A respected scholar specialising in platform competition and strategy, Prof. Joost Rietveld has authored numerous papers and curated an extensive library of platform research. In his role as ‘Academic in Residence’ at Launchworks & Co, Joost provides a crucial link between academia and our research-driven strategy consulting.
In their second conversation on network effects, Benoit Reillier speaks with Joost about the practical complexities of unlocking and harnessing network effects.
Harnessing the power of network effects
The complexities of unlocking and harnessing network effects are due, in part, to how they can be influenced by both the underlying market characteristics and the design choices made by platforms themselves. While some markets are characterised by strong network effects, others are more flexible. In the latter case, Joost notes:
“We see firms making heterogeneous choices in terms of how much they want to rely on those network effects.”
Indeed, Benoit points out that some markets that were traditionally not considered network effects-driven, such as taxi companies, are now being disrupted by players who build their businesses around network effects and digital platforms. The type and strength of network effects play a significant role in shaping the success and disruption potential of such platforms, whether they are local like Uber or more globally oriented like Airbnb.
Regardless, platforms need to be designed with potential network effects in mind. Joost reminds us that there are certain pitfalls for new entrepreneurs who are not familiar with network effects, such as trying to scale without building a critical mass of users within a target community. If the platform requires local network effects to thrive, but the users are globally distributed, then metrics like the total number of users are not capturing the real picture, he says. The key is to ensure that users on the platform have a reason to engage and transact with each other, as this is when network effects truly manifest.
New platforms or products can also find themselves facing a “chicken and egg” problem when it comes to network effects. For example, without games, a video game console like Microsoft’s Xbox offers little value to gamers, but the reverse is also true. One approach is to launch a product with strong standalone value and then add network value once the product has attracted users. Joost adds:
“One way to compete against a product or a platform with strong network effects is by being significantly superior when it comes to your standalone value to your competitor.”
Another way is through pricing, for example, by subsidising the product more heavily than competitors as a way to attract users.
In the ongoing research on network effects, Joost points to three major areas of interest. First, there is the question of distinguishing between different types of network effects and the appropriate timing for each:
“When should I design for network effects? How can I strategically influence the type of engagements that I want on my platform?”
Second, although the jury is still out as to whether machine learning and AI are responsible for network effects. In Joost’s opinion, there is a definite overlap between having an engaged network of users and being able to learn from those users to safeguard market position.
“Network effects and learning from data go hand in hand,” he says.
However, as technological paradigm shifts and changes in consumer behaviour continue to influence market dynamics, academics are considering how to structure and categorise the differences between network effects and learning effects from user generated data.
And third, there are many questions arising from Web 3.0, for example, the extent to which decentralised platforms can supplement or substitute centrally governed platforms. Decentralised Autonomous Organizations (DAOs) and their use of Non-Fungible Tokens (NFTs) as a form of subsidisation to attract a critical mass of users are among the new developments in this area. Benoit notes:
“Although these are really emergent trends and it’ll take time before existing firms start to harness these new models, interesting experiments are going on.”
In this thought-provoking exchange, Joost and Benoit move beyond basic definitions and delve into the intricacies of network effects. To learn more, read the full-length interview below. And be sure not to miss their previous conversation, where they explored different types of network effects, common misconceptions, and much more.
Harnessing the power of network effects — in-depth interview
Benoit Reillier:
When I tried to get my mind around the research on network effects from the early days, I remember that academics were looking at whether markets themselves were subject to network effects rather than firms—which were considered black boxes operating in markets. What I observed in the murky business world however was quite different. In fact, firms were actively trying to design products and services in such a way that they would give rise to network effects. Clearly both the underlying market characteristics and the platform design drive network effects so I used to call this the ‘nature’ vs ‘nurture’ debate. Where are we with that debate? Has research progressed much in that area?
Joost Rietveld:
A little bit more. But I think that as academics, we’re also just coming to grips with the fact that not all network effects are created equal. And you’re right. We very much started out from the argument of early mover advantage where we would think about network effects as some exogenous feature of the market itself that you operate in. Some markets are characterised by network effects, others are not. I think if you look back at some of the seminal papers in this space, that’s what they say. In markets characterised by network effect firms must behave differently. So, the assumption for many years has been that either you play in a space that has them or don’t. If the space has network effects, then you need to think about subsidisation, early market entry and so on.
With the advent of digitally distributed products, software products, I think we’ve come to realise that you can make certain choices in your product design that would either have you rely more or less on network effects but also could have very serious implications for the type and strength of network effects.
When you develop a video game, you can have a single player video game—for example Tomb Raider—or you can have a multiplayer video game like—say Fortnite. Fortnite clearly has network effects and those single player games do not. So, this is a market that may or may not be characterised by network effects. In this market, we see firms making heterogeneous choices in terms of how much they want to rely on those network effects. Neither is better or worse than the other. It goes back to this value bar. If you rely on multiplayer, then your initial value bar will be lower. But if you manage to get that player base, then your value bar will grow quite significantly, and you might have a larger value bar than those single player games that only have standalone value. So, it’s a risky strategy, but the payoff is potentially much bigger.
Benoit Reillier:
That’s interesting. We increasingly see markets that were initially thought of as traditional markets–like taxi companies, that didn’t see themselves at all through the lens of network effects–now being disrupted by players that build their entire business on network effects and digital platforms. In some cases, these businesses even go global. So, it seems to be placing increasingly network effects or decisions that impact network effects at the firm level as opposed to as an exogenous driver.
Joost Rietveld:
It goes back to the type of network effects or the strength of the network effects that you might be designing for. If you’re Uber and you’re a user of Uber, you need a very dense network in your local geographic area. Therefore you need to design for a very local network effect. If you are Uber, you’re going to enter different cities at different points in time. You’re going to try and create the network in each city to ensure that a local user actually can benefit from those network effects, that the network effects materialise.
Whereas if you are Airbnb, you need a much more global network because I might live in London, but I’m looking for holiday accommodation in San Francisco. The network needs to be developed at a global scale. It’s a very different type of network topology and a very different type of network effect. And you might think that some of these choices are kind of endemic to the type of business that you’re in.
But if we think about Facebook, Facebook started out as a social network for Harvard College students. What did Mark Zuckerberg and his team do? Engineer for a very tight dense local network. That’s because it’s easier to get a large share of a small network. And rev up the network effect. And then we can open up to students from other colleges. And probably, there’s already some connections in that network because we might have friends studying in other universities. It’s easier to get that network going once we have the Harvard network. Once we have the college network, we can say, okay, now we’re opening up to everyone in the US and then we’re going to expand globally, including my mother.
Mark Zuckerberg probably had the strategic foresight: which networks we want to play in and how much of the network we need to get on board to get network effects kicking in.
Benoit Reillier:
I think this is a fundamental point. When we help firms design and launch platform businesses, very often people who start up these new businesses are not familiar with network effects… They tend to go for scale rather than think about what a critical mass of a given niche community may look like. They end up with vanity metrics like the number of people on the platform that may look great. But it’s people who have nothing in common, or actually people who are not in the same geographic area. It’s not helpful if the platform needs local network effects. So, you really need to think strategically about what kind of community will help you test your platform, and the interactions and transactions you need to test. This is very different from a traditional product launch in retail for example and a big shift in management thinking…
Joost Rietveld:
Absolutely… Should you have a very small slice of a very large pie, or a large slice of a smaller pie? And that’s just generally something to think about when it comes to how you think about value creation and capture in a platform world. But we need to be very clear about what it means to have network effects. You have network effects when money is being exchanged, or when users are engaged, and they’re spending a lot of time on the platform and have value-adding interactions with each other.
It gets to this very issue. If you are Uber, and you have a million users on your network, but they’re scattered across the world, then you’re going to have zero interactions between them. You may have a large network with zero network effects. So, you need to ensure that the users on your platform, whether they are all consumers or consumers and merchants or other types of complementors, have a reason to engage and transact with each other. Because that’s when network effects manifest.
Benoit Reillier:
That’s why one way of looking at platforms is to look at them as portfolios of market positions. When Uber launches in a new city, they’re effectively a new entrant without critical mass and without network effects. The only thing that helps them is their brand as well as the playbook they’ve developed on how to very quickly ignite and launch a country—sometimes by paying drivers to be available in the early days to kick start the network. It’s the same with lots of platforms that rely on local network effects like Tinder, Deliveroo, etc… Some social platforms like Facebook seem to have fewer local network effects…
Joost Rietveld:
Yes, although Facebook had very local network effects when it started… just classmates from Harvard. I think it’s a strategic choice, a design decision, much more so than a feature of the market that you’re playing in.
Benoit Reillier:
So how do you embed these network effects into platforms? Are there specific features you can add to your products or services to benefit from these network effects?
Joost Rietveld:
If you’re a platform, so a two or multi-sided organisation, then I think you need network effects by design. If there’s more participants on one side, then it’s going to be more attractive to participants on the other side because they will have more opportunities to transact. When it comes to direct network effects, broadly speaking, anything that sparks social interaction or social engagement is something that can trigger network effects. And I think that a strategic consideration here is when you play in a space where your product has some standalone value and potentially some network value, you could be somewhat strategic about this. For example, you could launch your product in a way that relies more on standalone value. Just market the product as something that has standalone value and is attractive on its own regardless of how many people use it. You can then start adding the network value as you go along and have accumulated some users.
Benoit Reillier:
It’s one of the ways to overcome the chicken and egg problem… it’s what OpenTable and others did. You offer a traditional product (a booking software that restaurants use themselves) and then, once you’ve got a critical mass of restaurants using your software, you can open up a booking platform and then you start getting the other side of diners connecting with restaurants and this creates network effects. It makes it easier because you already have a critical mass of restaurants to start with.
But how do you compete and win in that world?
Joost Rietveld:
If we think about the framework of the value proposition and the value bar, one way to compete against a product or a platform with strong network effects is by being significantly superior when it comes to your standalone value to your competitor. It’s easier said than done but if your standalone value bar is greater than your competitor’s standalone network value bar, then you’re giving users a strong incentive to shift to your product even in the absence of network effects. An obvious way to win is to offer significant value, in terms of innovative tech or superior quality for example.
The other way is through pricing. You can subsidise at a higher rate than your competitor so that, even if you don’t have the network, you are priced very competitively and therefore you incentivise users to switch and you can build up your network. Another way is, and I think Nintendo has done this very well, you could decide, I’m not going to rely very strongly on complementors for the supply of content on my platform, but I’m going to produce it myself. And so, I’m going to make Super Mario, I’m going to make Zelda, I’m going to make all these great games that will attract users to my platform and that then will eventually attract third party sellers.
Benoit Reillier:
Nintendo is an interesting example. I guess the danger is that you must make sure you know how to develop good content yourself. I remember a portal platform called Vizzavi that was a joint venture between a media group (Vivendi) and a telecoms operator (Vodafone). They thought they would do everything themselves and let’s say it didn’t work out as intended.
Joost Rietveld:
It clearly depends on who you’re up against, what market you’re playing in, etc. And it’s incredibly difficult to give blanket statements such as “you should always do this or you should always do that”. And I think that’s where experts like us can come in to help platforms understand what some of the intricacies of the space they’re playing in are, and the pros and cons of the various strategic options available to them. It can be superiority, subsidisation, vertical integration, producing your own complements, or a mix.
Of course, there is also the envelopment strategy, where you seek to make your platform or product compatible with that of your competitor to make it really easy for their users (on both sides) to also be on your platform. And then, if along the way you find something that allows you to differentiate yourself from your competitor, because you manage to be better/cheaper, etc. then you might find yourself in a position where ultimately, you’re going to beat them.
Benoit Reillier:
I remember an excellent paper from Tom Eisenmann on enveloppement strategy a while ago.
Joost Rietveld:
Yes.
Benoit Reillier:
Any other thoughts or anything you find interesting in the literature recently about network effects or things that you think should be researched more?
Joost Rietveld:
Well, so one thing that we didn’t really touch on is the role that machine learning and AI play in this. And I find myself slightly in murky waters here because there’s a debate going on in the literature right now. Are those network effects or not? Some people refer to them as data network effects, whereas others simply refer to them as learning effects, but they are very, very interrelated. If you don’t have any users on your platform, then you cannot really train your algorithms for the purpose of making recommendations or for the purpose of learning about your users, or for the purpose of understanding where you should take the product next in terms of expansion.
On the other end of that spectrum, when you have so many users on your platform, you not only have network effects that can manifest your dominance, you also have all this learning from the data that you’re generating that may allow you to better service your existing users by giving them better recommendations, by better understanding where you should take your product next in terms of maybe expansions or add-ons. And generally, you can focus on bigger trends, at least when it comes to users on your platform that could help you to fend off competitors.
And so, the jury is out whether these are network effects or learning effects, but I think they are very much intertwined with this notion of having a large network of users and especially when that network is engaged, you can learn from them and safeguard your competitive position in the market.
Benoit Reillier:
It’s super interesting. Everybody has been signing up to OpenAI to play with ChatGPT and DALL·E 2 at the back end of last year… We are at the beginning clearly of these AI capabilities and it’s already producing pretty impressive results.
I read a poem explaining some complex economic theory that was quite impressive and produced in a few seconds by ChatGPT.
So how these networks or learning effects play out with AI will be very interesting… firms like Palantir for example are doing very interesting things with data because they get the metadata and the intelligence from the data of their customers. Even if it’s not their data and if they don’t retain a copy, they benefit from the learning effects since the data is processed by their systems. So, when people talk about the NHS giving their data to Palantir, the data is secure and it’s not given to Palantir. But by simply processing the data, Palantir can gain insights, can train new neural networks and can actually derive value as an entity, as a company. And people are still trying to get their minds around these learning effects or network effects, data network effects…
Joost Rietveld:
And at a more mundane level, for example, when you and I go to our Netflix accounts and we’re interested in watching The Crown, we’re going to see different thumbnails for that show that are driven by our viewing history on the platform. It’s the same logic on the iOS app store. If we go to the storefront on our respective mobiles, we’re going to see a different selection of apps depending on the apps we have downloaded and used in the past. If you’re Amazon—they say they don’t do it, but some people say they are—you could use customer data to better understand which product spaces Amazon should be active in as a seller themselves, rather than just facilitating a marketplace. So, this is not rocket science, but these insights require a network of users that you can learn from.
Network effects and learning from data go hand in hand. We learn from users and I think as long as we’re in somewhat of a world that’s in equilibrium, these two can really lead to a position where dominant platforms become more dominant and maintain their dominance over time. Obviously, when there’s a paradigm shift in the world, then those algorithms may be less effective. But yes, I think we’re still trying to figure out what the implications of these effects are on market dynamics. I don’t quite have a framework yet for structuring and categorising them.
Benoit Reillier:
When you say a paradigm shift may change market dynamics, are you referring to things like a shift in the interface used so everybody starts using VR and then suddenly the players are different and the network effects that have been created in previous platforms may or may not translate in the new one? These are the kind of market shocks or paradigm shifts that you’re talking about, right?
Joost Rietveld:
Yeah, just a sudden shift in consumer tastes or perceptions.
Benoit Reillier:
So it could be a change of taste like TikTok making people fall in love with five second videos…
Joost Rietveld:
Yes. Short form videos changed things and YouTube had to become a big player in the space. Many of these large tech companies have their feelers out and they’re just hedging their bets so that whenever a new paradigm comes along, they can quickly adopt it and co-opt it even as something of their own. But obviously that requires, again, a lot of learning, a lot of investments, a lot of awareness of the world / market scanning. It’s not easy.
Benoit Reillier:
No, it’s not straightforward but it makes a lot of sense. It’s still early days but I will be very interested to see how these new types of AI mediated network effects or learning effects play out effectively. For example, the extent to which they’re effectively subject to diminishing returns will partly determine the minimum efficient scale of market participants and competitive dynamics in some of these markets.
Joost Rietveld:
Yes, absolutely.
Benoit Reillier:
I’ve read about an app based on ChatGPT that basically uses data from thousands of people to segment the market and present to them the best sales pitch with the information available. For example, you have parameters like gender, time of day, location, etc. and the app generates the most effective sales pitch and keeps learning. So that’s impressive when you see the time it takes to craft a newsletter or an email or something, having an AI capability that automates that and learns from it and leverages all the previous customers, that’s quite powerful and that’s probably the kind of data network effects you’re talking about.
Joost Rietveld:
I think this is a good example of why some academics feel that these are not network effects per se, but more so learning effects because this helps to improve the offering of the firm but it might not necessarily create more value for the specific user using the product. And that gets us back to the start of the discussion. It’s really what the network effect is as when there’s more users using a product of value that either becomes bigger.
Benoit Reillier:
So, unlike network effects, AI learning effects may have more to do with producer surplus than consumer surplus then. If it’s more effective at selling to me but doesn’t add value to somebody else. Any area or anything you haven’t talked about? Any research topic that is a burning platform if I may?
Joost Rietveld:
I see what you did there.
Benoit Reillier:
Yes 😉
Joost Rietveld:
My PhD student, Joe Ploog, and I are really thinking about this distinction between different types of network effects and when they arise as a result of nature versus nurture to use your language. So that’s something that I think we’re going to see a lot more of in the coming years in the academic space. When should I design for network effects? How can I strategically influence the type of engagements that I want on my platform? If we’re to just have a better understanding of the strategic thinking about network effects, that goes beyond do I enter a market that has or doesn’t have network effects? I think that’s a very interesting theme research wise.
The second is this distinction between network effects and learning from data. Under what conditions can those learnings be network effects? When are they not? And how do they inform each other? I expect that they are mostly complimentary but perhaps in not all cases.
And then, I think the third one that is interesting to think about is, even though we are in the middle of a crypto winter, just the extent that decentralised platforms can supplement or substitute centrally governed platforms. I don’t think they can completely, but we’re probably going towards a space where we’re seeing hybrid models that partly rely on decentralised governance mechanisms and partly rely on centralised globalised governance mechanisms. And then the question becomes when you should go for which.
Benoit Reillier:
And we’re seeing that with Distributed Autonomous Organizations (DAOs), when votes get triggered by members of the community and governance decisions get made by people with tokens. Some platforms are also experimenting with tokens to overcome the chicken and egg problem we talked about. Giving tokens is a form of subsidy that can be distributed for a very low marginal cost if it’s an NFT. And so, we are thinking about this as well with some of our clients. Although these are really emergent trends and it’ll take time before existing firms start to harness these new models, interesting experiments are going on. As we’ve seen with value chains and platforms, different business models will co-exist. We’ll probably see DAOs operating in some specific markets. It would be super interesting to have a framework to look at predictive powers of where DAO models can give better outcomes in welfare terms when common goods are at play.
Joost Rietveld:
Yes, you’re right. When we think about two-sided platforms where you need to have this subsidisation both across sides and across time periods, generally I think that you’re much more likely to rely on centralised governance. That’s because you need something or someone that has oversight on both of the sides and also has foresight in terms of if we do this now then later we’re probably going to benefit from. So, it requires decisions that probably look bad at least to one side of the platform in the here and now, but might have benefits over the long run. And I think that’s much harder to design for in a decentralised manner than in a centralised fashion. But again, this is just me conjecturing here and now. We need more research on that.
Benoit Reillier:
So that’s what the best people do at the moment, conjecturing and thinking about the research that can be done and doing it! So, thank you very much Joost for this fascinating discussion.
The first part of the interview covered network effect definitions and why network effects matter. For those who like to go deeper on the topics discussed in this interview, here are relevant research papers:
Rietveld, J., & Ploog, J. N. (2022). On top of the game? The double‐edged sword of incorporating social features into freemium products. Strategic Management Journal, 43(6), 1182-1207.
Rietveld, J., & Schilling, M. A. (2021). Platform competition: A systematic and interdisciplinary review of the literature. Journal of Management, 47(6), 1528-1563.
Rietveld, J., Schilling, M. A., & Bellavitis, C. (2019). Platform strategy: Managing ecosystem value through selective promotion of complements. Organization Science, 30(6), 1232-1251.
