A look at Systems of Intelligence, barriers to entry, and enterprise VC investing in the post-SaaS world
Special thanks to Lenny Pruss, Tomasz Tunguz, Eze Vidra, Eric Wiesen, Max Claussen, and Ron Yachini for reviewing early drafts of this post.
What follows is a pretty long post, so here’s the TL;DR in seven points:
The “Systems of Intelligence” thesis captures a lot of what is happening in enterprise software
In assessing any meta-theme, barriers to entry is perhaps the critical issue to understand — as no company can build massive value without them
We’re starting to come out of the era of “Network Effects” investing, where barriers to entry were built into the thesis.
Because network effects made barriers to entry relatively easy to establish,VCs and founders have become a bit lazy when thinking about barriers to entry.
We are in the midst of a “technology value inversion,” in which a lot of the technologies that once created value for new entrants (like SaaS or AI) will potentially destroy value for new(er) entrants because they have been fully adopted and commoditized
“Systems of Intelligence” companies will create massive value, but it will be challenging for “Systems of Intelligence” companies to establish barriers to entry.
At the end of the post, I argue that barriers to entry for System of Intelligence companies will involve: domain expertise, complexity, hybrid machine-human architecture, networks where possible, hard tech where applicable, and data in some cases.
To summarize: In assessing the value of a “Systems of Intelligence” company, we’ve got to focus on the barriers to entry created by the specific system that company is creating. The value will not come from the “intelligence” alone. The value is in the system itself.
Here’s the full post:
This weekend, I saw a tweet by Semil Shah that pointed me at a recent blog post by Jerry Chen, a partner at Greylock. In the post, Jerry lays out a thesis that “Systems of Intelligence” are the next defensible business model — and provides the beginning of a framework for thinking about (1) why this transition is happening and (2) what to make of it.
Like Semil, I love Jerry’s post, and I highly encourage everyone to read it.
The idea of “Systems of Intelligence” captures a few things that I’ve been thinking about and struggling to articulate for a while. In this post, I want to try to use this idea as a jumping off point for a few observations on enterprise software investing and technology investing more generally.
Part I: Never forget that barriers to entry are the critical driver of venture returns
In thinking about VC meta-theses, its important to keep barriers of entry in mind. Venture-backed companies must erect barriers to entry in order to create value. This is pretty important, often overlooked, and can set the stage for thinking about meta-theses.
The higher a company’s barriers to entry, the greater its ability to charge high prices and generate a profit. Over time, profit drives cash flow, and cash flow drives value.
There are a lot of things that are critically important in determining the quality of an investment and the likely outcome (team, market, competition, execution, etc.). But in actually developing investment theses around specific sectors or verticals, VCs tend to emphasize technology. We do this becausetechnology is often assumed to be a barrier to entry. And for most of history, that was true.
When you go study the origins of the venture business in the 1960s and 1970s (back when Silicon Valley was really about Silicon),venture capital was really about technology. VC emerged as an “asset class” because holders of capital wanted an efficient and effective way to take a ton of risk to finance deeply innovative companies that had a reasonable shot at erecting really meaningful barriers to entry in large markets.
In short, if you were to sum up the venture capital mindset from 1960 to 2000:
Projected future cash flows are a proxy for valuation
Ability to generate profit is a proxy for projected future cash flow
Barriers to entry are a proxy for the ability to generate profits
Technology is a proxy for barriers to entry
Hardware and software are proxies for “technology”
Therefore, if you had enough “technology” (whatever that meant), you probably could create enough real value to create a “VC style” return. Or so the theory went. And that is why VC and “technology” are so closely tied up.It’s obvious, perhaps, but it’s important for the rest of the story we are building here.
Part II: “Network Effects” was the ideal VC thesis, because barriers to entry were built in — but they weren’t technological.
To really understand what the “Systems of Intelligence” thesis means for VCs, we need to contextualize it. And to do that, we need to go back in time a bit to take a hard look at the era of “network effect” investing.
Not through “technology” alone. Around 2000, it became clear that “technology” wasn’t the only way to build really meaningful barriers to entry.By 2011, Brad Burham famously tweeted USV’s brilliant thesis in 140 characters: “Invest in large networks of engaged users, differentiated by user experience, and defensible though network effects.” This captured the zeitgeist of the past two decades extremely well. From a VC perspective, if the last two decades of the last century were focused on building physical and logical networks (eg. Cisco, Netscape, AOL), the first two decades of the current century were focused on building companies that constructed very valuable network effects on top of those physical and logical networks (Yahoo, Amazon, Twitter).
The “network effect” thesis so perfectly expressed by Brad Burnham’s tweet is probably the best VC thesis ever. Why? Because one of the side effects of a business with a “network effect” is that as that business grows it becomes harder for competitors to compete: rate of growth is positively correlated with size. In theory, if you are the largest peer-to-peer hospitality network (e.g. AirBnB) you are therefore probably the peer-to-peer hospitality network that is growing the fastest. And therefore, your competitors will (at least in theory) never catch you, because at any given moment in time, they will be by definition growing slower than you. For a VC, that’s nirvana because it means that once you bet on a winner, the upside is uncapped and the winner’s status unchallenged. It’s “winner take all” on steroids. Normally, the bigger you get the slower you grow (law of large numbers). But with “network effect” businesses, the bigger you get the faster you grow and themore of a monopoly you become. Brilliant, right? And, as far as anyone can tell, these dynamics seem to have been borne out in the big winners of the “network effect” era.
The end of “network effects?” I would argue that we are gradually movingout of the “network effect” era and into a new era: the era of “Systems of Intelligence” investing. Opportunities to invest in “network effect” business will always exist — but our focus as enterprise technology VCs is gradually shifting to encompass “Systems of Intelligence” as the next big VC investment opportunity. (This doesn’t mean that network effects are dead. Far from it. Just that they are not the driver of the investment thesis.)
To put this in historical context, one could break down the history of tech VC investing into a series of major eras:
1970–1985: The “Silicon” Era (eg. Intel, founded 1968)
1975–1990: The “Information” Era (eg. Microsoft, founded 1975, and Oracle, founded 1977)
1985–2000: The “Physical Network” Era (eg. Cisco, founded 1984)
1995–2010: The “Logical Network” Era (eg. Netscape, founded 1994)
2000–2015: The “SaaS” Era (eg. Salesforce, founded 1999)
2005–2020: The “Network Effect” Era (eg. Facebook, founded 2004 and AirBnB founded 2008)
2015–2030: The “System of Intelligence” Era?
I think “system of intelligence” is potentially as big a deal as “network effect” because it’s a phrase that has the potential to define the era we’re in for some time to come.
Part III: “Systems of Intelligence” is a great meta-thesis, but the barriers to entry question is more complex.
“Systems of Intelligence” is a compelling meta-thesis because it couples the tremendous breakthroughs we are seeing in terms of data, analytics, and machine intelligence with a framework for how value is created/captured in the enterprise. Advances in artificial intelligence (AI) are profound enough that “Systems of Intelligence” will take its place alongside “systems of record” and “systems of engagement” as a critical product category — one that will lead to some huge venture outcomes. As an enterprise software VC, my focus is going to be looking for those opportunities — for companies that have the opportunity to build a “system of intelligence” profound enough to drive great value. As a meta-thesis, it’s a powerful one.
The challenge, however, with “Systems of Intelligence” as a meta-thesis is that — unlike “network effects” — the barrier to entry dynamics are not built into the thesis. With “network effects,” the network effect itself was creating the barrier to entry. But with “Systems of Intelligence,” it’s not that simple.Artificial intelligence is not the barrier to entry today that it might have been even a few years ago. We are in the midst of a boom in AI talent, technique, and technologies. Ability to deploy, configure, tune, and run a massive AI application at scale is no longer much of a meaningful competitive advantage.If anything, the proliferation of AI technology makes “Systems of Intelligence” less defensible over time.
In “Systems of Intelligence” we have a cup that is only half full: We have a great compelling meta-thesis that will drive huge outcomes, but we are still missing a built-in barrier to entry. This meta-thesis tells a great story about how value can be created, but it doesn’t yet tell us much about how that value can be defended. And unless we can figure out how value can be defended, we’ve only made it half way.
Part IV: We are in the midst of a technology value inversion. Someday soon, cloud/SaaS/AI technologies will be less about erecting barriers to entry and more about tearing them down. In fact, we might already be there.
Before drilling down into how to think about barriers to entry in the world of “Systems of Intelligence,” let’s take a look at why those barriers aren’t intrinsic to the meta-thesis itself.
In my mind, one of the most profound shifts the software world is going through right now has to do with the nature of technology itself: We’ve gonefrom a world where technology created value to a world where technology on its own often does not create value and — in many cases — reduces value.
Let me quote Jerry here:
Why does it feel like there are “no more moats” to build? In an era of cloud and open source, deep technology attacking hard problems is becoming a shallower moat.The use of open source is making it harder to monetize technology advances while the use of cloud to deliver technology is moving defensibility to different parts of the product. Companies that focus too much on technology without putting it in context of a customer problem will becaught between a rock and a hard place — or as I like to say, “between open source and a cloud place.”
This is pretty serious stuff. In the previous eras of software investing — theability to build the software was in and of itself a barrier to entry. But in today’s world, value is being squeezed out of the “technology” part of the software stack. On one hand, the proliferation of open source code makes advanced software both easier to build and harder to monetize. On the other hand, the easy availability and mass adoption of cloud-based (SaaS) technology makes advanced software systems so much easier/cheaper/faster to build that “value” is rapidly bleeding out of the software stack. Yes, software is eating the world, but software’s very ubiquity is starting tothreaten the ability to extract value from software. In other words, the ability to write and deploy code is no longer a core value driver.
Technology value inversion. I’ve been struggling with what to call this, but let’s refer to it as “technology value inversion.” The idea here is that as soon as a set of technologies (“cloud,” for example, or “SaaS” or, heaven forfend, “AI”) becomes ubiquitous enough, the ubiquity of that technology is more likely to destroy economic value than to create it. This happens constantly — it is anongoing characteristic of technology that something that was novel yesterday is ubiquitous today and will lead to value destruction tomorrow.
From a VC point of view, however, we’re probably pretty close to a very profound inflection point: The massive revolution in cloud/SaaS that powered so much value creation over the past decade is going to start destroying value instead. The cloud/SaaS revolution has been nearly completely successful.There is virtually no area of software that hasn’t been conquered by cloud/SaaS. Over the next decade, the easy availability of cloud/SaaS technology — the fact that these formerly magical things are commoditized table stakes — is going to instead suck value out of companies. And it’s not just SaaS. The same is going to be true of microservices and AI itself in fairly short order.
Part V: We are going to need a framework for assessing Systems of Intelligence. What would that look like?
On one hand, we’ve now got a useful descriptive name for the coming era in software value creation: “Systems of Intelligence.” On the other hand, a massive technology value inversion is underway that is stripping down the barriers to entry around “Systems of Intelligence.” What, then, is a VC to do?
First off, let me say that I don’t know the answer. I do, however, think that we can focus on the right question: when it comes to “Systems of Intelligence,” what constitutes a meaningful barrier to entry?
Let me start by ruling out a few false barriers to entry:
The ability to run a large cloud/SaaS/on-prem application at scale.Total commodity.
Killer user experience. Sorry. This was a barrier to entry in 2010, but not anymore. It’s hard, but not hard enough to create massive value.
AI chops. Absolutely not. This stuff is getting commoditized at a rapid clip — and the more value migrates to “deep” learning, the less of a meaningful barrier this is likely to be due to data portability (subject for another post).
So what might real barriers to entry look like in a “system of intelligence?”
(Vertical) domain expertise. There are certain domains that are specific enough that human capital and expertise can be a meaningful barrier to entry. Systems built by true experts can potentially maintain an edge over systems built by just anyone.
System complexity. Over the years, I’ve come to understand that — when it comes to enterprise software — the productization of complexity can be a very meaningful barrier to entry. Enterprises are complex. Their data and workflows are complex. Edge cases can be significant and critical. To be a true viable “system of intelligence,” a system would have to be able to accommodate this complexity without devolving into a customized professional services project. That’s not easy to do.
The ability to go truly hybrid. For many applications, the best outcome will be achieved through a mix of human and machine intelligence. For a system of intelligence to truly shine, it will need to enable and capture both types of intelligence: powerful machine algorithms and the specific wisdom and experience of the right human employee at the right time.This requires deep and sophisticated enterprise workflow integration.Building such a system is very complicated and can become a sort of barrier.
“Systems of Network Intelligence.” This a fusion of the original “network effect” thesis and the “Systems of Intelligence” thesis. It’s possible to imagine a “system of intelligence” that works across various parties in a supply chain. When a system of intelligence creates incremental value by sharing intelligence across customers, it can be said to benefit from a network effect. Those opportunities are going to be few and far between — but when and if they emerge, they are going to be extremely valuable.
Hard technology. There are always going to be areas of software that are on the technology frontier. Image/video processing is currently one of them, for example. In many cases, this will be niche opportunities. But some of those might be very large.
Data. Some argue that data creates network effect. This is only partially true. Data is increasingly portable, so the fact that a company accessed multiple customer data sources and might have even created new data sources is less of a barrier than it might appear. As long as that data is owned by the customer, that data can be exported and imported freely.Sometimes, data is a moat, but sometimes it’s a gaping hole in the castle wall. I would argue that truly proprietary data is a barrier to entry, but is rarely sighted in the wild.
The bottom line here is that in assessing the value of a “Systems of Intelligence” company, we’ve got to focus on the barriers to entry created by the specific system that company is creating. The value will not come from the “intelligence” alone. The value is in thesystem itself.
One final note on technology. I love the “Systems of Intelligence” meta-thesis. It describes a big part of what’s going on in enterprise software. But ultimately “barriers to entry” are what really defines a company’s opportunity to create and maintain value. At the end of the day, VCs (and founders) need to focus on great teams, operating in huge markets, and with a clear plan to build barriers to entry. I think that over the next decade many of these opportunities will turn out to be “Systems of Intelligence.” But many will not be. At its core, venture is about finding unique opportunities to build barriers. The ability to erect those barriers is the true definition of technology. It’s not “technology” just because it’s software (or SaaS or AI). It’s technology if no one else can do it as well as you. If you’ve got a plan to build value behind some barriers — let’s talk…whether you are building a “System of Intelligence” or not.
Special thanks to Lenny Pruss, Tomasz Tunguz, Eze Vidra, Eric Wiesen, Max Claussen, and Ron Yachini for reviewing early drafts of this post.