a16z Podcast - a16z Podcast: The Future Of Decision-Making—3 Startup Opportunities

Episode Date: April 29, 2019

As companies digitize, they change the way they make decisions: decisions are made lower in the organization, based on data, and increasingly automated. This creates opportunities for startups creating new ways to collect and analyze data to support this new style of decision making. In this episode (which originally aired as a YouTube video), Jad Naous (@jadtnaous) ‏and Frank Chen (@withfries2) discuss this change and the startup opportunities these changes create. The views expressed here are those of the individual AH Capital Management, L.L.C. (“a16z”) personnel quoted and are not the views of a16z or its affiliates.This content is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. References to any securities or digital assets are for illustrative purposes only and do not constitute an investment recommendation or offer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investor or prospective investor, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by a16z. (An offering to invest in an a16z fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund which should be read in their entirety.)Past performance is not indicative of future results. Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. Please see a16z.com/disclosures for additional important information.

The content here is for informational purposes only, should not be taken as legal business tax or investment advice or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund. For more details, please see A16Z.com slash disclosures. Hi, welcome to the A16Z podcast. This is Frank Chen. Today's episode is called The Future of Decision Making, three startup ideas. It's a conversation I had with Jan Nouse, originally as a YouTube video. You can watch all of our YouTube videos

at YouTube.com slash A16Z videos. Now on to the episode. Hi, welcome to the A16Z YouTube channel. I am Frank Chen, and today I am here with Jad Nouse. Jad is part of the enterprise investing team, and he's noticed something, and so let's just get right into it. So you've noticed something around the way

that big companies are trying to do digital transformation. So why don't we start there? What are the big companies doing? What is digital transformation? Yeah, digital transformation is something that gets thrown around quite a bit. I think there's a big shift now. We're starting to see a lot of industries actually starting to go through digital transformation, and I would bucket the things that people do

in digital transformation into two areas. The first one is around moving from these manual paper processes to more digital ones that are easy to change faster to modify more agile. The second thing that people tend to do when they're doing digital transformation is move from these manual processes to more automated processes, and so automation. And I think that this shift is now starting to happen in earnest, and we're going to start seeing three things pop out. The first one is people's roles and functions to a certain.

degree are going to start shifting around. The second one is we're going to start seeing new demand for new technology and new tools as these new functions and roles actually emerge and start to change. And third, that's going to also lead eventually to a change in market dynamics and how companies run who become successful, who wins in certain spaces. Interesting. So anywhere there is a fax machine or a clipboard or sort of of a big bundle of papers, there's opportunity, right? We're going to go from analog to digital, and then we're going to automate whatever business process

was behind that piece of paper that you had to fill out in triplicate. So why don't we talk about a couple of examples of these? What are good examples? So I'll talk a little bit about product management. So earlier on, people, the way they used to decide what products to build, how to prioritize features or bugs to fix,

is they'd go and they'd run these surveys that are manually and they send them out to people or the product managers go and talk to people. They spend a ton of time doing these, collecting all the data, and figuring out, okay, this is the segment of people I care about the most. Here's the issues that they care about. Let me figure out what the problem is and so on. As an old product manager, I went on those calls. Yeah, exactly.

You flew to a customer and you dutifully listened to what they wanted and you'd sort of come back and try to sort them all. And that, I mean, it's a huge time sync. A lot of the product manager's time used to be that. What's happening now is we have a new generation of tools that actually allow the automation of data collection from the product. What's actually happening? What features are people using?

Where are they getting stuck? And so where the product manager now, instead of having to go and do all these surveys, would look at a dashboard that describes what people are. doing in their product. And then they would be able to analyze it and figure out from that what features, what areas of the product are they getting stuck in, and be able to communicate with engineers. Here's the things that we need to do. And then once they fix some of these, they can actually roll them out and gradually and do A, B, tests to figure out, did this actually

fix the problem or did it not fix the problem? And decide that if something actually did fix the problem, then continue to roll out to the rest of the population. So that's on the product. management side. We see another example actually happening on the marketing side. I'm sure you've heard of growth hacking. So for a long time, marketers used to be this madman kind of thing where you spend a lot of time figuring out the creative aspect of what you do. You spend a lot of time on a lot of money on advertising, advertising campaigns, and you kind of spray and pray for the most part. What has happened over the past few years is the rise of this kind of marketing engineering role to a certain degree. This is one where a marketer who understands numbers, who understands

engineering systems, who understands pipelines, would work with these data systems and actually try to figure out ways that are low cost that would actually increase growth in a certain segment of the population. And that requires a lot of data instrumentation, a lot of understanding people, and a lot of creativity in figuring out how to spur growth

or how to get traction in a certain area. So Don Draper's tools were typewriters and stories, right? And so the tool set around this is going to change dramatically if we make this transition from sort of the old world analog

unautomated to the new world. And by the way, I think you have a name for the new world. Yeah. So let me first say, like, what's actually going to happen. Yeah. So as people's jobs become more and more automated, a lot of the things that they get, a lot of the things that they used to do that are work will go away. And what's actually left in their jobs is mainly decision making, figuring out, like, what am I going to do, what am I going to focus on how should I do it? And communication or other things that are actually related to their job, like creative work, human aspects that can't be automated, buy-in,

alignment, etc. But the rote work goes away. And so that means that there's a ton of decisions, a lot more decisions that they're doing more frequently on a daily basis that they have to go through. So what that means is that to a certain extent, everybody is going to end up becoming more of an analyst in that sense in the enterprise. When I say everybody, I kind of mean like the middle of the enterprise. And what that really means is they're going to have these questions that they're going to need to ask on a daily basis, but with no tools to actually help them do these. So you might say, okay, well, you know, people used to do this for a very long time.

They used to use BI tools to actually answer questions. Yeah, business intelligence, right? So you built the data warehouse, you build the tables on top of it, right? Then you build your reports. Exactly. And so I think that BI tools are not going to be enough in this world. And I've come up with a term for like the type of tools that we need that I'm calling operational intelligence because it's actually targeting the operational people.

It's questions that people need to answer on a daily basis, and they have to answer them immediately. Questions like, where is the bottleneck in my funnel right now, and how do I eliminate it? Or I have, my competitor is having a flash sale. How do I figure out how much of my revenue is impacted, which customer segment should I target, and what should I put on sale? And those are things that you're going to have to answer in the moment. You can't have, so for BI, you would need this army of analysts where you would just ask a question and then they would go off into your enterprise and rummaged through all the data sources,

try to understand kind of like what the question that you're asking is, kind of try to understand what the business context is, and then show you, build you a dashboard and hope that that's the one that you want. Yeah. Well, there's the old joke about BI, right, which is it's $10 million to your first report. and then you realize, oh, I didn't want this question answered anyway. Oops, wrong question. Exactly.

And so the solution there is kind of what I'm calling operational intelligence. And there's three pieces to it. The first one is that it has to be immediate. It can't be eventual like B.I. You can't just say, oh, I need to answer this question and then get an answer like three months later. It has to be answered in the moment. And that involves a few things. Like, first, that you have to actually be able to do it yourself.

Like, you have to actually get the data in real time, as opposed to it being laid. The second piece is that it has to be kind of continuous. It has to be real time. You can't have your data being sent into these systems on a batch basis every day or every week or whatever. The data that you actually see to make your decisions has to be what's happening at this point. Right now. So the classic example of this would be sort of social listening on Twitter, right? Which is that's got to be an ongoing process because things can blow up with your brand either in a good way or a bad way at any time, right?

So you can't say, hey, I'm done analyzing Twitter for the quarter. I'm done. Exactly. Exactly. Another example, I mean, I said this earlier about AB testing. I mean, if you're looking at, if you're trying to do AB tests, you can't just let it go and come back, you know, next week and see whether the thing worked. or not. You actually have to be continuously monitoring what's actually happening in the AB test space and figure out, did the B test work or did the A test work better? And I'm going to flip the switch now? Because if, I mean, you're doing an A-B test on a segment of the population, you don't want them to completely fail at the end, in the end. In fact, we're seeing with some of the more sophisticated machine learning systems that you actually have multiple models, machine

learning models that are live at any given time, and you're actually doing nightly bakeoffs against these models, right, which is model A will get 40% of the traffic, and then model B will get 20% of the traffic, and then we'll just sort of let them run, and the best models get promoted to receive more of the traffic over time, right? So that's an example of what you're talking about, which is a sort of continuous process. It's really interesting that, like, what we've seen is this kind of monitoring, this kind of continuous monitoring, like what I'm calling operational intelligence, has actually been kind of standard

on the engineering side for a very long time. People have been monitoring systems and engineering systems for a very long time. And they would kind of run A-B tests continuously to try to improve performance. And now we're actually seeing these kinds of engineering disciplines kind of migrate into other functions of the org, right? Like marketing seems to have been the first one

to go after that and then product management. And we're actually seeing now people trying to do this for salespeople, trying to, like, look, okay, here are the things that salespeople have done. And in order to close a deal, let's actually learn from that as a pattern and, like, figure out how to get everybody, every salesperson on the team to get to that level of the top performer. Yeah, Cresta.a.a.a, a great example of this, right? So you're chatting and you're getting real-time advice about, hey, maybe this is the time to mention we have a product in this space. Yeah, that's a real-time recommendation. Exactly. Yeah. Yeah, so in the old days,

engineering typically was first because websites were coming online and you needed to watch those things, right? Because everybody knows the statistics that if like, you know, your web page loads this much slower, you're going to lose that much more people through the conversion funnel. And so like you had to watch all these things in real time. And now that's getting outside of IT, right? Yeah. It's interesting also that. So I used to work at App Dynamics. I was there for for a few years. And App Dynamics sells APM tools, application performance monitoring tools. It's probably one of the easiest things to sell.

Because you go up to your customer and you're like, well, how much does it cost for your engineering systems to be down for, you know, five minutes, ten minutes, an hour? And then you say, hey, we prevent that from happening. That same kind of sale hasn't yet? happen in these other orgs. It's a little harder to prove the ROI. But I think it'll get there. Right. So now this is about sales performance, marketing performance of those people. Exactly. And we're going to sort of treat them as if they were websites, right? What's the downtime? What's the dollars lost if you have a salesperson being non-optimal at this point in time?

Exactly. Exactly. Exactly. Right. And so to recap sort of the tool change from business intelligence to operational intelligence, sort of, I need it now. I don't need it in three months. Three months is too late. That's one. Two is I need it ongoing. I don't need a one time, hey, I'm done. I need to.

And then I think there was another aspect of the tools that you expect to change. And what is that? Has to be self-service, not full service. Oh, I see. You can't have somebody else going and doing all the work for you. Those tools have to actually give you insights that are catered to you. And you have to actually be able to ask the questions yourself out of these tools. They have to enable you to do all these things by yourself.

Yeah. So basically, the tools need to be easy enough to use such that the average business analyst can basically just poke in the data and then an answer comes out, as opposed to you think of a question some team later, six weeks later, turns that into a very complicated SQL query, and then the report comes back. I wouldn't even say it's an analyst that actually is doing this, right? These are tools for the actual operational people as opposed to like, I call the meta-operational

because they're like analysts. They're about the business. They're not the business. I see. So what a good example of somebody who now needs to consume these tools directly, which is different brand, a marketer. Like growth hacker, the product manager, the customer support manager, the salesperson, these are all the actual functional operational people

that need to consume this data. Got it. So that would be a big change, right? Because in the past, it was sort of a very sophisticated technical consumer, right, who would be the interface between the business person and the system. And now you're saying the business person needs direct access to the system. Exactly. So that's not going to be easy, right?

So if we think about the entire stack of how it came to be that you've got a BI answer. Yeah. Right? There was ETL. There was storage. There were data cubes. They were analytics. there, right? So, do you think the, like, each layer of the stack is going to need to change?

Or do you think, like, these are just features that the incumbents can add? Yeah, good question. So I think that the breakdown of the stages of data pipeline is a functional breakdown, not really so much legacy. Like, you've got ETL at the top, you've got, well, maybe at the bottom, depending on how you like to draw your pancakes. From the left to the right. You've got ETL kind of right after your data sources. You've got storage where all the data that you've processed goes in. These are your data warehouses, your databases, data lakes, etc.

You've got processing that happens to extract the data from the storage layer and turn it into insights or whatever. You've got analytics that's actually used to turn a question into actual execution. You've got the access layer which controls and governs who is allowed to access what. And then you've got processing at the end, sorry, presentation at the end. That actually... That's where your answer comes out. This is the dashboard.

I think every layer, functionally, each layer is going to remain the same. Like at the core, it's going to be doing the same things. But each layer is going to have new non-functional requirements. Each layer is going to have to be usable by a non-technical person who is trying to ask their own questions. And we see that happen in large companies. These large companies have already built these stacks. So Airbnb, for example, built Superset, and they luckily open-sourced it to the world. And now it's used by hundreds of companies.

It's a presentation layer product that's focused. toward the more technical engineers or data scientists to be able to get ad hoc access to their data and answer questions immediately. One of our investments imply is doing this for the analytics and the processing layer. So they're able to store streaming data directly into their database and allow you to do OLAP

types of queries and analytics on top. And they provide a presentation layer that allows you to slice some dice on problems. Databricks is another one. They're focused on the processing layer. So we're seeing a bunch of things happening in each of these layers. And I think probably the layer that hasn't yet

seen the most changes is the ETL layer. And why do you think that is? Is that the hardest layer? Is it just, well, that's going to be the hardest to turn a business user into a direct customer of, because traditionally that's been very wonky. Yeah. I think two reasons why ETL has been so hard.

The first one is it actually requires domain specificity. Like, ETL for healthcare is not going to look the same as ETL for financials. Ride sharing or for ride sharing, for whatever. Like the ontologies, the things that they care about are different. And so any company that does these has to really get deep into that domain. The second one is it's a lot of integration and a lot of kind of heavy manual work and engineers don't really like to build these kinds of things. So they're going for the lower hanging fruit at this part. Got it.

But it seems like overall you're arguing there are a lot of startup opportunities here that the incumbents are going to have a hard time retrofitting their products, right? So it's pretty hard to change a product that was designed originally for a technical user. to turn that into a non-technical? Is that sort of a fair summary of where you're going? Yeah, so if you think about the opportunities in operational intelligence, I'd probably break them into maybe three categories. The first one, actually the first two are maybe related to each other.

It's basically you want to become an operational intelligence vendor. So you sell software and tools that enable existing incumbents to become operationally more capable. You enable them to do operational intelligence. And within that category, there's a breakdown. So you can either target a specific role, so I'm going to enable the salesperson to become successful, or I'm going to enable the product manager,

or I'm going to enable the customer success manager. And we see products in each of these categories today. There hasn't yet been a complete, breakout success in any of these, but it's super crowded, and I think it's probably the hardest one to win in at this point. The second category is, within that vendor superset, is segment-focused vendors. So companies that sell operational intelligence tools to existing incumbent, for example, companies that sell sensors and analytics for oil and gas companies. So these are people who will collect data from your wells, optimize it, and then collect that data from

your wells, put it into dashboards, tell you how your wells are doing, and tell you how to optimize it in order to improve efficiency. So like a vertical solution for oil and gas. For oil and gas. So those those are still vendors selling software, maybe some hardware, into an existing industry. And then finally, you have the vertically integrated, you know, operationally intelligent company that competes against the existing incumbents. And so we've got plenty of examples of that at this point. So we've got Airbnb that's in the hospitality business. We've got some Sara in the logistics industry. We have Lyft and Uber in transportation. And I think that's where the biggest value is, but also one of the hardest to go into. Yeah, the classic full-stack

startup, right? Which is I'm going to build these operational intelligence tools, but nobody else gets to use them. I'm using it to serve my own business. And I'm going to win the market by winning the customers directly. And I think that the industries that are going to win the most out of operational intelligences are going to be these kind of like traditionally non-IT buyers. So oil and gas, groceries, construction, these are businesses that are really trillion-dollar industries or trillions, but they have very low margins. They've existed for such a long time that they've operationally become really efficient and at the same time commoditized.

So I'll give you an example. The largest construction group in the world is called the ACS group. The revenues are about like $34 billion per year, but their margins are about 6.5%. And so a small change in the gross margins for these businesses, a small change in how operationally efficient they are translates into huge increases in their profit margins. Another example is Costco. So in 2017, their revenues were about $12.5 billion,

and they were operating on about 11% gross margin. Again, another place where a change in operational efficiency can lead to huge changes in revenues. The final example is a little different. This one is less about gross margins, but more about capital deployed. And so the example here is ExxonMobil, the mobile. If you were to guess what they're,

like the value of the capital that they have deployed around the world, what would you guess? Oh, Exxon Mobil? Hundreds of billions? Is it the order magnitude? So Exxon Mobile is about, $230 billion in capital, and the way they measure their performance is on return on capital employed.

R-O-I-C, right? R-O-I-C, yeah. It's very different. It's different than how, you know, the grocery example I gave earlier, which was based mainly on the gross margins. And their return is about 9.5% or so. So, again, a small change in the operational efficiency of the, of the capital that they have deployed can translate into huge additional gains.

I mean, they're deploying about like $23 billion additional capital this year. That's a lot of spending. Yes. And that's the, I mean, it's really interesting, like helping these companies that are capital heavy as well. So it sounds like you're excited about a whole sort of gamut of startups. One would be, hey, look, I'm going to sell a particular technology to enable. you to be more operationally intelligent, right? You're also interested in the full stack

startups, right? Which is, I can sell an entire solution to a customer directly and nobody else gets my OI goodness, so to speak. What are some examples of sort of startups that you, or what are some examples of things that you're personally excited about? I can give you some examples on the infrastructure side. So I'm excited about the Superset project. I'm excited about what imply is doing. I think there's a lot of, I think a lot of what's actually happening is people are now starting to see analytics

and observability as urgent, as necessary to running their business. And so I think that there's a really great opportunity in that space. I'm also really interested in companies or vendors, software vendors, into incumbents, into large existing industries, like into construction, companies that sell into construction

or companies that sell into groceries. We've seen a few startups in that domain. Some of the hardest problems here is that these are startups that are going to have very different economic profiles than the traditional you know, Silicon Valley startup that we know. How so?

So, first off, these are, you're selling into markets that are stagnant, that are very low margin. They don't have a lot of margin to go around, right? They can't afford to pay a lot.

Exactly. And they're not used to buying new technology. They kind of understand one, two, and three, and like they don't really know about four, or they don't know how to digest And so a lot of the effort there is going to be around educating and the sales cycles are going to be very long. The pie at the end of that, like the other, the flip side of this is that these are huge businesses, right? Like construction, oil gas, retail chains.

Once you're in, you're in. And so when you're actually starting a company in this area, there's a few things that you want to keep in mind. One, you need to educate your investors. Usually investors are not going to understand these businesses really well, and they might not know the difficulty of actually selling into them, like what it takes. And so you need to prep your investors for this long-haul thing for the long term. And they need to understand that at the end of this,

there's a really bright light. The second piece is you need to get, domain expertise. Like, you need to become the expert in that business, and you need to become a kind of trusted advisor to these companies. And so when they say things like, oh, you know, we want to go through digital transformation, you need to help them understand, like, here's what that means. We're going to be here for you. We're going to guide you through it and actually help them with both a significant amount of services as well as software on the back end. So don't shy away from the services.

Don't shy away from the services, especially in these industries. Well, Jan, thank you so much for coming and sharing this idea. The good thing about this is that the world really is changing fast. If you are a retailer, Amazon has scared the bejesus out of you, right? And so what used to be a very long tedious sales cycle has gotten a little quicker because Amazon's in the rearview mirror. And so everybody sort of knows that they need to go faster. to make decisions sort of lower in the organization. They need to make them in real time.

And so it's exciting to see startups helping that transition to real-time decision-making pushed lower in the organization. So thanks for joining the YouTube channel. If you liked what you saw, go ahead and subscribe. Feel free to leave comments. Maybe the question that I'll use to prime the comments is, what are your favorite examples of decisions that now need to be made in more real time? and look forward to joining the conversation there.

See you next episode.