information cost is the cost of surfacing information relevant to either party making an informed decision as to whether they should execute the transaction.

Challenges: Cold-start, liquidity.

Solution: Marketplaces with network effects more quickly overcome these.

Simply stated, network effects are present when the value delivered to users scales with the network size. OpenTable, a restaurant-booking marketplace, illustrates this dynamic. Started in 1998, OpenTable now processes ~99% of all online reservations in North America [16] OpenTable is aggregating the world's restaurant goers and cataloguing their tastes. This aggregation of foodies attracts more restaurants, which in turn attracts more consumers. More consumers means that OpenTable has more data, making it more valuable both to consumers (recommendations) and restaurants. This “flywheel” drives favorable economics, as fixed costs can be shared over a larger and larger customer base (their margins went from 37% to 47% between 2010 and 2013) [17].

2. Bargaining and Negotiation Costs (“transacting”)

Definition: Bargaining and Negotiation costs are the cost of determining out what is being sold and for how much.

Challenge: For goods or services with poly-defined value, it is difficult as a marketplace to reduce negotiation costs.

Solution: One way to address this is to target higher fungibility goods or services that more easily admit to pricing automation.

Fungibility is the degree to which a good or service is interchangeable with another such good or service. An example of a highly fungible good would be a piece of mass-produced electronics, such as an xbox 360 video game console. There are 78 million such consoles worldwide, and purchasing one of these consoles new is as good as purchasing another. A highly fungible service would be something like retained automobile transportation—as long as a rider is delivered from point A to point B, one ride from a driver is as good as another.

Higher fungibility means that more suppliers are capable of performing a given service (e.g., drivers on Lyft), relative to cases in which each individual supplier is unique and more difficult to evaluate (e.g., designers on oDesk). If goods or services are not fungible, it is more difficult to automate transaction infrastructure (reduce bargaining and negotiation costs) because by definition comparables do not exist. When a fundamentally unique item or service must be priced for the first time, negotiation and bargaining costs are inevitable. Contrarily, when comparables do exist, goods and services can be priced according to precedent according to a limited set of known variables. In the case of electronic goods, these variables may be make, year, condition; in the case of a Lyft ride, these variables could be time of day, length of trip, location.

3. Policing and Enforcement Costs (“fulfilling”)

Policing and Enforcement Costs are the costs of ensuring transactions are prosecuted fairly and favorably for both parties.

Challenges that this category often address is dis-intermediation.

Solution: Low likelihood of recurrent transaction between same supply side and demand side agent.

When you enter an Uber or meet your AirBnB host, what keeps users from turning off the meter or transacting offline? Typically this is because the interaction is high frequency with low likelihood of recurrence of the same service provider. Bill Gurley explains this best: “failed marketplaces attack purchasing cycles that are simply way too infrequent, which makes it much more difficult to build brand awareness and word-of-mouth customer growth. Another repeated mistake is attacking verticals where a satisfactory supplier “match” end’s the customer’s need to re-enter the market in search of an alternative” [51]. Unpacking this further, if trust is easy to build offline (e.g., through a repeat interaction) then the marketplace is more easily dis-intermediated by offline transactions once trust is established. In contrast, the value of marketplace-enabled security is higher if the user must engage frequently with unique service providers (e.g., ride-sharing).

Business features of successful online marketplace -

Clearly, functional online marketplaces must address at least one of the above-mentioned costs. Yet, that does not make it a successful business! Other considerations are relevant when evaluating the potential of a marketplace business.

Transaction volume and defensibility

Uber’s transaction volume is high in at least two important ways: ride sharing is a repeat-use service with high-ticket sales. Lower repeat-use services (e.g., real-time car washing, which was tried by a now defunct company called Cherry) suffer from the fact that consumers rarely require the service that the marketplace offers. Other failed marketplaces, such as Kozmo, suffered from (among other things) transaction of low-value items. A similarly interesting case study in this issue is education marketplaces, such as Teachers Pay Teachers (TPT). TPT currently has ~ 10% of US teachers and had total transactions ~$50M per year (in 2013) with revenue then a fraction of this. Yet, the sale item (lesson plans) may not be a repeat-use item, which would mean that recurring sales at the 2013 level are harder to obtain without expanding the size of the customer base.