A Framework for Network Effects in Web3

The terms ‘Web3’ and ‘Network Effects’ have now become even more frequently misused than ‘Machine Learning’ or ‘A.I.’. When these terms are haphazardly thrown together by signaling-addicted Twitter influencers, blood cortisol levels rise even further, heads pound tables that much harder, and out of frustration both terms are disregarded as buzzwords. Even among the so-called ‘crypto-native’ community, where there is strong consensus that web3 networks will become the dominant internet paradigm, seems to be an odd lack of precision in the discourse surrounding how network effects are actually generated by these networks.

In this essay we will attempt to stomach the noise, boil terms like ‘web3’ and ‘network effects’ into their simplest components, and then use these components to build models of how web3 networks unlock an entirely new species of network effects.

The contents of this essay are broken into three sections. We start in section one by establishing the primitives that comprise web3 networks, as well as the technical and economic design differences between web2 and web3 networks. Then, in section two, we explore how these primitives are used by web3 networks to supercharge network effects. Finally, in section three, I put the frameworks developed in the previous sections to the test by using them to generate practical insights about the future of web3 networks.

Section One — Modeling Network Effects, Web3 Primitives and Arbitrary Web3 Networks

We start at square one. What actually is a network effect? Put simply, a network effect is an increase in a platform’s value caused by an increase in its usage. I find it useful to model network effects by the following process:

New user adoption has to generate enough value to convince existing users to stay, and higher rates of retention need to sufficiently increase the value proposition for new adopters, and so on. In the early stages of a network, the rate of adoption and rate of retention need to be accelerating, otherwise this flywheel will begin spinning in the opposite direction.

Founders who discovered the secrets of network effects in the ‘web2 era’ have scaled social networks, dating apps, and marketplaces to billions of users, but the post-2010 era of internet networks has been one of zero-sum stagnation, evidenced by the rapidly declining rate of new web2 networks managing to reach this scale.

So then, why is it that web3 networks have demonstrated unconscionable growth over the past 12 months? Why is it that a simple video game like Axie Infinity could grow its monthly revenue from ~$10mm to over $350mm over a two month period, or a ‘creator token’ network like Rally can be worth almost $1bn just a year after launch?

To the dismay of web3 skeptics, answers like ‘hype’, ‘ponzi-schemes’, and ‘virtual circle-jerking’ don’t sufficiently account for this. To move beyond these kinds of lazy explanations, we have to be precise in establishing the key design differences between web3 and web2 networks.

Clarifying the Definition of Web3: ‘Shift To Web3’ = Adoption of Web3 Primitives:

The definition of ‘web3’ has become about as polluted in the public sphere as it possibly can be. For our purposes, by ‘web3’ we’re specifically referring to the set of new discrete digital primitives enabled by decentralized virtual computers (a.k.a blockchains). More specifically, we’re referring to smart contracts, the fundamental primitive blockchains which support Turing Complete programming languages like Ethereum unlock. We consider ‘arbitrary smart contracts’ as the root node of a tree of web3 primitives, each of which themselves are specific instantiations of smart contracts:

The composability of smart contracts allows them to serve as the plumbing that enables these various primitives to work together within and between web3 networks. It is the interplay between all of these primitives, facilitated by smart contracts, that forms the basis of ‘web3 networks’.

The Economic Divergence Between Web2 and Web3 Networks:

We can model both web2 and web3 networks as combinations of raw materials and capital flows. Networks have two kinds of raw materials — users and technology platform, and two kinds of capital — user capital and platform capital. User capital covers any scarce user-supplied resources, including user generated content, physical inventory, money, or even just attention. Platform capital also extends beyond platform generated revenue, and includes the non-financial utilities platforms return back to users.

Users supply ‘user capital’ and technology platforms convert it into ‘platform capital’, some of which is distributed back to users and some of which leaves the network — this is the fundamental action of any internet network:

Where web3 and web2 networks substantially differ is in their handling of platform capital. While web2 platforms siphon away almost all of the capital they generate, smart contracts allow web3 networks to return platform generated capital to users in proportion to the value of the user capital they supplied via tokens.

From an economic design standpoint, this means web2 networks are skewed Authoritarian whereas web3 networks are skewed Libertarian/Capitalist:

Web2 Network Design (Authoritarian):

Web3 Network Design (Libertarian/Capitalist):

This is neither to say that all web2 networks are perfectly authoritarian nor that all web3 networks can be labeled ‘libertarian/capitalist’. Networks can certainly use web3 primitives to create ‘authoritarian’ economic dynamics. However, enabling users to capture the value of their digital labor has become the guiding ethos for network design in web3 (and hence is the breed of web3 network we’ll focus on).

The assumptions made above about web3 networks hinge on the idea that ‘governance tokens’ capture the value of the platform capital generated by web3 networks (imbuing them with non-zero financial value). At first glance, it isn’t intuitive that tokens should have value at all.

Indeed, tokens can (and mostly do) have zero financial value. Those that capture value, however, possess at least one of the following traits:

1. Platform Utility — web3 networks use smart contracts to provide token holders with special in-network privileges, which drives token demand and hence financial value.
2. Ownership of Community Treasury Funds — revenue generated by web3 networks is sent to a community owned treasury smart contract. Governance tokens derive value from the future revenue captured by the network’s treasury, just like shares derive value from a company’s future cash flows.

Governance tokens can also combine both of these traits, bundling together application utility with network ownership. For instance, holders of Uniswap’s governance token UNI receive voting utilities and ownership over the~$7bn currently in their treasury (https://openorgs.info/). Even though the network isn’t currently generating revenue, Uniswap’s smart contracts also contain a ‘fee switch’ that can be turned on by a vote of the UNI holders which would redirect transaction fees to the treasury. The voting utility, ownership of existing treasury assets, and the potential for future revenue drives financial value for UNI (currently priced at $17). By implication, tokens which don’t possess these traits don’t play a substantial role in the network effects we’re about to cover.

Section 2 — How Networks Use Web3 Primitives To Drive Network Effects

Now that we’ve built a robust model for arbitrary web3 networks and their associated primitives, we can begin fleshing out the role these primitives play in influencing the various components of network effects (adoption, retention, value creation).

Not All Users are Born Equal — Acquiring ‘Value Producers’ via Token Reward Mechanisms

Counterintuitively, when it comes to user adoption, the objective any network is trying to maximize is not the sheer number of new users. Referring back to our model of arbitrary internet networks, networks actually aim to maximize the net user capital invested by new users.

Nascent networks are far more concerned with acquiring ‘value producers’ than ‘value consumers’, since value producers invest substantially more user capital and are required to attract value consumers. The types of value producers also vary from network to network. For networks like YouTube, value producers are the content creators. For ride-sharing apps like Uber, they are the drivers. For dating apps, as unusual as it is to use this framework, the highest value producing users are attractive women.

Although the distribution of value producers and consumers varies in its flatness for different networks, because value production is inherently much harder than consumption value producers are generally much rarer than consumers.

The incentives for value producers to join web2 networks are based solely on the payoffs they receive from value consumers on the network. Without guests on Airbnb, there are no incentives for renters to upload listings. Without riders on Uber’s network, there are no incentives for drivers to register. And without substantial audiences on Twitter, there are no incentives for high profile users to produce tweets. This is otherwise known as the infamous cold start problem.

By contrast, web3 networks can acquire value producers without a critical mass of consumers by rewarding them with governance tokens in exchange for their upfront user capital. For instance, DeFi protocols like Sushiswap rewarded early liquidity providers with SUSHI tokens and web3 social applications like Rally Network disproportionately rewarded early fast-growing creator communities using their network via weekly RLY token airdrops.

This means web3 networks can acquire value producers purely on the promise of a share of future platform capital, since the financial value of this promise is priced into the governance tokens being distributed. Again, this dynamic doesn’t manifest unless the tokens being distributed possess the characteristics we outlined in section one.

While many types of token reward mechanisms have been implemented, effective mechanisms generally disproportionately reward users that invest user capital early since, for any network, all forms of capital are most scarce and valuable at inception. Web3 networks often decrease token rewards as a function of the number of new value producers they have acquired to incentivise value producers to join as quickly as possible.

By bootstrapping the adoption of early value producers, token reward mechanisms increase the utility of the network’s technology platform. In addition to this, the network’s increasingly scarce governance token increases in value (price) alongside the network’s platform utility. This means token rewards drive network effects through the dual effect of increasing the network’s non-financial platform utility and increasing the financial incentives for adoption.

Token reward mechanisms play a significant role in selecting the types of users who join web3 networks. Networks with effective token reward mechanisms aim to optimize their token holder hierarchy, an important concept we will revisit later which largely determines the ability for web3 networks to establish defensible moats.

Second and Third Order Network Effects of Token Reward Mechanisms:

Because a user’s capital is scarce (users have limited time, money, attention etc.), the actual investment they are willing to make in any given network over time is weighted by their subjective willingness to invest. The capital any one user invests into a network can be modeled by the following equation:

For both web2 and web3 networks, W is a function of relative payoffs a user expects to receive in one network over another. Both kinds of networks must compete for user capital by increasing the payoffs users expect to receive, hence increasing W.

Token reward mechanisms are one way in which web3 networks do this. When users earn governance tokens, their net incentives to continue investing their capital into the network (to increase the value of their existing ownership stake) increases proportionately. Governance tokens are scarce by design, and networks that preferentially reward higher value producing users with these scarce rewards maximize the future user capital that will be invested into the network. More precisely, such networks maximize the following sum, where U is the set of all users in the network and I is the total user capital invested into the network:

By contrast, web2 networks often acquire early value producing users through cash based incentives. Networks like Paypal have spent hundreds of millions of dollars on strategies like this. However, governance tokens fundamentally differ from cash payments because they provide exposure against the long term financial upside of the network. This means token reward mechanisms confer web3 networks with much greater capacity to intervene on W than their web2 counterparts. Another way of saying this is the web2 version of this strategy only incentivises adoption, whereas the web3 version incentivises both adoption and retention.

The compounding impact token rewards have on network effects continue to extend beyond the actions of individual users. In web2 networks, value producers must engage in zero-sum competition for a limited share of platform generated capital. YouTube content creators will compete for views, Twitter users for followers, and Uber drivers for riders, and so on.

This dynamic inevitably exists in web3 networks as well, but is offset by the fact that governance token holders all mutually benefit from the upside of the network’s value together, a dynamic popularly dubbed the ‘alignment of user incentives’. This is also why governance tokens are often called ‘digital equity’. Unlike traditional equity, however, the number of ‘digital equity owners’ can reach the scale of thousands very early on through token rewards. What this means is net amount of cooperation/positive sum interactions between users is substantially higher in web3 networks, ultimately improving the efficiency with which user capital is deployed.

Composability Driven Network Effects in Web3:

Unlike web2 networks which are closed-source and siloed, web3 networks are open-sourced and composable. Smart contracts, tokens, and other web3 primitives in one network can natively interact with those in others. Crucially, the rules of these interactions are governed only by the code contained in the relevant smart contracts, and require no coordination between people to carry out.

Instances of network-effect inducing composability boil down to mutually beneficial transactions of capital between web3 networks. In other words, network composability can make the capital users invest in one network productive in another.

DeFi protocols are a perfect case study for composability driven network effects. For instance, the crypto assets users supply Yearn Finance are used to provide liquidity for networks like Compound in exchange for yield, simultaneously increasing the returns for Yearn Finance users and improving application utility for Compound users. As the number of protocols in DeFi that reach maturity continues to grow, the opportunities for Yearn Finance to compose itself with additional yield generating strategies (and the accompanying network effects) in other networks continue to increase.

As the broader web3 space matures, we can expect composability to be a substantial driver of network effects not only for DeFi protocols, but also web3 consumer applications. We’ve already seen several examples of composability strategies at the intersection of DeFi and web3 social, like Rally Network’s use of Yield Delegating Vaults on Yearn Finance to bootstrap their community treasury (https://amit-rally.medium.com/introducing-yield-delegating-vaults-f861a11afb0b). If we push this trend closer to its conceptual limit, it would not be surprising if the growth multiplier inter-network composability creates becomes the main driver of web3’s transcendence over web2.

The Relationship Between Open-Source, Composability, and Defensibility/Retention For Web3 Networks:

Although the composable and open-sourced nature of web3 networks can produce strong network effects between projects, it clearly increases the surface area for attack by new web3 networks. For instance, despite gaining substantial traction after mainnet launch, Uniswap was still vulnerable to a ‘vampire attack’, temporarily losing liquidity to Sushiswap (a fork of Uniswap with an adapted governance token reward mechanism). Rather than facilitating mutually beneficial capital transactions, attacks like this involve extracting capital from one network and inserting it into another. Examples like this are legitimate causes for concern.

Given that web3 software isn’t proprietary and there are very low barriers to exit for token holders, is it possible for web3 networks to establish defensible moats? Before answering this question, we need to understand what ‘retention’ and ‘moats’ mean specifically in the context of web3.

For any web2 or web3 network, users can be considered ‘retained’ provided they continue to invest a minimum threshold of user capital into the network. Web2 networks measure this by analyzing application usage data. Web3 networks have access to the same usage data, but retention really boils down to how likely users are to hold (hodl) their governance tokens. When users exit their token stakes, we can consider them churned.

The durability of any web3 network’s moat is tested whenever there is a significant negative perturbation in the price of the token, which can be caused by events like vampire attacks, token holder collusion, security breaches, changes in market conditions etc.

Web3 networks successfully establish moats when they convince a critical mass of users that holding onto their ownership stake is a dominant strategy. As we mentioned earlier, some users are substantially more important to retain (value producers) than others (value consumers). Another useful way to segment users in web3 networks is by the dichotomy between ‘missionaries’ and ‘mercenaries’. Missionaries are generally more value producing, intrinsically believe in the mission of the network, and have a disproportionate impact on the subjective culture within the network’s community. Mercenaries on the other hand are more value consuming, lack intrinsic motivation to grow the network, and are motivated exclusively by financial gain. We can measure a user’s position on this spectrum, in part, by evaluating how they respond to a negative perturbation in token price:

‘Missionaries’ hold long-term views on the growth of networks they participate in and hold onto their token stakes even during large negative price perturbations, whereas ‘Mercenaries’ hold short-term views and will exit their stakes when token prices begin to falter.

We now introduce the concept of a network’s token holder hierarchy, a term which refers to how governance tokens are distributed between a network’s missionaries and mercenaries. Different token holder hierarchies have different levels of ‘robustness’, where robustness refers to the ability for a network to retain its token holders after a negative perturbation in its token price. We can approximate the robustness of a network’s token holder hierarchy by computing the conceptual sum below:

The premise of this sum is that networks are maximally robust when their largest token holders are the least price sensitive users (missionaries), and the smallest token holders are the most price sensitive users (mercenaries).

This sum demonstrates the crucial role token reward mechanisms play in establishing the durability of a network’s moat. If mercenaries can hijack reward mechanisms and earn substantial network ownership, the network’s token holder hierarchy won’t be robust, the majority of users will churn when the token price is perturbed, and the network won’t be able to sustain its moat. If, on the other hand, a network’s token reward mechanisms is geared towards disproportionately rewarding missionaries, the network can gain huge durability advantages over its competitors.

This explains why networks like Axie Infinity have established durable moats without having substantially differentiated products. Sky Mavis, the company who created Axie, was in the trenches developing the network for years during the post-2017 crypto winter. Inevitably, this meant their early users were disproportionately value producing missionaries who played in spite of the bear cycle.

According to Jeff ‘Jiho’ Zirlin, co-founder of Axie Infinity, “The community needs to be small first to work when it gets really big. That’s what makes Axie hard to copy — if you build a competitor now, you’re going to attract the kind of people who want to find the next Axie, not the kind of people who are actually interested in moving gaming forward”.

In the context of our models, what Jiho is really saying is that Axie’s userbase has a strong foundation of missionaries and the net incentives for its largest token holding users to hold their ownership stakes is much higher than that of its competitors. This has allowed them, in part, to establish a durable community driven moat.

Section 3 — Applying Our Models of Web3 Network Effects to Generate Practical Insights

At this stage, we’ve generalized the major ways in which web3 primitives facilitate new kinds of powerful network effects. However, our frameworks are only as good as how useful they are at providing us with practical insights about the future of web3 networks. Here are some of the insights I’ve arrived at, using the frameworks developed in sections one and two as the basis.

Capitalist Design Will be the Dominant Network Design Pattern in Web3:

Web3 primitives finally make it possible for users to earn a cut of their digital labor in the form of governance tokens, producing a much fairer and compelling user experience. Networks which choose not to adopt tokenomic designs with these dynamics will lose on user experience and will be outcompeted.

The corollary of this is that web3 networks will compete on the amount of value they allow users to capture, which means we can expect web3 networks to become decreasingly adversarial to users over time — the exact opposite of how the relationship between web2 networks and their users has evolved.

Moats in Web3 Will be Easier to Establish but Harder to Defend than in Web2:

Web3 networks can (and do) establish moats when they sustain growth and optimize their token holder hierarchies to be sufficiently robust. However, new web3 networks have a powerful armory of new primitives that make attracting high value producing users, and hence disrupting existing networks, much easier. This dynamic increases the level of volatility we can expect to see in the establishment and disintegration of web3 moats.

Web3 Networks Will Be 10x as Valuable Per User Compared to their Web2 Counterparts:

As we’ve discussed, web3 networks unlock capitalism on the internet. Throughout history, every major shift towards capitalist economies has led to immense increases in total production (and hence GDP). Web3 networks will increase the average value produced per user by a similar magnitude. From the outside, it will appear as though web3 networks are achieving valuations at a >10x greater multiple than web2 alternatives on a per user basis.

Token Reward Mechanisms (Airdrops) Will Become Increasingly Granular:

Early experimentations of token reward mechanisms were fairly simple. Uniswap’s initial airdrop, for example, was a flat 400 UNI tokens for any wallet that has used the protocol by a set date. As the space matures, web3 networks will compete to optimize the amount of future user capital their users invest. This means token reward mechanisms will become much better at measuring and rewarding value producing behaviour, inevitably making these mechanisms more granular and complex. This increase in granularity will be compounded by the new growth in web3 social applications where the investment of ‘user capital’ is, by nature, conducted through extremely granular interactions.

Hybrid Networks (Networks With Centralized and Decentralized Components) Will Perform Effectively In Spite of Scalability Problems:

As the web3 ecosystem moves beyond the DeFi boom into an era of web3 consumer social applications, it will be much harder for all user interactions to scale on-chain. Because these new kinds of high-touch web3 networks ‘aren’t fully decentralized’, concern has grown that these networks won’t succeed since web3 networks must be ‘community owned and controlled’.

Although centralizing network components places limitations on user control, provided networks still allow users to earn network ownership via governance tokens, the lack of decentralization won’t significantly undermine growth. This is because users, on average, place a substantially higher premium on ownership over a network’s revenues than control over the specific changes in its technology platform. In other words, properly designed hybrid networks will still have access to the all of the ‘web3 network effects’ we outlined in the earlier sections.

Niche Web3 Networks Will Have Better Odds of Surviving than Niche Web2 Networks:

Web2 consumer internet networks could only self-sustain once they reached the scale of millions of users. Users in web3 networks generate and capture sufficient value to self-sustain with far fewer users, benefit from composability driven network effects, and evangelize value producing users much more powerfully.

Most crucially, composability with the DeFi ecosystem reduces the barriers to entry and exit for token holders of these communities. Users know their ownership stakes are liquid regardless of the size of the network, which increases their willingness to buy and hold their tokens. Ultimately, this means niche networks that don’t reach massive scale can survive more easily in web3, and that the distribution of communities we can expect in the web3 space will include a higher percentage of niche communities than in web2. This, in part, explains the recent wave of Discord based DAOs.

Final Word:

I’d love to hear your thoughts on the models laid out in this essay. Are there other types of network effects in Web3 that you think are worth considering? Are there hidden assumptions being made in the models I laid out that are worth dissecting? Do you agree/disagree with my predictions? Let’s discuss.