Simone Cicero 

Hello, hello everybody and welcome back to the Boundaryless Conversations Podcast. In this podcast, we explore the future of business models, organizations, markets and society in our rapidly changing world. I’m joined today by my usual co-host, Shruthi Prakash. Hello Shruthi.

 

Shruthi Prakash 

Hello everybody.

 

Simone Cicero

Nice to see you. And we also have a special guest today that shares the passion for platforms and ecosystems. And it’s like a little bit of a legend in space. Welcome to the podcast, Sameer Singh.

 

Sameer Singh 

Thank you Simone and Shruthi, glad to be here.

 

Simone Cicero

Thank you, Sameer. Sameer is a long time writer and explorer of platforms, ecosystems, marketplaces, network effects. He also writes on his blog, breadcrumb.vc, which many, of you I’m sure know. And he’s also a partner at Speed Invest where he looks into marketplace and especially network effects enabled organizations. We’ll talk about this more in the future minutes. And he’s also an instructor at Reforge where he teaches a course about network effects. 

 

So glad to have you here with us today. As a starting opening question, let’s say, right? Maybe we can look into your practice. You have been into this world since more than 10 years now (since 2021), and in the last, I think it’s more or less five years, you’re doing investing so strongly. Okay. 

 

And so I would be curious to hear how your perception of the market of this industry, let’s or the sector in the industry has changed over time, and especially what you have seen in terms of changes and impacts in your role as an investor, right? So maybe that’s a good starting point.

 

Sameer Singh 

Good question. I’d say network effects are kind of always hard to find. And one thing I’ve consistently seen is that people always mistake what a network effect is. And it happened in sort of every technological era. It happened in mobile apps. It happened once people sort of ran out of marketplaces and started labeling things like scooter companies and 10-minute grocery companies, marketplaces. It has happening in the current AI era. The biggest change is sort of the frame of reference through which the mistakes happen as opposed to the mistakes. the core fundamental constructs of network effects are basically based on math. They don’t really change. They’re fundamentally always the same. And the number of companies that have them will always look small minority. The only recent change I’m seeing is that as the volume of dollars available has grown, it increases the noise in the ecosystem.

 

So it’s even harder to separate signal from noise and kind of the right companies of the companies that kind of have that glimmer of a network effect at their core.

 

Simone Cicero 

Do you make more mistakes? Because the noise is broader, right? So do you make more mistakes or maybe you are just more attentive in order to separate the signal from the noise?

 

Sameer Singh 

Good question. I mean, in this job you always make mistakes. In fact, you make more mistakes than get them right. At least the nature of mistakes tends to vary. If you make the same mistake consistently, then something is fundamentally wrong. And so the things that I try to not make mistakes on are the fundamentals. 

 

Is there a a seed of a network effect in this product. Is there a unique structure of multi-player interaction at its heart? I think that that bit is more of a science. So that part’s kind of easier to get right now. How much of a problem does that interaction solve? That’s so halfway between science and art. Like there is definitely scope to make mistakes there. And then there is how committed is the founder to seeing out and truly solving that problem, there’s even more scope for mistakes there. And so that’s kind of how you would look at it. The science part of it, you should not make mistakes. The art part of it, you kind of underwrite some mistakes.

 

Simone Cicero 

So it looks like there is maybe a broader scope of application of these principles in the last few years. With this broader set of application spaces, it’s easier to find maybe non-optimal teams or non-optimal opportunities. So pick the real the real ones that really have the scalable network effects, it’s a bit of a challenge, it’s becoming a bit of a challenge.

 

Sameer Singh 

Yeah, and also it’s useful to remember that for consumer startups in particular, which is sort of my neck of the woods, generally you don’t find a massive correlation between things like founder background and eventual outcomes. Whereas that correlation tends to be a bit stronger, let’s say, know, deep tech startups, biotech startups, and maybe even SaaS startups, like not really, that’s not my area, but in this world you don’t. 

 

So you’re often looking at first time founders and the learning curve for them is pretty steep. And so trying to separate the folks that have the slope, so to speak, are on the right trajectory, really understand this problem at a very, very deep level, have a certain missionary zeal to solve it and not just mercenaries. I think that’s where some of the art sits.

 

Simone Cicero 

I know Shruthi has a brewing question, but I wanted to ask you maybe a last follow up on this, especially focusing on consumer tech and consumer companies.

 

What is changing? Because it looks like the markets are changing a lot. Changed in terms of, how can I say, it’s definitely a market where there are lots of alternatives now. It’s overwhelmed with options. The major categories have been captured already. Distribution is becoming extremely more competitive. Social habits are changing rapidly. AI.

 

Of course, we’re talking about AI more deeply, but AI is impacting the type of interactions we have. 

 

So what are you seeing in the consumer space? What are the two or three things that you are excited about these days when you look for companies?

 

Sameer Singh 

Good question. That’s it. The two main things that seem interesting right now. 

 

The first is I mean, products that are leveraged technology changes are always interesting. So are there multiplayer interactions that can be built today, now that AI exists, that could not be built a couple of years ago? And that’s an interesting area. I made one investment this year off the back of that thesis, not announced yet, but essentially think of it as a game, a multiplayer game that could not exist without AI, mobile multiplayer game, built off of image models.

 

So that’s sort of one interesting area. There’s a couple of other sort of areas I’m looking there in sort of productivity enhancement and stuff like that. The second interesting area is just sort of behavioral shifts in new generations. If you look at, let’s say, mean, Gen Z is a very overused term, so I kind of grimace when I say it, but if you look at Gen Z and compare them to folks that are 10 or 20 years earlier, they are more willing to share stuff about themselves with their friends than those folks are. So stuff like sharing your calendar with your friends, your location with your friends, your dating life with your friends, even shared payments. There’s a whole lot to unpack there that didn’t really exist a while ago. So while I think quote unquote, a social network as a construct has been played out, it’s been done. There’s probably not a whole lot of scope there, but there is scope for building out something at the intersection of let’s say social and utility or social and productivity. And we’re seeing some interesting plays there.

 

Shruthi Prakash 

If, yeah, if I can jump in right just on the last point before we sort of digress. Yeah. I mean, I’m curious. I’ve always been curious on the behavioral side of things. So, I mean, when you spoke about behavioral shift in, let’s say the Gen Z, how does that sort of adapt in traditional systems or systems that are, you know, many a times built by incumbent companies or traditional, let’s say B2B sort of perspectives which might not be accommodative of, I don’t know, newer mindsets and shifts.

 

Sameer Singh 

I mean, that’s the beauty of a consumer product, right? When you build a consumer product to solve a particular pain point, you’re going straight to the user. The user tends to adopt these products and spread these products with their social circles. 

 

And that bottom up organic user growth is kind of what we’re looking for. And so it tends to circumvent existing system beyond sort of distribution. If you’re following traditional distribution channels, then yes, that’s a bit of a problem. And sometimes, you do need to leverage them for parts of your acquisition flow. 

 

But the higher your K factor is, the higher your word amount coefficient is, the less you need to rely on traditional modes of distribution. Now that is the holy grail. Like if you can get to a K factor of 1.1, like that’s exceedingly rare. That’s really happened that often. But there are companies that have gotten half way there, close to that. And that does go a long way towards minimizing the impact of sort of saturated paid marketing channels and stuff like that.

 

Shruthi Prakash 

Got it. Okay. So I mean, I wanted to ask this before jumping in, which was on, let’s say the AI integration, which you brought up as well. So sort of the language around it when it comes to a platform shift or a not platform shift that’s sort of happening. What does all of this mean in an ecosystem design? I think we’ll be curious about it from our side. 

 

How does this impact existing platforms? Is it sort of spawning in new platforms, how does that look when AI is integrated?

 

Sameer Singh 

I mean, I’ll start with this, right? So if you bring a room of investors together and ask them: “hey, do you think AI is a platform shift?” about nine or 10 will raise their hands. If you ask them: “hey, can you define what a platform shift is?”, you will not see as many hands. And the answers will not really be consistent. So you start with a: “What on earth is a platform shift anyway? What does that mean? How does this fit in?” And therefore: “how does this then impact existing platforms?”

 

The way I’ve defined most platforms is you need sort of four components: You need an underlying product, a product that someone uses. You need a way to build applications for that product. You need to then match the users of your product with those applications. Then you need an economic benefit for those developers. Without those four components, you do not have a platform. And in this case, I’m using a platform in sort of a very, very specific way. In academia, platform is a broader term. To me, platform is a much narrower term.

 

So if you’re looking at a category-level platform shift, the underlying product has to be a product category that people use, let’s say, browsers or smartphones. And then you have developers building on top of those browsers or smartphones, or developers are building websites, or developers are building mobile apps. And then sort of the rest of it kind of followed from that. So the more users you had off browsers and smartphones, the more valuable it was for developers to build on those platform categories. And the more websites and mobile apps there were, the more useful browsers and smartphones became to end users. 

 

So that is what created a new platform. Essentially, you had a new platform on which to build on. And there’s a network effect between both those sites. Now, if you look at AI and LLMs, none of that is happening. Yes, a lot of people use ChatGPT, but nobody is building applications on top of ChatGPT. People do not access or use applications through ChatGPT. That does not happen.

 

This is a new technological capability for sure, it potentially has society-altering consequences, but you are not seeing a platform shift here. You can argue that it is a technology shift. So essentially what it is, there’s a new technological capability which you can bake into existing products. 

 

So let’s say you’re Salesforce, right? You are quite a sprawling build a software right now. Can you create more automations to be able to pre-populate a lot of these fields that need to be filled by end users? So that’s sort of one way sort of AI can sort of remove bits of friction from existing bits of software. And that can happen across lots of different types of products. So in that sense, there’s definitely a lot of benefit that incumbents gain from this. 

 

The question is if you are creating a new company, what do you do? Creating a new company is very, very easy when there is a new platform shift because, hey, nobody’s built on browsers yet. So you can go build something on a browser. You can go build a mobile app. It’s an entirely new interaction paradigm. And so the products look and feel different. The form factor is different. None of that is happening right now. 

 

And so the question you need to ask yourself is, the way you build products does not actually change. Fundamentally, it is the same thing. What you have is a new piece of your tech stack. And so what sort of novel experiences can you build with this new piece of your tech stack? I think that’s the fundamental question you need to ask. The wrong rabbit hole to go down is, hey, everything is kind of going to happen through a chat interface. And I think over and over we’ve discovered that the chat interface is pretty high friction. 

 

And so it is a great replacement for things that already require lots of very high cognitive load activities. I need to go talk to someone. I need to send emails back and forth with someone. need to go do a lot of research. think chat replaces that very, very effectively. What it does not do is replace clicks effectively. And it should not. Clicks are taps. And so try not to do that. And instead, within the existing paradigm of building products with GUI, what can you do with this tech stack? And if you’re an incumbent, what bits of your user interface have the most friction and what can you remove with this text.

 

Shruthi Prakash 

Do you see today that let’s say startups or founders are currently at least overbuilding thinking it’s a platform shift that’s sort of happening and is that sort of avoidable? How have you seen it sort of take shape?

 

Sameer Singh

Oh absolutely. I mean, I can give you a couple of examples, right? So one of the areas that we see a lot of hype about is agentic commerce. The idea that people are going to be making transactions with AI. So increasingly, a lot of the pitches we see are, hey, here’s a way to book flights or book hotels with an LLM is you kind of search for, show me flights from San Francisco to London. 

 

And then it’ll ask you, hey, do you have a time preferences? Like, yeah, I want it to be from this time to that time. Do you have a date preference? Like, yes, from this time to that time. In case you haven’t put all that in the initial prompt already. Do you prefer a particular carrier? And then it’ll kind of, once you kind of go through all of that, it will give you a link at the end and you can click through it and go book it. 

 

And effectively all it’s done is replace search filters on a GUI with the chat message. That has just sort of increased, all it’s done is add friction to the user interface. Like that doesn’t really accomplish anything. And then you have to kind of remember that these systems are probabilistic. I there’s a of research going on into reinforcement learning to kind of work out if you can make them deterministic. At least as of now, I don’t see any line of sight to making them deterministic. 

 

And without that happening, automated commerce is, let’s just say, not very realistic and all you have left with is this sort of super high friction chat interface and so instead you’re going to be thinking about okay if you do have a chat interface – How is that different from a chat GPT or a Gemini? What can you not accomplish there? And two, what use cases is it suited for? Where an alternative mode would actually have very, very high cognitive load. 

 

And if you don’t have something like that, essentially you’ve got to work out, what problem are you solving? What sort of quote unquote product, the GUI-based interface would go solve that. And then what pieces of this tech stack, of the new tech stack can you leverage to get you there faster to create a great experience. It’s obviously a lot easier once you dive into specifics and talk about specific products and then it becomes easier to kind of unpack what makes sense, what doesn’t make sense.

 

Simone Cicero 

So typically, when you say you cannot really build applications on top of ChatGPT, right? How is that true? So there are a lot of people building applications on top of this technology. Maybe what I can see is that they are building these applications on top of this technology as a commodity essentially as a pervasive, how can I say, ubiquitous technology.

 

Sameer Singh 

Yeah, so I’ll put it this way. The difference is, let’s say, the difference is between, let’s say, using an application on an iPhone versus using iOS SDKs to build on any sort of web-based product. So when you’re. The idea of a platform shift is that one, the underlying product has ownership of the customers, of the users. So let’s say people are using ChatGPT as a product, not just GPT-4 or GPT-5, the technology. I’m talking about ChatGPT, the product. 

 

And they are not using applications through ChatGPT. People are not. Developers are not building applications for chat GPT users. What they are doing is creating a third party application that you can go access it via the web or the app store. And that product’s got sort of GPT-5 embedded in it. GPT-5 or you know, Geni 2.5 or clouds on it or whatever it might be.

 

Simone Cicero 

No, that’s a good point because to some extent I was thinking that that’s where the current product strategy of the various LLM providers is failing, you know because – what is the point? Context is extremely important in the interaction with LLMs, right? And so far there is no real portable context across applications. 

 

So, for example, if I use three applications that are powered by ChatGPT APIs, but they do not use my context as a user, which is into the ChatGPT OpenAI platform. So basically they are using some sort of white label, let’s say general, how can I say? Yeah, they’re using the API. So a blank state ChatGPT agent.

 

While maybe the value would be that if I can bring my context. So maybe imagine that if I log through an OpenAPI account, an open, sorry, an open AI account into a service that uses open AI platform, I can bring my context. And that would be building products on ChatGPT according to your thesis, more or less, right?

 

Sameer Singh 

I mean, there is a hypothesis, but when you kind of look at usage behavior, it’s kind of hard to make that argument sometimes. So I’ll give you an example, right? There was a report from OpenAI, I think this week, which basically kind of broke down how open AI, how chat GPT is used the biggest chunk of let’s say chat GPT usage is kind of asking for information, searching for information. 

 

Essentially things that are not really possible via Google or it’d be sort of too high friction to go do that. And so you’re trying to get a quick summary of information. Now, is that necessarily going to be very, very useful for an application that is generating images? Probably not. And so the challenge is like the use cases can be so fragmented. Like I’m not sure how much portability of context would actually have, maybe for some applications it would.

 

And for some applications that it won’t, sometimes people think it’s a bit of a home run and I’m not so sure that it is in fact a home run. And so this is where I come back to the idea of it is not just context. It is the fact that the product, the end product, like the iPhone, like Chrome, essentially owns the end user. And the access of the application is happening through that product. If you do not have that, it is very, very hard to make any argument for any kind of network effect, platform dynamics that actually end up happening. Without that, it’s basically hope.

 

Simone Cicero

All right.

 

Simone Cicero 

I mean, they tried with plugins, right?

 

Sameer Singh 

I mean, it did work, right? And one of the reasons it didn’t work is like the, let’s just say discovery of applications is kind of hard with the chat interface. Now they’re trying again with MCPs. We’ll kind of see how that goes. Like, frankly, I don’t see it being dramatically different.

 

Simone Cicero 

I mean, MCP is completely different by the way. I think plugins are really what more or less you were talking about in terms of having an app distribution.

 

Sameer Singh

Initially I was. Initially, they tried the GPT store. I mean, yeah, I mean, the reason I mentioned MCP is because right now they’re trying to automatically route to sort of different MCPs. And the whole idea is developers are building MCPs on the side. But again, there’s a lot of nuances and application and sort of problems there that sort of break the discovery floor, which really is central to any sort of platform dynamic.

 

Simone Cicero

But I think we have a little interesting point here that I see this friction with. You said, for example, it’s a real platform shift if the experience goes through the product. But at the same time, the fact that OpenAI and other players are trying to make this happen. So they want people to interact with their interface. They’re not creating portable context – which is at the end of the day, probably one of the enabling bricks of building products on top of AI. There is this tension between they don’t wanna create portable context because this is gonna enable experiencing their intelligence on other websites. But at the same time, it’s paradoxically the real piece that may somewhat let them become an enabling – a real platform, right, where you actually own part of the user experience. 

Because if I think about the context I gave to my AI, it’s a lot of context. I can see that, of course, I see your point in use case fragmentation, but I see that it could be useful in few applications.

 

Sameer Singh 

I mean, potentially, I mean, that’s certainly, you might say one building block, as you mentioned, but I think the bigger roadblock is just the fact that you have a chat interface and we know that the chat interface is inherently limiting. Until you kind of solve that roadblock, I’m not sure the other roadblocks, other building blocks matter as much. The biggest reason sort of plugins sort of failed is because in that interface, discovery is hard. So how do you get people to even know that they can use plugin X, Y, or Z?

 

Simone Cicero 

And what about the potential of, you know, there’s a lot of talking about, you know, AEO, so essentially SEO for LLMs. But I’m wondering, what do you see happening in terms of the potential for LLMs to actually become at least a distribution channels?

 

Sameer Singh 

I mean, complicated question, but I’ll put it this way, right? 

 

So there is certainly some traffic being generated via LLMs. I think we’ve seen it across some of our portfolio companies as well. A lot of companies are noticing it. So as an emerging channel, it certainly makes sense to optimize for it. I mean, we saw this with TikTok a while ago, where a certain amount of search was happening on TikTok and you sort of try and optimize sort of by building certain kind of videos to ensure that one certain search terms were sort of included in search box that your videos would show up in others. 

 

So I think there is definitely room for that. Then again, when you actually dive into the data that I’ve seen, the share of traffic from LLMs while it’s growing very, very fast is still very, very small. Like we’re talking low single digit percentage. And so that would have to be meaningfully into the double digits before it genuinely moves the needle on anything. And at least so far, like there might be a handful of products here and there where that’s happening, but at least across the board, I’m not seeing that happening yet. I’m not saying that can’t happen. Like if that does happen, then certainly the value of AEO goes up exponentially. But at least so far, it’s still a small share of traffic. I’d say it makes sense to optimize as much as you can, it probably should not be an outsized share of your efforts.

 

Simone Cicero

And you don’t see the possibility for advertising models to enter the picture. 

 

Sameer Singh

I mean, they could, they could, they could, you kind of have to work out what the right format would be. I think which is non-trivial and I people are sort of understating how challenging that can be. But I think that’s an inevitability because I mean, if you look at the economics of LLMs, you will have to gain some advertising share to make these economically viable.

 

It’s not an option. Now what that does to, you know, LLM marketing optimization, we don’t know because it depends on what the format is. We don’t know what the format is. The AEO in my mind is different. It’s not organically trying to get the LLM to mention your product versus one of

 

Simone Cicero (29:04.8)

Yeah. Yes, like SEO versus blue links, advertising links. 

 

Shruthi Prakash 

Yeah. Yeah. I was actually going to say from my company itself, I could see that around, I think 10% of all of my, let’s say, enquiries today come from like LLM searches. So that I found interesting. So I was curious, maybe if this is a very industry specific, let’s say, you know, challenge or difference or whatever. So maybe it could be that.

 

Sameer Singh

No worries.

 

Sameer Singh 

It could be. There’s definitely some variation. I’ve seen a couple of instances where that has been true. And so if you are, then definitely you should try to put more effort into AEO.

 

Shruthi Prakash 

Yeah, potentially niche industries is my, let’s say, assumption in terms of what could be, you know, sort of leading in that direction. 

 

Actually, I wanted to, you know, sort of touch upon this and one of the points that you brought in earlier one was that maybe all of this is, let’s say, getting more views, but potentially the action is limited. That is, let’s say people clicking on a certain action is limited. 

 

So, even though accessibility to information has increased drastically, maybe the intent action, things like that has not changed very highly from a consumer perspective. So ya, mean, for me, the question is then there is a lot of creation, generation, all of that that’s happening, but retention is obviously low, quality of the content, let’s say being produced is very low. 

 

So how do you tackle that? Should we, let’s say, you know, create frictions or create constraints. How does all of this sort of take place and how do you improve your quality of how you work with AI?

 

Sameer Singh

Are you talking about an LLM or you’re talking about an application developer that’s using sort of AI as part of their tech start?

 

Shruthi Prakash

I’m just talking about LLMs, yeah.

 

Sameer Singh

I mean, in LLMs, you’re sort of less in control, right? So what it generates is what it generates. And so as an end user, you’re putting in a prompt and hoping for the best. Now, they are getting much, much better. 

 

But I didn’t quite understand the question in the context of an LLM. how could…

 

Shruthi Prakash 

No, so I mean, I’m curious from the businesses side, like an organization side, right? Like how do they, they are generating, let’s say, lot of content through AI and other tools. So I’m curious, how can action increase? Because you were also speaking about how a lot of these platforms today are used to generate a lot of content, but essentially the action is relatively low from a consumer side.

 

So how can businesses, let’s say, improve the quality of how they work with AI? Yeah, that’s what I’m, yeah.

 

Sameer Singh

Okay, mean, at least when I was talking about, I was talking about more from an application standpoint that using AI as part of this tech stack. From an organization’s point, if you actually look at the behavioral data of ChatGPT, about two-thirds of writing  requests or content generation requests for via AI are for editing existing text as opposed to generating net new text. And I think that’s in general a much stronger use case for it. I mean, AI struggles with creativity, essentially what it’s doing is predicting the next word. And so there’s a baseline level of quality it’ll have, but the ceiling of it is also quite limited. So if you do have something meaningful that you as a human have written or generated, and using AI to edit it, that’s probably a much stronger use case of it. And the best developers are spoken to kind of use AI the same way in a coding context, in the sense of they don’t necessarily let it run wild they kind of give it snippets of text to kind of edit and improve and fix as opposed to if you kind of, know, vibe code an entire product and trying to figure out where it’s broken is much harder.

 

Simone Cicero 

Yeah, do you have a follow up question? I was reflecting in the background, know, because you seem to be skeptical on this AI frenzy.

 

Sameer Singh 

On the frenzy, yes. On the benefits it has less so.

 

Simone Cicero 

Yeah, because my question was really about the structure of benefits that you see, because you’ve been critical about the distribution capabilities. You’ve been critical about the UI impacts, UX impacts. So what are you, instead, what do you think is going to have an impact, especially on consumer and network effects-based businesses in your perspective? 

 

In the investees that you have invested in recently, what were the usage of especially GenAI. I’m really curious because that’s the radical shift now.

 

Sameer Singh 

Yeah, good question. like I mentioned, right? When you have a construct where a user is interacting with AI, you have very, let’s say, non-existent opportunity to create network effects. At the end of the day, a network effect is a unique structured multiplayer interaction between users. When you have that, you can create a network effect around it. When a user is interacting with AI, by definition, that is a single player interaction. The AI is not a person. It is not a user. It is software.

 

And so the only way to enable that unique structured multiplayer interaction is to work out, okay, what kind of interaction can you create between two or more users with AI under the hood? And this is where lots of novel experimentation is taking place. And again, keeping in mind that AI is probabilistic, there are use cases in entertainment, gaming, where having AI generated constructs is an interesting basis of an interaction. I’ve got two portfolio companies that I unfortunately can’t talk about in detail. One is at the intersection of AI shopping and gaming, and the other ones are the intersection of AI social and game. And they’re both enabled by GenAI under the hood.

 

And they use it in very different ways. One of them sort of uses image models. The other one essentially uses AI-enabled scoring of submissions and AI judges. And if you took AI away from both products, both products would not exist. 

 

And so that to me is a more interesting application of AI where you’re kind of staying away from the frenzy because when you’re interacting with AI it is a magical experience and sometimes people expect it to be a lot more than it is. Whereas when you kind of treat it like a novel piece of your tech stack and you just sort of think about, what problems and experiences can you create now that you have this piece of your tech stack? You’re generally far more free to open an experiment and come up with multiplayer interactions that make sense.

 

Simone Cicero 

Something like, for example, don’t know, music generation, it has a social, inherently social graph, fans versus creators, and it’s enabled by AI. Also something where AI is enabling maybe a new generation of creators to connect with their fans, something like that.

 

Sameer Singh 

I mean, that’s more what I’ve called AI 1.5, is basically the way that works is user A comes in, generates AI, creates AI generated content of some shape or form. In this case, that’s music and that gets shared with other users. It’s viable and you’re seeing certain some products that do that. 

 

The problem you run into is that in general, there’s not many you might say limitations on creation. So a lot of people go in and sort of create something and you have a very, very high volume of often low volume, low quality content.

 

Simone Cicero

So you lose that scarcity element that makes it vulnerable on the first stance.

 

Sameer Singh 

I mean, not only do you lose scarcity, you also lose interest because there’s so much of it and most of it is kind of blah, right? And so that affects retention on the consumer side as well. And so usually for products like this, I’m fairly insistent on having a good degree of creative friction on the creator side. So yes, you can generate stuff with AI, but don’t make it sort of one-click generation. That sort of one-click or one-sentence generation sort of defeats the purpose of creation.

 

Where it’s not that satisfying for the creator, it’s not that satisfying for the consumer. You need the creator to actually put some work into it. And you’ve got this tool that can help him along that journey, but don’t make it too easy. And so once you kind of add that creative friction in, then it of becomes more of a multiplayer interaction, which is why I always say the AI is part of your tech stack. Like, strip it out of your user interface as much as possible. The more you have it in your user interface, the more tempting it is to have one click generation users interacting with AI directly, which sort of breaks every form of network effect you can have.

 

Simone Cicero 

So you’re looking into something that is more like creating an entirely new type of interaction or like have a 10x impact on the type of interaction. When you say, for example, Gen.AI driven gaming, your creation of worlds or having non-human characters that can interact in the system, making it much more attractive for the gamers to join or something like that.

 

Sameer Singh 

I’m not even going that far, right? So think about Minecraft, let’s say. In Minecraft, you have a ton of combinations of products that are essentially manmade, right? So you just sort of combine this and this and you get this. There are probabilistic creations you could take. So the one example I take, just as a very simple example, this is not multiplayer, is a game called Infinite Craft.

 

It’s one of my portfolio company of mine showed this to me. And literally, it’s a very, very simple game. You open up the game and there’s like four elements on the screen. I remember correctly, it’s water, fire, earth and something. And so as you combine two of the elements on the screen, there’s a new AI generated element that pops up. That again, you can combine with any of the elements and you basically have infinite combinations. So that construct is certainly an AI generated construct that could be under the surface of of a game, right? And that’s not one example that I’m using. There’s probably multiple ways to implement this logic. And so again, if you have a way to combine things in infinite ways, because the output is always probabilistic, what can you create? What sort of multiplayer interaction can you create? I’m not a founder, I’m not that creative, but other founders are. 

 

And so when you point them in this direction, their minds go wild and they kind of start experimenting with stuff. And so that’s what I’m encouraging people to do, to stop thinking of this as a product. AI is not a product. It is a part, probably a critical part of your technology stack that you can use in very creative ways, at least from the point of view of a consumer product. It’s kind hard to see AI as product, a paradigm five years from now from my vantage point.

 

Simone Cicero 

I mean, I think your point is extremely clear and interesting, one of the points that we have raised in a completely different scope. Imagine organizations, right? So what we say these days is that as labor and time become less of a measure of value in organizations, organizations are prompted to think through what is the value that they want to create in the market and society more in general, right?

 

So similarly, your point is, OK, AI is not creating value for the user per se. You have to think of it, you have first to think about value, and then think about how AI can maybe make it possible to create value in a much faster way or a much more diverse way or enabling your creative thinking to maybe, I would say, gaps that before generative AI would have costed you lots of money because you would have needed a human actually to do some of the steps that connected the two users in the systems.

 

Sameer Singh 

Correct. the very basic example I use is someone’s invented a hammer, stop going and handing everyone a hammer and give people tables and chairs.

 

Simone Cicero

I wanted to ask you, so if you look at the companies you are looking into for investing, network effects, these new avenues of value that you have identified, behavioral changes, what are the characteristics, some of the common threads that you see in these companies that entrepreneurs should be thinking about or looking into?

 

Sameer Singh 

Do you mean sort properties of the products or traits of the founders?

 

Simone Cicero 

I think both, know, property of the products, traits of the funders, type of business models, distribution capabilities, I don’t know.

 

Sameer Singh

Okay. All right. I’ll start with a caveat that this is probably more specific to, again, consumer startups and sort of the area I invested. To start with the founders, missionaries, not mercenaries, is a pretty safe rule, which is the founder has some sort of irrational obsession with this problem, sort of borders on illogical at least to the average observer. And so that is super important to have. Without that, it’s very hard to keep working on the same problem for like seven, 10 years. Eventually you will give up. This is not easy stuff. 

 

Second, the founders have to be numerate. They don’t have to be data scientists, but they need to be able to think through numbers. Like when you’re selling to an organization or a business, you can go ask them what their problems are. When you have even 10,000 active users, you can’t ask people what their problems are because their problems are not necessarily consistent. 

 

And so first, you’ve got to start with behavior. What are you seeing in behavior? What positives and negatives are you seeing in behavior? Are you seeing drop-offs in a particular page on your onboarding screen? Are you seeing a drop in retention after a certain period? Are you seeing lower activation rates among a certain cohort? 

 

And then you kind of dive deeper, and eventually you kind of narrow down a small enough cohort of users that you can talk to. And so you kind of have to be numerate by definition. And you have to be very, very, very open to feedback. One of the challenges with building consumer products are consumers are notoriously fickle. 

 

The problem you thought of might have been valid, but the initial solution might not be. And so you’re gonna need to iterate until you get there. And you have to be very open to seeing what the market is, what users are telling you through their behavior. So that’s on the user side. 

 

Second, on the product side, if you’re building a marketplace, ensuring that you’re not doing direct sales on both sides is insanely important. That’s a big failure pattern. You need to ensure that you can have scalable acquisition on one side, which basically means one side of your marketplace behaves like a consumer. 

 

Either they kind of search for you online and they find you, or when you onboard one side, they invite the other side in. Or you kind of operate some sort of SaaS enabled marketplace where you initially just start selling just to one side of the marketplace and then you have loads of them using the product, then you can figure out how to onboard the other side. But trying to do direct sales on both sides is a bit of a problem. And so when you kind of look for those three factors in general that really winnows down the number of marketplaces that actually exist. And that’s part of the reason why marketplaces are such hard businesses relative to B2B SaaS where you can direct sell to one side and be just fine, right? 

 

If you’re building a consumer network, these are even more sciency than I think the average marketplace at least, where you have to be even more data-driven. And you kind of have to work out what your route to liquidity is because there are far more paths to go to liquidity here. The easiest one if you’re building a friend network is hey you sign up a user they invite their friends boom you’re done right so at least they’ve got two or three users using this together all the same time. Slack use that model Facebook less so Facebook maybe WhatsApp use that model Snapchat use that model so pretty popular.

 

The other sort of network you have is sort of a discovery based network where users sort of show up to not use it with their friends, but kind of find other things or people. And this is where things are infinitely harder because you have to have liquidity first before you onboard users. And so what you kind of have to do is create some sort of tool that users are willing to share the output of in some way, shape or form. So as an example, Instagram, like the first use case was creating filters right and those filters got shared on Facebook and they kind of had a bit of an Instagram logo which brought users to Instagram and so that – A) is a bit of a single player use case but also B) popular distribution so it’s called piggybacking. 

 

YouTube also did something very similar people don’t really see the parallels between Instagram and YouTube but it happened YouTube the earliest users were at least you know once they found success with was musicians that had Myspace pages and they had to embed they had to find a way to embed their music videos on their Myspace pages and that ended up being a YouTube embed which sort of populated YouTube in the early days. So those are networks. The implementation is always much harder and very, very specific to each network, but it’s always the same structure, so to speak. 

 

But that’s also why you find so few of them. So you will always find more ones that sort of rely on friend invitations versus others. So those are sort of the distribution paradigm. And then finally, behavior. Often, you kind of come across founders that say, hey, retention is not great now, it’s going to get better later. That never really happens. Usually retention is a function of the core problem the product solves. If it’s not a big enough problem, you’re not going to have retention. If it’s a big enough problem, you will have retention, especially if you’re going after the right kind of users. The important thing to keep in mind is kind of what, how you measure retention. Retention, you can measure it daily, can measure it weekly, you can measure it monthly. 

 

There’s no right frequency, there is only a right frequency for your product. Like for example, if you measure daily retention for Airbnb, it’ll probably be terrible. It’s not meant to be used daily, it’s meant to be used three times a year. And so you kind of figure out again, how frequently is this product meant to be used and then kind of create a retention measure around that that makes sense.

 

Simone Cicero 

I was thinking that I feel like from what you say, the focus on retention, the obsession, almost illogical obsession with the topic, you also said something, it’s hard to understanding if you’re resolving a problem with the customer. I think it tells us about the consumer market, which is somewhat saturated and you really need to be digging deep to understand where the value is in consumer at the moment, right? What are the pockets of value that are still there in consumer? Maybe before we move into the breadcrumbs, you can add, I don’t know, two or three spaces where you see, you mentioned gaming.

 

You mentioned social interactions in general with, you know, before you mentioned, you know, social finance or social utility, anything that you want to say a bit more specifically about what pockets of industries are being interesting at the moment.

 

Sameer Singh 

I mean, be honest, if I had the answer to that question, I would be a founder. The fun part of my job is I don’t need to have an answer to that question. I will say, consumer always looks saturated until someone finds a problem that is interesting. Historically, you look across the last 10, 15 years, it’s always seemed like consumer is either saturated or the new thing is just a toy. So really what you’re trying to figure out is what is the next thing that looks like a toy that might not be one. 

 

Simone Cicero 

Right. mean, maybe I’m biased because I remember when I started my career in platforms and ecosystems, was the age of the social sharing economy. Right. So everybody was doing Uber for X and that seems more, seemed more easy note to find.

 

Sameer Singh

Well, now it’s Cursor For “X”. it’s a… Now people say it’s Cursor For “X”. It’s something slightly different. They’re easier to build, but there’s like 20 competitors for the same thing. Yeah, and with Uber for “X”, the problem was it was very hard to get liquidity for a lot of these super local marketplaces.

 

And so a lot of those things wouldn’t get off the ground. Now the problem is a lot of these things do get off the ground, but there’s like 20 others on the same thing. So like, how do you, how do you stand out and how do you differentiate in the markets? You’re sort of always going to go back and forth in these things.

 

Simone Cicero 

And this brings to the internet theory, So much content that nobody cares about and is made by agents and that’s it.

 

Shruthi Prakash 

I mean, yeah, I was just, I think one of the points that Simone was speaking about earlier as well. think for me, interesting is how participative I guess, consumers are. think that’s something that stands out for me in this sort of, let’s say era, let’s call it. So I’m curious how that amount of vast user data, quick feedback loops, things like that can make a company’s, let’s say product or offering stronger. So that’s something I’m interested to see. 

 

Sameer Singh 

I mean, that’s definitely true. You consistently see the companies that make the most progress irrespective of all other factors being the same. the consumer tend to be ones that learn from data the fastest. So like I have two different examples of companies. One that sort of is doing well, one that did not do well is where they were experimenting with some landing pages and one company kind of figured out that, something on the landing page wasn’t working.

 

So within a week, they had a new one. They had some early data on it, it be working and it was up and running and swapped out. And the other one took about four weeks to hire an agency to redo their landing page. It was like faster iteration cycles based on data will always take you much further.

 

Shruthi Prakash 

Awesome. So firstly, thank you. Thanks, Sameer. I think we had some really interesting opinions and thoughts shared, which are rather different from the industry. So we always appreciate that. So towards the end of the podcast, we have this section called as the breadcrumbs where you could share maybe some podcasts, books, anything that you take inspiration from that our listeners can also sort of look into.

 

Sameer Singh 

I’m more of a book person than a podcast person. So I’ll start with a book recommendation. I think the book, Technology, Revolutions and Financial Capital by Carlotta Perez probably is one of the most consequential ones for the current era we’re living through. Like if you want to understand AI, and this book was written like 25 years ago, it’s probably one of the best reads you can. 

 

Now the caveat is, like it’s, the book is kind of written like a textbook. So it’s not necessarily the most engaging read, but it’s a very important one. So I’ll always give that caveat for this book. That’s one. 

 

If you want to understand how startups work, or rather sort of what makes interesting startups. This book called Pattern Breakers by Mike Maples. Who was a floodgate founder is a very interesting one. He sort of breaks down a lot of the key elements which especially for folks that are not up to speed with this world is probably a very very informative read and a blog or rather a set of blogs that I will refer to is – everyone should read Bill Gurley’s blog. 

 

He wrote many of these posts 15 20 years ago. All revenue is not created equal is an important one again for the current era. It’s funny how many old books and blog posts are applicable right now. Whenever we are in cycles, I always refer back to old books. There are universal truths that you can always rely on. You just need to work out what they are from first principles. And some of these books can help you understand what those first principles are. And so with that I will always tend to interact.

 

Simone Cicero 

Thank you. I mean, as she said, I mean, Carlotta has been one of our target guests for ages now. She didn’t confirm it yet, but we always keep our dream to get her on the podcast. But first of all, thank you so much. I mean, it’s been very dense. I think a few ideas we discussed are really, really important and generative. You know, I don’t want to make the list, but.

 

I think I need to really listen to the podcast. Our listeners probably will have to do the same. So thank you for your time. I hope you also enjoyed the conversation.

 

Sameer Singh

Yeah, always fun. I mean, talking about network effects is not work to me.

 

Simone Cicero 

Right, right, right. And I mean, for our listeners, of course, you will find on our website all the details, the notes, all the books that Sameer has suggested, some other pieces that maybe resonate with the conversation we had today. So go to www.boundaryless.io/resources/podcast. You will see the podcast there.

 

And of course, until we speak again, remember to think Boundaryless.