a16z Podcast - a16z Podcast: The Rise of Full Stack Startups

Episode Date: March 27, 2014

Suppose you develop a new technology that is valuable to some industry. The old approach was to sell or license your technology to the existing companies in that industry. The new approach is to build a complete, end-to-end product or service that bypasses existing companies. Andreessen Horowitz’s Chris Dixon, Balaji Srinivasan and Benedict Evans discuss the reasons behind, and advantages of, going “full stack.”

Hi, this is Chris Dixon. This is the A16Z podcast. I'm here with Benedict Evans and Bologi Shrinivossin. So we've talked about this concept a lot internally. We call it full-stack startups. I wrote a blog post about it recently. I guess just to get out of the way because some people say, what is this isn't just another fancy word for vertical integration. Let me define what I mean by these, which are basically, there's sort of a new, I don't know, trend in startups to have a kind of a core insight that that's around technology. But then to build a company that does multiple activities beyond just that technology, not just sort of, you know, so instead of building software and then going out and selling that

software, building a complete service around it, that includes things that we would traditionally not call technology functions in order to go to market. So I'll give some example. So, you know, Uber, I think is a good example. where prior to Uber, there were actually companies out there that built software that they tried to license to taxi companies that would help do automated dispatching and things. For a variety of reasons, those companies were not technology forward thinking. You were selling them something that they was going to replace part of their staff, maybe even the people buying it. They didn't catch on.

And then you have someone like Uber come along who does it sort of in this full stack way. They actually, you know, pay the drivers and do all these things we don't think of it's technology companies. you know, Tesla building a full car versus a lot of companies before that tried to just build electric batteries and sell them. You know, you take Netflix. They're now, you know, both funding content, movie development. They have a website where you can discover it and then they also, you know, and you can watch it. And they're sort of providing the whole stack there, whereas prior to them, there were lots of companies that tried to do things like streaming services that they licensed to, you know, Hollywood distributors or something, or recommend

services for Hollywood websites or something, so Netflix going kind of full stack. Another example might be BuzzFeed. BuzzFeed actually began as a company that was trying to provide services for media properties to be more viral and then figured out that actually it would be better to apply it to their own media property and build that up. So just a variety, we recently announced our investment in Alt School, which is an attempt to kind of, instead of taking technology and selling it to other schools, actually build a new school using that technology.

So I call these full-stack startups. I blogged about it. A lot of people said, isn't that vertical integration? I think vertical integration is a term. I think you could argue this. Maybe it's a boring point. The reason, one, I think vertical integration is a term that maybe is overloaded, and most people's minds refers to kind of non-technology companies that might go and

acquire their suppliers or buyers, so, you know, standard oil or something like this. I think it's, you know, I think my mind, that's what it implies to people that it's good to have a new term. Number two, I think some of these functions that these full-stack companies are doing aren't really sort of strictly the buyer or supplier. It's sort of more almost like horizontal integration in some cases. So I thought a new term was warranted. Maybe it's not.

But regardless, I think it's a new phenomenon. Yeah, so briefly on that point just for a second. So I think there's these two reasons why I think the term full-stack is interesting. So one is that it is sort of in keeping with just how people think about software, right? You've got a full-stack developer who can do the, the front end, the back end, and everything in between. And the interesting thing about software companies in particular is that they have the option to build rather than buy many more things than previous kinds of companies.

That is say, many more things are within their core competency. As software eats a world, acquiring a new skill is learning a new API, or being able to develop a new thing is learning a new API frequently. And so they have the option to do more of the stack. That's kind of point number one. And point number two is that also means that they would grow into these other layers. by building more than acquisition and more than the earlier thing. So that's kind of, I think, one of the reasons this is interesting.

The third is that after kind of a decade in which every layer of the stack, the computing stack, was sort of split off. And so you have Cloudflare and you have AWS and you have all of these services which kind of are based on a fair degree of maturity, you know, of, for example, X-86 and ARM architecture and so on and so forth. there's some degree of maturity of those individual pieces, so you can, in theory, focus on a very thin slice and just work on that. But when you're doing a real-world application, it's almost the opposite, right? So you actually want to be very integrative. And I think extremely importantly,

when you're going into a new industry where software has not actually been present for a while, often it is because there's a few different parts that need to get, or a few different layers that need to be replaced at the same time. Now, with that said, one thing that I saw... It's interesting, yeah. So you're saying it sort of the... So 20 years ago, if you wanted to build a you know, Amazon had to go build their, well, not Amazon. Let's say whatever, random startup had to go build their own Kolo or their old data center and do the whole thing. And then as each of the tools matured, you got, and you had good APIs, you could build

a thinner slice. And now we're sort of seeing the same pattern repeat in the world of Adams. That's exactly right. And I think that then again in like 10 or 15 years, you'll be able to build a logistics company on top of Uber or Lyft for logistics or something like that, because those things will also be produced APIs. But I think when you first move into a new vertical, there's so much stuff that is broken. There's reasons that software is not yet eaten that vertical.

And you often need to go after a few different components at the same time. So with that said, one of the things that I noticed in some like the comments, so I all said this, like, you know, long Twitter series. You had a P mark a tweet storm. 41 tweets. But, you know, the reason I like actually those like multi-part, you know, tweets is that... You should just blog. I don't understand.

Yeah, yeah, well, so I would blog, except first you can tweet, like, in the car, and so you can tweet partially, you know, you can tweet in the house, you can tweet like a mouse, like a Dr. Seuss kind of thing. But second, you can also just cut it off at any time. Okay, okay. So that's like, you know, a blog, you have to edit it more. But anyway, they're all drafts of like MOOC lectures. But regarding the kind of some of the responses to the Twitter post, I saw a lot of people who are saying, oh, you know, the cool enough, yes, I want to build a uranium mine. Now you were telling me, yes, we can go in fund it. I'm not against, like, extremely ambitious initial things, but I want to clarify that

doing a full-stack startup does not mean, like, take Uber, for example. Yes, they did the full stack, but they didn't buy 100,000 cars on day one, right? They, you know, they figured out, okay, here is an ultimate ambition. Maybe they weren't even thinking about 100,000 cars at the beginning. But they picked the minimum number of layers to replace at the beginning. They had the ambition of doing the whole thing ultimately. I think that's really what distinguishes it. You still want to have a minimum viable product.

Maybe it's not quite as minimum, but you want to have a plan for getting to those other layers. And in the interim, you use the stop gaps. You use the things that are commercial off the shelf, use the things that are best-practicing. Some of these things like Tesla versus A123 batteries or something, like the Tesla, you just have to, the minimum viable product is a really nice car.

And it's not, and it's pretty ambitious. But even then you can stage it. You can say, hey, look, we're going to show the engine works. And they start at the high end for like the niche. wealthy people or something. Presumably they're going to have a very mainstream inexpensive car at some point. Like, you know, even starting with, I'm sure

you know, I didn't see Tesla's, you know, roadmap and so on and so over. But I'm sure that there was staging where they said, okay, first let's prove out superchargers in California, right? And let's do it a network in California. But actually, it's even beyond that because, you know, they started the first car, they didn't make half the compents.

So there's been a steady road out. Yes. Okay. You know, they start with like, well, let's do a roadster and we'll buy in half the bits. and then you kind of step slowly through and you know there's an end point somewhere in Alan Musk's mind that you know he's making as many cars as Toyota but you know how do you get there

I mean there's a sort of in a sense this isn't really unique to technology and there's always been kind of asset heavy and asset light models in here you know obviously Uber doesn't own the cars and it's interesting to look at they're not buying cars but they're now thinking about leasing but it's interesting looking at Airbnb and comparing

that with say a big hotel chain like you know a Marit or Hilton because they don't own the hotels. You know, they've got, you know, dozens or hundreds of franchisees who've gone out, who are individual real estate developers who've gone out and raised some money. And so there's always that kind of tension of, well, how far into that do you actually need to go? If you look at Apple, we would think of Apple as being the ultimate integrated company, but they don't own any of the factories. I mean, you know, they own a few things. But, you know, most of the places where iPhones are made, they don't own, and they put capital in all they own. Well, I think, and to

it's an apple where do you where do you draw the ownership line where do you put the inventory where is it held how what can you do you know back to apple you know apple is the world's biggest sap installation so that that degree of of um disaggregation of manufacturing that apple is an internal function of software yeah so and you're right so the apple is clearly the the canonical full stack technology company the um i think the way the the principle they use to decide which parts of the chain they have to own and which parts they and outsource is just ultimate product, what ultimate product experience I need to deliver and which of the, you know, which of the layers, Tobology's point, are mature enough that I can use their quote API. So maybe the China factories, they decide

the API is mature enough. But, you know, for the, clearly for the main core processor, that wasn't mature enough. Yeah, they brought that in. So there's kind of a bundling and unbundling thing going on. I mean, the thing that you see in media and telecoms is, you know, clearly an unbundling of things

that were previously bundled that's enabled by the internet and enabled by smartphones and so you have um you know magazines and newspapers so you know on the one hand or you know data and SMS and voice on the other getting pulled apart into the component pieces one of the things i think that's underappreciated about um full-sex startups in particular is so why do you need them to move into a new vertical often not always but often um Airbnb for example is moving into new vertical without doing the whole thing but they're doing probably more of it than um other web competitors have have gone to, for example, they sent photographers to people's houses and they photographed them. And they were a lot more invasive about that than people were before.

And that was an enormous component of making the listings look attractive and so on and so forth. And now they're getting even more invasive or maybe invasive is not the right term. Let's say they're being more considerate of the full experience by thinking about it as a hospitality experience. They're giving guidelines for their hosts and they're saying, you know, here's the cleaning staff and we'll provide cleaning services and so on and so forth. So, you know, there's, I think the logic of providing a good customer. experience often means that full stack. And one of the reasons for that is that the more layers

you can operate, you can propagate feedback for one layer to another. So I'll give two examples I think of a lot. One is, so the development of like the MagSafe, you know, cable, as I understand it, because Apple controlled the Genius Bar, they had people who had come in and they would be able to log all the defects that happened with Apple computers, right? And they found that a relatively small number of defects, but with a very high dollar value, were caused by people yanking the computer off the desk. And so they propagated this from the retail level, right, the like highest level of the business, all the way down to the very lowest level of the business, i.e. solid state physics, right, to help a new device, right, to then sell and propagate all the way back up to

this layer that sensed it, right? As another example, so when running counsel, the fact that we controlled the entire stack all the way from mixing reagents and procuring reagents and doing that ourselves, up through robotics and sequencing and so on so forth, up through the national sales force and like clinical integration and so on meant that you know for example when doctors asked us for a turnaround time improvement right because people needed to have a baby you know rapidly um if it's like an IVF you know cycle or something like that uh we could go and we could you know take certain reactions and put them into triplicate rather than cereal and so we could optimize things at the reagent level to fix things that were at the patient level and so like controlling all the layers like that i think is

very helpful and in the uh a new business um the one challenge in doing it is is you need managers of these different layers that are very competent. And, you know, initially as an executive, you need to actually be able to articulate that you need a robotics expert when nobody in the industry has hired a robotics expert in quite this way before, right? So I think, like, I think of recruiting is one of the most difficult challenges associated with building a full stack startup because you need to have expertise. You need to, you need to know what you lack expertise at.

Well, that's, the other interesting thing, though, is if you can do it, it's very, very hard to compete against because you've got to do sort of this, like, it's this and function of like this and this and this like you know you want to go head to head with you know Apple like I guess you can take the radically different approach like Google which is you try to break the stack up in different ways with Android but in general like you got to be good at branding you got to be good at semiconductors you got to be good at operating systems you got to be good at apps like there's a lot of things and it really is uh you know goes against the whole kind of business school theory of like have a single core competency and do

it well and Google is arguably pretty full stack themselves. I mean, they build their own data centers, they build their own hardware and chips and routers and all this type stuff. I was going to say, I mean, looking specifically at startups isn't part of this a function of the collapse in cost of actually getting something out of the door. So, you know, you could look at Airbnb or Uber or, you know,

any of these companies and say, well, 10 years ago, it would have taken your $100 million just to get the app out of the door. Today, you can get that out of the door for five. And so actually the amount of capital that you can deploy to spread out in all directions is completely different.

And in parallel with that, you know, just the size of the addressable market now means that it's actually feasible to do something like Airbnb. You know, you could not have done Airbnb 15 years ago. You couldn't have done 15 years ago. Well, and it also affects like our job as VC's. You know, I think people like us are willing to, because the markets are so huge, you know, write these $100 million checks and knowing these can be, you know, many tens of billions of dollars in value companies. Yeah, exactly. Whereas if you try to do Airbnb. be 10 years ago, you'd have got to, you've got an in a tiny addressable market, and it

would have cost you 10 times as much to get to the product that you see today. So you just couldn't have done it. So the next question is, where do we think the next wave of, like, what industries will the next wave of full stock startups to be? I'll just give you my bias view, which is, or my bias, which is, so I think the best lens to look at this through is the Carlotta Perez lens of, basically she divides every technology revolution into two phases, the installation phase, which always corresponds with the financial bubble. there's always a crash and there's like a long period of the deployment phase where you're kind of then taking the fruits of that technology and deploying it throughout the economy so the automobile

in the automobile revolution the first phase happened in Detroit it was around building better cars the second phase happened over the next 50 years which is about building the highway system and suburbs and trucking systems and logistics and fast food and best big box retail and all these other things that happen as a result and happened by the way all over the place in different industries. And to me, what I think of full-sac startups is kind of the, what we've learned to be one of the optimal deployment phase strategies. Now that we're going into healthcare, healthcare is a very, very complicated industry where you could argue a lot of parts of it have misaligned incentives, have people who are versed to technology. In many cases, you know, one of the

trickiest parts is it's one thing to have a breakthrough. You invent a new, there exist expert systems that can diagnose illness better than any doctor, and yet they're never used. Why? Because what are you going to do? Sell it to doctors to replace themselves? Like, there's all sorts of technologies just aren't out there now. And I think the full stack approach is the answer. Yeah. I think in particular with healthcare, right, it's an industry where you have third party payment and fourth party pricing and fifth party regulation in the sense that pricing is set by CPT and regulation is by a number of bodies and, you know, payment is by insurers. And so you have like when you're talking to the doctor there's actually like three other people maybe

you know four in that in that room with you effectively right doctor doesn't know the price you don't know the price nobody knows the price of anything and so going full stack like something like kaiser for example actually does start to get aligned incentives because they aren't just paying for it they're also doing the treatment and so they start to actually you know close a loop on this um i would say that three industries in particular that are very good candidates for um you know full-stack startups are finance, education, health care. And these three have a few things in common. First, they're information-heavy businesses,

which are good for a software approach. Second, they have regulations, which can be partially automated in terms of compliance via software internally. Third, they're very large markets. Fourth, we're seeing a lot of entrepreneurs in them. And fifth is they don't require that much in the way of physical components. They're still actually pretty friendly to software. I mean, old school is actually building.

schools, but it's not quite at the same level. I would also say that those three industries are the ones where the striking how little they have changed in the greatest invention in the history of information delivery, the Internet, you would have expected it to have changed, to have had a major impact on those information-heavy businesses, yet it hasn't. Like think about the way health care is really not that different or finance than it will. I mean, yeah, they have online brokers and things. But the fundamental structure has not changed in the way, for example,

example, in media, is just dramatically different, right? And so I think that, to me, the disparity between those two things, the fact that they're information intensive and the fact that so little has changed means they are, they are by far the ripe targets now. That said, they're very difficult. I think you refer to them as boss mode, boss level on the video game. Like, this is the tough one. Yeah.

This is the big, big scary dragon that you have to, you know, fight on the last level. Yeah, I think so. I mean, like, I think, so what would be like, I think, a relatively, so finance education, healthcare, I think, are quite difficult. I think that's like about the level of sector that we can take on right now. I would say that what I think of is probably infeasible right now is like nuclear power plant, right? Like energy and transportation, like aviation, like at the level like Boeing or something like that, very large physical businesses are still not, I think, good candidates for this right now.

I mean, there's obviously exceptions. You've got your, you know, musks and, you know, you've got people who are doing interesting things with drones and so on and so forth. but generally speaking like i think that right sweet spot is like finance education health care things in that area