Marketplaces with more than two sides

Marketplaces — peer to peer networks that facilitate the buying and selling of goods and services are increasingly central to software’s continued march into all aspects of our lives. Uber helps us get around and Airbnb helps us find places to stay. Interestingly, these services and most other prominent marketplaces feature only two sides — the buyer and the seller.

However, marketplaces do not have any intrinsic forces that limit the number of distinct sides to two [1]. In fact, marketplaces with greater than two sides have greater net utility than those with just two sides. These marketplaces can model real world firms which make use of multiple actors to create value for the customer. For example, Google has salespeople, engineers, designers and product managers, Deloitte has consultants, salespeople and engineers and Applebees has waiters and cooks. In each case multiple actors must work together to deliver value.

Thus, beyond the intrinsic additional value, certain problems currently solved by firms simply cannot be solved in a scalable way without additional primary actors in the marketplace model.

Now, let’s try to unpack why these kinds of marketplaces have more value☺

Marketplace Value

A Marketplace has value because of the connections that can be formed between buyer and seller. More generally, a network is valuable because of the connections that can be formed between nodes within the network.

Therefore the net utility that customers derive from a network is roughly the number of connections that can be formed between the customer and any other primary actors on the network. In the analyses below we assume perfect symmetry — that all actors on the network can perform the roles of other actors (e.g. all buyers are sellers and all sellers are buyers).

In a two sided marketplace of size N, each buyer can connect with ~ N sellers and thus derives a value of ~N from the marketplace. With ~ N buyers we get a net value of N*N. Thus, the net value of a perfectly connected two sided marketplace is ~ N^2. This number is essentially the number of pairs we can select from a network of ~N nodes.

What happens when we increase the number of sides?

For a marketplace with m sides, we simply count the number of groups of size m we can select from a total of N. Thus, the net value of a perfectly connected m sided marketplace is ~ N^m.

So for a 3 sided marketplace of size N, the value is ~N^3 i.e N times larger than the value in a 2 sided marketplace [2]. At scale, marketplaces comprise of millions of people and so in theory the utility of a 3 sided marketplace could be ~1M times that of a 2 sided market.

Before we go on to discuss why we should or should not explore 3 (or more) sided marketplaces, it’s interesting to consider what one might look like.

Three sided Uber

One strategy for getting to three sides is to split up one of the two sides. The two sided version of Uber is the rider and the driver. The three sided version could split up the driver and the car and we end up with rider, driver and car provider.

One obvious advantage of this version of Uber is that it expands the market to drivers who do not have cars. If we can pair people who can drive but have no cars (a large number) up with unused cars (also a large number) then we end up with a larger number of Uber drivers on the road and thus increase the value of the network to each customer.

We got here from an abstract and general strategy (i.e seek out 3 sides by unbundling one side) but interestingly there’s an actual startup (Breeze) started by ex Uber employees trying to play the role of car provider in our three sided scenario. GetAround also works on filling this need. The takeaway here perhaps is that Uber and GetAround should have a baby ☺

This supply side unbundling is relevant to other on-demand services like Instacart, SpoonRocket and Sprig. Within a service like Instacart we can also separate shoppers from car owners and end up with more shoppers.

Alternatively you could operate a service like Instacart or Sprig atop a third party network of cars and end up with more efficient operations that focus on the core value propositions related to grocery shopping and food preparation and selection respectively.

Caveats

If the intrinsic value in these “higher order” marketplaces is that much higher then why aren't there many visible examples? The reasons are:

1. Increased complexity & coordination costs — This one is fairly obvious. If it’s really hard to acquire and manage two sets of people in a marketplace then it has to be nearly impossible to do the same for more than two sets of people.
2. Users in a network do not typically interact with all other users in a network and so the theoretical utility gains are highly overstated. See [2] for details, but the gain over a two sided network turns out to be ~ logN and not N. Combined with #1, it’s possible that gains get greatly damped or even washed out if the marketplace is not carefully tuned.

Proceed Anyways

The issues of increased complexity and coordination costs can be mitigated via technology. A combination of better algorithms, more data and real time communication tools allows us do a lot.

Regardless of whether their costs can be brought down by technology, marketplaces with multiple primary actors are not merely interesting to speculate about. Many real world scenarios in the workforce naturally lend themselves to having multiple distinct nodes e.g retail (customer, manager, worker), music (producer, songwriter, singer, consumer), movies (producer, screenwriter, director, actor)and ridesharing (customer, driver, car provider)

As online networks and marketplaces continue to penetrate the world, these higher order networks will be needed and, accordingly, technology for managing coordination will be critical. Networks with the highest amounts of symmetry [3] will be the easiest to bootstrap.

Unbundling the supply side of existing two sided networks could also provide interesting first generation examples. However, these kinds of networks will likely have the largest impact on industries where buyers have highly complex requests and/or units of work necessitate collaboration between multiple parties. These sorts of work are the wheelhouses of firms and have so far resisted automation and marketplace approaches.

I am a hacker and physicist in San Francisco, who loves working on problems at the intersection of labor, networks and artificial intelligence— follow on Twitter.

FOOTNOTES

[1] While ecosystems like Android and Microsoft are multi-sided (developers, users, OEMs), this article refers mostly to peer to peer service marketplaces.

[2] One correction we can make to our calculations, is to take into account the 80–20 intuition that the majority of the interactions an individual will have in a network would typically occur with just a few of the nodes. Zipf’s law says that if we ordered the list of people we interact with on the network by how much we interacted with them, we would find that the second highest interaction would be half as much as the first and the third highest interaction would be a third as much etc. Thus for a two sided marketplace each customer gets a net value of (1 + 1/2 +1/3 +…..1/N) = ~ logN

The total value in such a network is N*logN instead of N^2. Applying the same correction to the three sided network gives a value of N*logN*logN

[3] All nodes can perform the functions of other nodes e.g. a buyer can be a seller and a seller can be a buyer on eBay.

Thanks to @jbreinlinger , @__aston__ , @ai and @freefrancisco for reading drafts of this