The Network Matrix: Bridges & Identity
Why network bridges and unique user identity are critical to building successful network-based startups
As I explained in my last post, the “marketplace matrix” is a great framework to get a quick understanding of the strengths and weaknesses of marketplace startups. Combining both defensibility and scalability of unit economics gives us a more holistic, but not yet comprehensive, view of these businesses. Interestingly, the core tenets of this framework are not restricted to marketplaces and can be extended to other types of network-based businesses as well (1-sided, 2-sided or multi-sided networks). But moving from the subset of marketplaces to the superset of networks requires the introduction of new concepts. I’ll explain these by looking at some well-known names in the social media space, and then extend these ideas to other network-based business in far-flung spaces, from payments to online gaming.
Scalability & Network Bridges
Hyperlocal Networks (Unbridged)
As we saw when we studied Uber, dating apps like Tinder and neighbourhood social networks like Nextdoor are hyperlocal, i.e. the addition of a new user makes the network more valuable for all users within a small geographic area (Tinder for travellers is an edge case). For example, when Tinder acquires a female user in West London looking for male matches, it makes it easier to organically attract (no pun intended) users with corresponding preferences in West London. Tinder requires new user acquisition investments when expanding to another geography, creating discrete network clusters with very few links or bridges between them. This makes their unit economics less scalable as they cannot leverage existing users to organically attract new users in a new geography (or at least not at scale).
Cross-Border Networks (Bridged)
Some networks (both social and content-based) exhibit cross-border network effects, like those seen in marketplaces like Airbnb. For example, the addition of a creator within a specific niche on YouTube or Twitter (e.g. a technology influencer), makes them valuable for all consumers interested in that niche across borders. Even when niches are localized (e.g. Brexit influencers in the UK), network effects extend across the region and are not structurally restricted to a small geographic radius. True social networks like Facebook that are built on users’ real-world social graphs also exhibit cross-border network effects, but with a “bridged” network structure.
The addition of a Facebook user in the United States does not automatically make the social network more valuable to a user in India. This is because social circles are concentrated by geographic location and people are far more likely to know others around them. However, some people also tend to travel and relocate, resulting in more distant connections. This creates a structure of overlapping social clusters, i.e. most of the people we know are in the same location as us, but some live further away. Users with dispersed connections act as “network bridges” and can help the network organically expand from one region to another.
This was a key reason why Facebook scaled so rapidly despite starting in a close-knit and co-located community, i.e. Harvard University. Harvard students come from all over the world and have pre-existing connections in their hometowns in the U.S. or home countries overseas. Gaining traction in this community made it easier for Facebook to organically attract new users when it expanded to other universities, cities and countries. The presence of “network bridges” is critical to scaling across geographic boundaries and this is a key trait I look for in new network-based startups.
Defensibility & Identity
Differentiated Networks (Identity Focused)
I have previously explained how the defensibility of a marketplace is a function of supply differentiation. Marketplaces like Airbnb, where individual units of supply are differentiated across a range of different attributes, tend to be far more defensible than those like Uber, where each unit of supply is interchangeable. This concept can be applied directly to “content networks” like YouTube, as supply diversity is analogous to content diversity (from sports analysis to repair guides). But what about social networks? What makes individual users differentiated on networks like Facebook? Their unique identity.
“Mainstream” social networks like Facebook, Instagram, Snapchat, etc. are built on connecting unique users, i.e. each user’s unique identity is core to the network. Users want to connect and interact with specific individuals who are not interchangeable. This is a very simple but exceptionally important concept. If I want to add Person A to my network, Facebook cannot swap in Person B as a substitute. This would defeat the entire purpose of the network. So when social networks amass a user’s entire social graph or a critical sub-section of it (or an interest-based social graph on follow-oriented networks Twitter), it becomes very difficult to abandon them.
Snapchat’s redesign fiasco is a great example here. In late-2017, Snapchat removed its much-loved Stories page, moving user stories into the Chat page and publisher stories into the Discover page. This resulted in intense and negative feedback from loyal users. The company’s stock price even declined by nearly 20% when Snapchat admitted that the redesign required tweaks. However, Snapchat’s daily active users only dropped by 3% over a few quarters before resuming growth. This happened despite intense competition for users from Instagram. Snapchat thrived because it was a critical channel for its users to interact with friends they had added to the app. This shows how important user identity is to building defensibility — another important attribute I look for in network-based startups.
Commoditized Networks (Identity Agnostic)
Other networks make user identity largely immaterial, and fall closer to the “commoditized” end of the spectrum. Anonymous social networks like Whisper, Sarahah and Yik Yak are extreme examples here. These networks allowed users to either join various anonymous interest groups or receive anonymized communication from friends (added via other social networks). They completely stripped out user identity in order to promote more “honest” communication. But once these networks gained a critical mass of creators and/or content, the addition of new users did not make them more valuable to other users. So the value of the network largely peaked once they had sufficient “anonymous content”. This removed any incentive for users to stick around once the novelty of anonymous content wore off, especially when the downsides of anonymous communication became clear. As a result, the popularity of these networks was very short-lived, ranging from a few months to a couple of years depending on the amount of venture capital funding they raised.
Short video apps like Vine and TikTok (previously, Musical.ly) are also examples of commoditized networks. Unlike anonymous social networks, short video apps do not obscure the identity of content creators. However, these apps have to rely on an algorithmic approach to discovery to maintain engagement because of limitations in the duration and variety of content. Notably, TikTok does not require users to create an account before they begin browsing videos and its algorithm recommends videos in an infinite feed based on user behaviour. This desensitizes users to the identity of creators, as content from creators they follow is combined with other similar videos in their feed. As a result, creators are commoditized and much like we saw in the case of anonymous social networks, short video apps peak in value once they gain a critical mass of content. The limited relevance of creator identity and content diversity also makes it easier for mainstream social networks (or competing short video apps) to poach creators and/or absorb their functionality. Consequently, “multi-tenanting” (users or creators using more than one service) is an even bigger risk than it was in the case of commoditized marketplaces like Uber. This results in declining engagement once the novelty of the content format wears off.
While the longevity of these commoditized networks tends to be longer than the extreme case of anonymous social networks, they still face the risk of gradual abandonment because of the dynamics described above. Vine’s history is a great example of this. The short video app went viral when it launched in 2012 and Twitter later acquired it to capitalize on its role in popular culture. However, user engagement began to decline and it was shut down 4 years later. Similarly, Musical.ly went viral in late-2015, but sold to TikTok just two years later when user interest began to wane. Meerkat’s brief time in the limelight is yet another example.
The Network Matrix: Scalability x Defensibility
Like we did with the marketplace matrix, we can now plot scalability and defensibility against each other to create a “network matrix”. As we can see below, this is applicable well beyond the narrow social media space and can apply to any startup connecting multiple participants. Again, this is by no means a comprehensive approach to analyse network-effect based startups. It does, however, give us a starting point to gauge strengths and weaknesses before doing a deeper dive.
Tier-1 networks include global networks or those with “network bridges” combined with a strong emphasis on user identity or differentiated content. These companies present the most attractive investment opportunities and, unsurprisingly, represent some of the most used networks across the globe. The pattern isn’t limited to social media, and includes publishing networks like Medium and payment networks like Transferwise and Paypal. It even includes B2B SaaS-based equity management tools like Carta that connect startups to their shareholders. This shows that the strongest network effects are by no means limited to social media companies or even the consumer sphere. To emphasize, startups that fall into this category are the highest priority for us as investors.
Tier-2 networks either have a fragmented, unbridged network structure combined with a strong focus on user identity or a bridged network structure combined with a weak emphasis on user identity. The top left quadrant covers local networks like dating apps and neighbourhood social networks. The bottom right quadrant includes some anonymous social networks, short video apps, live-streaming services, ad networks (where the identity of publishers isn’t critical) and online games (where the identity of players is only relevant in some cases).
Tier-3 networks combine both a fragmented network structure with a weak (or non-existent) focus on user identity. The lone example I can think of here is Yik Yak, the social network that allowed users to share anonymous posts with those in their immediate vicinity. Hyperlocal network fragmentation made it very challenging to scale the network sustainably and stripping out user identity made it very difficult to keep users interested, especially as it became a go-to channel for bullying. This is a recipe for creating the worst kind of leaky bucket, i.e. users leave faster than you can add new ones. As a result, Yik Yak went from raising over $60M at a $400M valuation, in a round led by Sequoia, to shutting down in just 2.5 years. Yik Yak’s history is among the strongest pieces of evidence to show the importance of user identity and network bridging.
Much like the marketplace matrix, the network matrix is a great framework to screen network-based startups. Of course, an in-depth evaluation requires the addition of more layers of complexity to this basic framework. The next layer of complexity to add is the impact of integrated software (or SaaS) on the strength of network effects. Watch this space for more…