Companies that build online marketplaces struggle to solve the dreaded chicken and egg problem—how do you get buyers without sellers and sellers without buyers onto the platform? Many of today’s familiar online marketplaces—like Uber, AirBnB, eBay, Xbox Live, PayPal, OpenTable, and Angie’s List—managed to crack this challenge and reach sustainable scales.  In addition to creating useful products for their users, most analysts would agree that these businesses also benefit greatly from positive network effects between their buyers and sellers. That is to say, each additional buyer or seller on the platform generally has a positive effect on his or her counterparty:

  • More sellers à more options for quality, price, and convenience for buyers
  • More buyers à more opportunities to transact for sellers

Many people are ready to give these companies the benefit of having solved for positive network effects on their platform. In fact, I too fell into this trap until Professor Gautam Ahuja of Ross School of Business (visiting at HBS last year) made me re-examine this belief. Rather, there exists an opportunity to more deeply understand these companies by exploring the network effects across all party interactions. We can gain a better idea of the costs involved in growing the business and what valuable follow-on products should be developed. The basis for this company analysis begins by understanding that multi-sided platforms must inevitably also have multiple network effects. To clarify what this means, it might be easier to first understand what network effects look like on a single-sided platform.

A Single-Sided Network Example: Early Facebook

When Facebook first came on the scene in the mid-2000s, it resembled a single-sided platform. In the time before marketers were peppering the site with ads—weren’t those the days!—users basically joined other users in a system where all users were alike. We can think about Facebook’s network effects by looking at what happened when a single new user joined (an increment of +1 new users).

The incremental user’s content provided other users with more to see and explore. This content came in the form of posts, photos, and music preferences and meshed seamlessly into other existing users’ experiences. An existing user now has more to explore and interact with on the site—like seeing pictures of the new user’s trip to Bali or discovering linked articles the new user shared. Existing users’ content might also get more interactions from the new user in the forms of more likes, comments, and tagged photos. These could all be described as positive effects as they generally enhanced the value for Facebook users.

However, the addition of a new user could make the Facebook experience turn negative as well. The new user might constantly post pictures of their cat and make it difficult for existing users to find the content they really wanted to see in their newsfeed. The new user could also be someone existing users didn’t want on the site, like a coworker or boss, which would make them less inclined to share photos or other content. Quite simply, the network effects induced by one new Facebook user could be positive or negative.

The insight into Facebook’s network effects helps us understand a lot about what followed in the company’s history. Because the effects were predominantly positive, people invited their friends and family to the site virally, reducing expensive marketing and growth costs. The site exploded and reached 1.49 billion monthly active users as of June 2015 and saw 1 billion different users log in on the sam day in August 2015. Facebook subsequently developed products like a filterable newsfeed (no more cat pictures from the new user) and privacy tools to reduce users’ pain points (only share photos with friends or make posts visible to friends “except XYZ” people) that resulted from its rapid growth and negative network effects. These developments could have been predicted by our simple analysis of identifying the network and its positive and negative attributes.

A Multi-Sided Network Example: AirBnB

When thinking about multi-sided networks, the model for analyzing network effects grows more complex. Rather than thinking about how an incremental user will affect the entire network, we should scrutinize who the incremental user is and what network is being affected.

To illustrate this more clearly, let’s consider a two-sided platform like AirBnB. When a new user joins the AirBnB site, we should first consider whether the user is a guest or host. Next, we need to explore how that additional guest or host affects other guests and hosts. For this two-sided network example, there exist 4 possible network effect scenarios:

  • How does an incremental guest affect hosts?
  • How does an incremental guest affect other guests?
  • How does an incremental host affect guests?
  • How does an incremental host affect other hosts?

Understanding network effects can quickly get complicated when dealing with multi-sided platform. The number of unique network effects necessary to consider is n2 for each distinct n type of parties on the platform. As we saw with Facebook, network effects can be either positive or negative, complicating our understanding of two-sided platforms even more. When considering positive or negative effects, the interactions that should be examined are 2n2.

Below I’ve illustrated how AirBnB might experience positive and negative network effects across all 4 of its network change scenarios:

AirBnB Network Effects Diagram

 

An Updated Model for Thinking About Network Effects: Uber

The AirBnB analysis is a useful starting point, but I find it easier to simplify each of the distinct networks into a more manageable characterization. This reduction allows us to quickly understand the dynamics in the networks of a company while maintaining an explainable simplicity. I therefore classify interactions into one of three buckets.

  • Collaborators (Positive)—Parties predominantly enhance the experience of other parties in the network. Examples of collaborators include funders on Kickstarter who together to support an idea or product or gamers on xBox live who play together in multiplayer Halo battles.  Such relationships encourage users to invite other users to a site, and can lead to organic site growth and lower user acquisition costs.
  • Counterparties (Positive)—Parties are involved predominantly in a monetary transaction or exchange that satisfies both sides. Examples buyers and sellers transacting for a deal on Groupon or a mother ordering food via delivery service Sprig. Parties exchange clearly identifiable goods and services, which, when priced at a point such that the transaction clears, creates value for both parties that supports repeat usage and high user lifetime value from multiple transactions.
  • Competitors (Negative)—Parties predominantly compete for resources or opportunities. Examples include applicants applying to Y Combinator where only a select number of applicants are accepted or eBay bidders competing against each other for a baseball autographed by Mickey Mantle. In both cases, competition is likely to give users a worse user experience as they might not secure the opportunity or good or end up paying a higher price. This can result in a lowered user experience, unsubscribing, and high sign-up or reactivation expenses.

By using this framework, it helps me understand the operational costs a company is likely facing and what products they might consider developing in the future. For example, if this analysis were applied to ride-sharing start-up Uber, it might look like this:

Uber Network Effect Diagram

For Uber’s two-sided platform, a large part of the company’s value comes from solving the most obvious network dynamic: matching drivers with riders and riders with drivers. The company was able gain a foothold in markets like San Francisco because cab companies were not keeping pace to satisfy this counterparty need. As more drivers joined Uber, riders benefitted with greater ride availability and more riding options (uberPOOL, uberX, uberXL, uberTAXI, UberBLACK, uberSUV, uberSELECT, uberPOP, uberBIKE, etc.). As more riders sign up, drivers are more likely to match with a pick-up request and earn money for their services. These services led to increased usage by both riders and drivers as value was realized.

However, this dynamic doesn’t necessarily lend itself to growth. Drivers aren’t actively inviting or converting new riders, and new riders aren’t energetically recruiting new drivers. The first time they usually encounter each other is during an Uber ride, at which point they’re both already on Uber’s platform. While a positive experience for each party—a clean, convenient ride for the passenger and a profitable transaction for the driver—will influence who how often the other party uses Uber in the future, they’re not actively growing the platform.

The story gets more interesting when you look at the network dynamic across the rider <-> rider and driver <-> driver dimensions. For current riders, each additional rider chiefly means increased competition for resources. In Uber’s case, this manifests itself in surge pricing when many people try and use the app at the same time—such as during a rainstorm or Friday rush hour. The experience is painful, and users are upset by either the wait time to find a rider or the total cost for the trip. Similarly, as more drivers join the platform, existing drivers face both increased competition for riders and reduced chances for earning surge prices. If driver’s aren’t able to find riders and drive around unoccupied, this cuts into the driver’s ability to earn for time worked.

By understanding the interplays occurring across its networks, it’s easier to identify and appreciate Uber’s growth, marketing, and development challenges over the past few years. While Uber has benefitted from word-of-mouth marketing for its remarkable service, much of Uber’s recent growth depends on promotions and discounts rather than virality. Because riders and drivers aren’t actively working to sign up other drivers and riders without an incentive, Uber bears the burden of these growth costs itself.

For example, Uber offers money to users (riders and drivers) who sign up new riders and use a unique promotion code (mine is below—feel free to join using it!). Additionally, Uber will give new riders one or more free rides upon joining. To sign up new drivers, Uber offers drivers a $500 bonus after completing their 20th rider, $500 dollars for signing up a Lyft driver, and fixed hourly income guarantees to ensure new drivers realize monetary gain immediately—with promotions often varying city by city. These acquisition costs can be large for a company looking to scale globally and help explain why Uber has raised massive amounts of cash to grow operations in areas like China, India, and other parts of Asia. In addition to building and operations costs, a lot of that money will likely go to signing up riders and drivers through aggressive promotions and discounts and launching citywide marketing campaigns. Given Uber rider’s lifetime value from its positive counterparty interactions, such costs can likely be easily justified.

Uber Promotion Code

On the product development side of the business, Uber’s network effects can explain a lot of what the company has focused on building. Paramount to the experience is maintaining a positive rider <-> driver and driver <-> dynamic. Anything that facilitates a quality service has taken precedent in the pipeline to protect the company’s advantages and keep users using the app. Such developments include credit card scanning for easier payment, license plate information to help riders identify drivers, written explanations for 3 star or below reviews to protect drivers’ reputations, and optimal route maps to make sure the most cost-effective route is taken.

Uber has worked to tackle negative network effects inherent to its business as well. Uber launched a fare split feature that aims to make the riding experience more collaborative and less competitive, easily allowing users pay each other and receive ride receipts. Additionally, a feature was added to surge that allows users to be notified once surge has dropped below as certain level, decreasing the pain from increased competition over resources. Finally, uberPOOL matches different rider pairs so that a each group receives a lower fare for carpooling with the other party. These features all subtly aim to shift the experience from competitive to collaborative.

Looking Ahead

In this post, I hope that I’ve helped lay out a new model to help understand network effects for multi-sided online marketplaces. By identifying all the distinct networks that exist and then understanding the interplay of people in those networks (collaborators, counterparties, competitors), one can develop a valuable tool for understanding much about an online business. This insight can be used to analyze a company’s growth and marketing costs—will users sign other users up or do they need to deploy resources? It can also give vision into a business’s product development priorities—what networks need to be protected with better products and which networks need pain points addressed?

 By: Ry Sullivan


read more

Over the past few years, we have seen an emergence of peer to peer “rental” sites that allow people to rent out their belongings to others looking for short term use of that asset.  Users can log on to relayrides.com to share their car, airbnb.com to share their house, or zilok.com to rent just about anything else.  Additionally, we have seen an exponential increase in the success of multi-sided e-commerce websites that connect sellers and buyers from around the world to facilitate global commerce.  Alibaba.com, Tradekey.com, and globalsources.com are a few of the sites that are attracting hundreds of thousands of users and million dollar valuations.

The single biggest threat to these business models, as I see it, is a loss of trust between the two parties.  The fundamental issue here is the balance of power – the perception that one side has a lot to lose while the other side has little risk in the transaction.  To explain:  In the peer to peer sharing sites, the lender risks renting to an irresponsible renter who trashes his or her valuable property; for the e-commerce site, the buyer risks paying for an item only to be delivered an inferior product, or, even worse, delivered nothing at all.

To further illustrate this point, two examples have appeared in the media recently.  Their potential to undermine the business models serves as a warning to this fledging industry.

Airbnb came under attack this summer when a renter nicknamed EJ lent her apartment to a vacationer, and returned to find it vandalized and ransacked.  EJ described the experience in a blog post that went viral, garnering the attention of Techcrunch, USAToday, and CNN.  She writes, “[the renter] and friends had more than enough time to search through literally everything inside, to rifle through every document, every photo, every drawer, every storage container and every piece of clothing I own, essentially turning my world inside out, and leaving a disgusting mess behind.”

Airbnb isn’t the only site where a breach of trust occurred between its two parties:  in February of this year, Chinese police arrested 36 people accused of fraudulent practices on Alibaba.com.  These “business people” are accused of scamming buyers out of an estimated $6 million by taking payments for items that they never actually delivered.  The event was made even more scandalous by the fact that these sellers were granted “Gold” supplier status by Alibaba employees who were allegedly aware of the scam.

Despite the verification systems that were in place in each of these cases, people still got burned.  Such examples bring the reputation of the world-wide, multi-sided platform business model into question.

So, what needs to be done to ingrain the integrity of the business model?  Below, I offer some advice on how to build trust between the parties:

1.    Take responsibility:  It’s not enough anymore to simply build a site that facilitates transaction.  Users expect more.  Perhaps expectations have been set by industry trailblazers like eBay, which acts in a no nonsense manner when dealing with questionable or suspect transactions.  Both buyers and sellers risk removal from the site if practices are called into question.  Multi-sided platforms need to protect both sides from the risks inherent in the transaction, through tools like insurance policies, escrow accounts, and post-transactional feedback tools.

2.    Find out what’s broke and fix it, immediately:When Alibaba discovered that its own sales staff had been involved with some or all of the 2,300 cases of fraud over the last two years, leaders were held accountable.  The company’s CEO and COO removed themselves from the organization after the fraud was uncovered to take responsibility for the “systemic breakdown.”  This, combined with their public admission of guilt, has gone a long way in keeping down bad press and building user confidence.

3.    Be proactive:  Rather than waiting for an unfortunate event to occur before acting, anticipate the risks and protect your users.  Relayrides, for example, holds a $1 million supplemental insurance policy for its car renters and installs an immobilizer on the vehicle to prevent cars from being started without a reservation.  It is necessary to implement stringent requirements for both sides of the platform – be it mandatory compliance to a legally binding Code of Conduct/Ethics or compulsory screening of rental properties and manufacturing sites.

In summary, there is value created for both sides using these platforms.  If a platform wants to remain a viable business, however, they must invest in security measures to protect their customers and be prepared to take responsibility when something does occur.  Building trust, taking responsibility and underwriting a product or service are just good and basic business practices.


read more

Multi-sided platforms (also known as two-sided networks), bring together two distinct groups who each benefit from the other’s presence on the platform (e.g. Monster.com is a classic two-sided networ

k where employers post jobs on the site because of the breadth and depth of the jobseeker base and jobseekers come to the site because of the quantity and quality of jobs on the site). Much of the existing literature on multi-sided platforms has focused on possible solutions to the inherent chicken-and-egg/Catch-22 condition of the model. The thinking goes that prospective users on each side of the platform will be reluctant to join without the presence of users on the other side. To skirt this issue, most solutions focus on one of four approaches:

  1. Subsidize one side of the platform to motivate those users to join.
  2. Establish exclusive relationships with key users on one side of the platform to lure users on the other side.
  3. Start off as a vendor to one side of the network and after building a loyal base of customers, begin facilitating transactions between those customers and other suppliers.
  4. Start off as a merchant acquiring inventory from suppliers and then resell that inventory to customers on the other side of the network before transitioning to a non-risk taking intermediary.

From my perspective, these strategies seem to be optimal for startups with completely innovative business models or those that reject the precepts of the lean startup methodology (http://www.startuplessonslearned.com/2009/08/minimum-viable-product-guide.html). To the extent that no content already exists on the web from either side of the platform, I believe the aforementioned would be the best place to start. However, given the vast availability of content on the internet, there may be an approach to starting a multi-sided platform that would be far more consistent with lean startup methodology, harvesting.

Harvesting is the collection and processing of information already stored elsewhere. RentJungle (apartment aggregator), Kayak (flight aggregator), theLadders (job aggregator), Yipit (daily deals aggregator) are all multi-sided platforms that harvest content already on the web to solve the inherent chicken-and-egg problem of the two-sided network without subsidization, exclusivity, or starting off as a vendor or merchant to one set of users. Since both sides of each of the platforms above already transacted over the internet in a fragmented way (two key requisites), the simplest approach for serving as an intermediary was to aggregate the data from all the “sellers” (content providers) and serve as a one-stop shop for all of the “buyers” (users). In this way, each of these companies was able to avoid the Catch-22 by eliminating the need for one side of the platform to join the site in the first place.

At this point it is worth taking a step back to think about how harvesting fits in with the context of the lean startup methodology. Eric Ries, whose recently published book The Lean Startup: How Today’s Entrepreneurs Use Continuous Innovation to Create Radically Successful Businesses has garnered much acclaim, writes that one of the most important techniques for a lean startup is the “minimum viable product”. Ries states that “the minimum viable product is that version of a new product which allows a team to collect the maximum amount of validated learning about customers with the least effort.” The four traditional approaches to solving the chicken-and-egg issue of multi-sided platforms all most commonly require substantial effort and/or resources and do not lend themselves to rapid experimentation and testing. In contrast, harvesting is an extremely fast and easy way to acquire content from one side of the platform (as outlined below) and the harvested data can be pruned for different users to test various iterations of the platform.

All that said, how easy is it to harvest? In the absence of an API or direct-feed into one side of the platform, the lean startup can develop a simple scraper to acquire data from one side of the platform which then can be used to attract users on the other side. That said, if an API does exist, it might not even be necessary to build a scraper as the data may be readily available from a content provider (Amazon, Google, CNET, Facebook, Twitter and even the World Bank all offer APIs). In either case/whenever possible, I believe entrepreneurs adhering to the lean startup methodology should first attempt to harvest one side of the multi-sided platform in order to eliminate the chicken-and-egg problem before moving on to one of the more traditional approaches for vetting the viability of a multi-sided business model.

 


read more