In the first part of my marketplace thesis, I explained how I think about the geographic scalability of marketplaces, i.e. I prefer marketplaces where a given unit of supply is accessible to customers across borders as opposed to those where it is restricted to a small local area. We saw that marketplaces with cross-border network effects (e.g. Airbnb), created a lot more value for investors than those with hyperlocal network effects (e.g. Uber). In sports terminology, this is a view of the “offensive” side of the game. But we also need to study the “defensive” side to gauge how well marketplaces can defend themselves from competition and copycats at scale.
Network effects are among the strongest sources of structural defensibility in today’s world. But as we saw when we compared cross-border and hyperlocal network effects, not all network effects are created equal. To evaluate defensibility, we need to add another layer of complexity and study the nature of supply itself, i.e. how the demand side views supply. I’ll build on a few ideas from some very smart people at a16z and NfX to explain this part of my thesis. And I’ll rely on another comparison between Uber and Airbnb, two of the most iconic and also easily understood, marketplaces of our time.
Uber’s suppliers are interchangeable (or commoditized), i.e. customers just want a ride and are not particularly sensitive to driver identity or vehicle brand. Uber only offers 2–3 vehicle types (in addition to UberPOOL) based on price, group size, luggage requirements, etc. and its rating system attempts to filter out drivers who don’t meet a baseline level of safety or cleanliness. As a result, riders gauge Uber’s value simply by “quality of service”, i.e. the amount of time it takes them to get a ride. In its early days, the addition of more drivers had a significant impact on value to the rider as it meaningfully brought wait times down from 10+ minutes to 4–5 minutes. But as wait times reduced to the 5 minute range, additional driver acquisition did not meaningfully affect rider experience. In other words, riders became indifferent to wait times below a certain threshold.
This meant that competitors did not need to truly match Uber’s supply density to steal riders. They simply needed wait times to be “good enough”. As a result, many ridesharing startups invested much of the capital they raised on driver incentives to poach Uber’s existing drivers and hit critical mass. In addition, commoditization of supply meant that riders were largely indifferent to which service they used. A ride on Uber is the same as a ride on Lyft or Bolt or Cabify, as long as the driver showed up within 5 minutes. And since drivers just needed a fare, they signed up with every active ridesharing marketplace. This behaviour is called “multi-tenanting” and is widely prevalent across marketplaces that have commoditized supply.
Extensive multi-tenanting on both demand and supply sides created a hypercompetitive environment even in Uber’s core markets. Lyft gained market share in the US, while Bolt, Kapten and Ola moved into the UK. Uber still has a much stronger brand than its competitors but that is a much weaker source of structural defensibility as compared to network effects.
Convoy faces a similar competitive dynamic because of commoditized supply. Convoy connects shippers to truckers, but shippers just care about moving their goods from point A to point B. A trucker’s attributes don’t matter beyond price, capacity and a baseline level of reliability. As a result, shippers are motivated to compare prices across multiple logistics marketplaces including Flexport and Uber Freight. Likewise, truckers are motivated to sign up with every other logistics platform to maximize capacity utilization. So despite the geographic scalability of these businesses, commoditized supply creates a very competitive environment leading to higher supplier acquisition/retention costs and lower pricing power.
Airbnb, on the other hand, has differentiated supply, i.e. each unit of supply is unique to some degree across a number of attributes: type of property, quality, nightly rate, location, capacity, etc. More choice adds more friction for guests which increases the responsibility of the marketplace to curate supply and enable better matches (and filter out scams, which can be challenging). However, this also meant that Airbnb needed to gain critical mass in many different niches as opposed to a single, standardized unit of supply. In other words, Airbnb needed enough hosts listing countryside cottages, small town B&Bs, centrally located apartments in major cities, beachside villas, etc. across every region they operated in and across a range of other attributes (price, capacity, etc.). Competitors could not just scale within one niche (e.g. apartments in a few cities) to poach guests as Airbnb offered far more choices. So as Airbnb scaled, it became nearly impossible for competitors to match their value proposition. As a result, Airbnb does not face anywhere close to the scale and volume of competition as Uber or Convoy.
Poshmark is another, even more extreme, example of differentiated supply. Poshmark is a C2C marketplace focused on clothing and accessories which are differentiated products. That seems simple enough, but Poshmark sets itself apart by enabling social network-like interactions between buyers and sellers. For example, it allows buyers to discover sellers based on event-based styling requests. It also hosts virtual shopping parties around a specific theme (e.g. a brand or category) to help buyers discover sellers with different interests. This allows sellers to establish a unique identity. As a result, Poshmark doesn’t just have differentiated supply, it also has differentiated suppliers. This makes it even more difficult for their competitors to provide a comparable experience.
The Marketplace Matrix
Now that we’ve covered both offensive and defensive capabilities of marketplaces, we can plot them against each other to create a marketplace matrix. While it is not comprehensive, this 2x2 provides more holistic framework for categorizing marketplace startups.
Tier-1 marketplaces, in the top-right quadrant, have highly scalable unit economics (thanks to cross-border network effects) and are highly defensible (thanks to differentiated supply). The marketplaces in this quadrant are likely to create the most value for every dollar in funding. For example, Airbnb’s valuation-to-funding multiple is 9–10x, while it is 13x for Numbrs and 8x for Poshmark. Therefore, it should come as no shock that tier-1 marketplaces are the highest priority for us as investors.
Tier-2 marketplaces are those that score low on either scalability or defensibility while scoring higher on the other. As a result, they have lower value creation potential than tier-1 marketplaces. For example, Convoy, Flexport and OfferUp all have a valuation-to-funding multiple between 3–5x.
Tier-3 marketplaces (commonly referred to as Uber for X) score low on both scalability and defensibility. This category mostly includes hyperlocal, service oriented (and delivery oriented) marketplaces like Uber, Deliveroo and Wag. These startups tend to create the least amount of value for every dollar invested. For the three examples mentioned here, the valuation-to-funding multiple consistently ranges between 1–3x. Anecdotally, this category of marketplaces also appears to have the highest failure rate with examples ranging from home cleaning startups like HomeJoy to on-demand laundry services like Washio. Their capital intensive nature combined with low defensibility means that they are only viable in a handful of product segments with very high frequency of use, e.g. ridesharing and food delivery. Of course, some of them are likely to be more viable without high risk venture funding.
The chart above shows the valuation-to-funding multiples (proxy for value creation) across a number of examples in all three marketplace tiers. Even though this list includes a mix of public and private companies, the pattern is unmistakeable. Tier-1 marketplaces have clear structural advantages over others.
Of course, this categorization is still incomplete as it does not take many other factors into account. Marginal costs, side switching, market fragmentation, frequency of use and “single player” software are all very important factors to consider. I will explore these topics in future posts and layer more complexity onto the basic framework described here.