The Exchanges

Q How willing are you to do small contract, and what would your advice be to do small contracts with mega customers where they're like, hey, land and expand?

A I think you need to be very willing to do that. Now, the key thing that we've learned, and I, you know, learned this at Slack as well, but at Glean, You've got to have success criteria. You've got to have, okay, we're going to do, let's say for Fortune, 100 company, we're going to do a 100,000, a 150,000 land deal, but you have commercial alignment and executive alignment on what are the success criteria that we need to hit to get to that seven figure contract. That's how we try to do it in structure. It's a very timely topic was we're like debating it even more as a, as a go to market team. We had one Fortune 100 companies say, we want to test how you're, how we're lowering call resolution time with Glean relative to the other solution we had and the competitors. It was very time bound. It was very specific. And we're able to tie that back exactly to dollars. So that land slash paid POC in some ways turned into the seven figure contract.

AI assessment note: “I think you need to be very willing to do that.”

Q How willing are you to do small contract, and what would your advice be to do small contracts with mega customers where they're like, hey, land and expand?

A I think you need to be very willing to do that. Now, the key thing that we've learned, and I, you know, learned this at Slack as well, but at Glean, You've got to have success criteria. You've got to have, okay, we're going to do, let's say for Fortune, 100 company, we're going to do a 100,000, a 150,000 land deal, but you have commercial alignment and executive alignment on what are the success criteria that we need to hit to get to that seven figure contract. That's how we try to do it in structure. It's a very timely topic was we're like debating it even more as a, as a go to market team. We had one Fortune 100 companies say, we want to test how you're, how we're lowering call resolution time with Glean relative to the other solution we had and the competitors. It was very time bound. It was very specific. And we're able to tie that back exactly to dollars. So that land slash paid POC in some ways turned into the seven figure contract.

AI assessment note: “I think you need to be very willing to do that.”

Q On that, sorry, let's just start on that. Sales segmentation, does that need to change as you move upmarket?

A Absolutely. Yeah, I mean, when I, when, you know, when I came into, at both Slack and Glean, when I joined Slack, I immediately, after we got to about ten million in revenue, Cut the team into SMB and enterprise, and I basically said enterprise is a thousand employees and above, SMB is below, and that way we just sort of put the resourcing, the more expensive sales resourcing, on those bigger accounts. Now, at Slack, it was a lot of inbound, but even that segmentation is important to do, so you at least get your true, like your sellers starting to think about upmarket Inbound at a minimum. At Glean, when I joined, we had real great success in that mid-market segment. We had strong sellers, about like eight of them, and we decided very quickly with the team to say, we've got to move up market, let's cut it at 2000 employees and above, and, and the below will be serviced by a smaller subset of the team. And that helped us start to accelerate moving up, but you're right, it's not like, because you only have eight AEs or 10 AEs, you're not all of a sudden touching every Fortune 100 and able to do deals overnight, and so what we did was strategically pick one or two key bets that we wanted to get to prove success in. T-Mobile became that account for us at Glean. Not just a sales team, as a company and leadership team, we said, we are going to prove success at a Fortune 100, Like T-M…

AI assessment note: “Absolutely. Yeah, I mean, when I, when, you know, when I came into”

Q It's a really, I hate kind of broad and generous questions, because they're generally for crap interviewers, but, you know, uh, as I said, I've done 2700, so hopefully I have some skills. But when you think about kind of the AI landscape today, how do you think about where the most value will accrue, and you want to concentrate most of your time and capital?

A We just talked about how everyone's over-investing right now into this cycle, and because none of us can miss, whether it's the large incumbents, Or us as venture investors back in companies. And so your question is like, where do we invest as venture investors? And I can tell you, we're, we've invested in a lot of application layer companies, and that are solving very specific pain points. Uh, and the way we've looked at it pretty simply is, if you think about, we took actually the top 20 jobs in the US, and who makes the most? Simple. And it's doctors, it's lawyers, and it's developers. How do we help supercharge these people who are highly scarce, highly skilled, and we're not producing enough of them? So you try to build software, AI, that helps them do their job better. And, uh, so we backed companies that help doctors, lawyers, and developers with co-pilots. So Harvey, Ambience, And Kodium.

AI assessment note: “we've invested in a lot of application layer companies, and that are solving very specific pain points”

Q When we look at the venture landscape, you have like, in my mind, boutiques, USV benchmark boutique, and then you have like capital accumulators, which is Tiger, KOTU, Andreessen, General Catalyst, Lightspeed, Sequoia now. Respectfully, and I say this with Tony, where does Kleiner sit in that? Because you kind of sat in the middle in my mind. How do you think about that?

A We are primarily early stage focused. Uh, we have a, an eight hundred million dollar fund for that. And then we have a, uh, 1.2000000000 dollar growth fund. And we, this team of seven folks, um, invest Out of both of those funds. I would characterize us as boutique because we're kind of a small team that believes in the craft of venture capital. Uh, it's, we, we think it's a business that doesn't scale, actually. Um, and so, uh, we're not scaling through people, but we have the scale of capital. Because our growth fund, even, half of the dollars are allocated towards, not allocated, but half those dollars are invested in our best companies from our early stage funds. So it doesn't require us to have a, a large, uh, team, so to say, because we're already involved with some of these companies like Rippling and Glean and Figma that we're doubling down into out of our growth fund.

AI assessment note: “I would characterize us as boutique because we're kind of a small team”

Q It, it's fun. It's also challenging from a pricing perspective. I saw three companies my moon last week that raised it over the seven hundred and fifty million pre-product. How do you think about navigating the pricing environment when there is such further pitch excitement for these companies?

A Great question, Harry. And I think we all sort of fall victim to those, uh, every once in a while, but that can't be the core part of the business. That can be the, the one like that got away and you have to get Into this pre-product company because the founder is so exceptional. That can be, you know, one out of the 20 deals you do this year. It can't be every single one of them. Because as you know, Harry, you know, we have to get our ownership at the early stages where you're investing five to ten million dollars for 15 to 20% for the math to work for, for our funds. And it can't be done if you're investing, you know, twenty-five million at 750 post out of an early stage fund.

AI assessment note: “That can be, you know, one out of the 20 deals you do this year.”

Q know, when we look at a lot of potential use cases, a lot could be subsumed by the foundational model companies, if they are big enough. An example could be talking translators, you know, talking avatars that you could talk to in a friendly enough way. How do you, Do you worry about application layer companies being potentially subsumed by foundation model layer if they are such a core competency?

A I don't. It's, it's a bit like, ah, the hyperscalers thinking they can do everything, and they've decided that that's a great business model, is to, ah, own the electricity, and, ah, or the, and the pipes, ah, and then just charge for, by the, by the hour, or the kilowatt hour, and I think that's a pretty darn good business model for the hyperscalers that provide the models, um, in OpenAI, and I was reminded by, you know, I was at OpenAI maybe a month ago, and, you know, we're Going through all these demos of cool products that are coming out, like O-one and Strawberry, and realize that their positioning is, we can't do everything. We're a 600 person company. We can't build the application layer stuff that you guys, we want you guys to build, or your companies to build. And so, I think that there's a great business to be had in LLMs and in providing the compute and the electricity, and there's a great business to be had by being very vertically focused Around applications.

AI assessment note: “I don't. It's, it's a bit like, ah, the hyperscalers”

Q Klarna announced that they were stopping their partnerships or, you know, customer, you know, contracts with that Workday and with service, with Salesforce. Um, and it led to this wave of, wow, this is the future. Companies will build their own SaaS tools. I'm interested. How do you feel about that? And how do you respond to that question and concern?

A I have no choice but to say that that's a terrible fucking idea. Uh, just like Intercom is eventually fucked by AGI, we'd also be fucked Buy companies building their own AI systems. The reality is hand on heart, independent of my association with Intercom. Um, I think that building your own AI agent platforms is going to be as smart as in a previous generation, building your own SaaS workflow tools. These things are super deep. Like I said, with Finn, our AI agent, it's the product of over a hundred different experiments. Many patented parts. We've got 30 ML engineers and counting, you know, a hundred other engineers working on all the application components. Um, you're just not going to build a highly performant, rich AI agent on the side, not going to be done. And I think any attempt to do so may yield some short term excitement, but even Klarna long term, I'm pretty sure we'll move on to a professional purpose built system.

AI assessment note: “I have no choice but to say that that's a terrible fucking idea.”

Q Do you buy the reduction in team sizing that AI has? I think if you look back to the seventies, people would say, God, if we had computers that did all these things, we wouldn't need anyone here. Yeah, there's still companies with hundreds of thousands of people. How do you feel about the reductive nature of AI on team size?

A Valuable businesses have got smaller with technology. You know, the size of Google or Facebook when they were worth a hundred billion dollars in employee size was, was substantially smaller than Um, you know, General Electric or Ford. So the, the long arc of technology and automation requires less humans, so it'll perpetuate that arc. You're not going to see mass layoffs. At Intercom, for example, we've been using our, this AI agent fin for nearly two years, and we've not let anyone go, but we've stopped growing the size of our support team. And so I think any future growth is going to be catered to by the AI agents. The AI agents will first do a lot of low level and simpler work, and the humans will do the higher level work. Now, when we hire for customer service, we're hiring a more talented, experienced person who can do the creative things that the AI agents can't. Eventually, as these teams get smaller and through attrition, um, the AI will be doing more and more of them, and there just will be less humans in the mix.

AI assessment note: “the long arc of technology and automation requires less humans, so it'll perpetuate that arc.”

Q The question that I am perpetually stuck by is will the infusion of AI into products lead to an increase in average revenue per user, or will it just lead to a better customer experience?

A I think it's already delivering incremental ARPU. Uh, and I think it'll happen first in companies like Meta, like I said, because you have an auction marketplace, this gets repriced immediately. And so if I look at Meta, we estimate that it's not just delivering revenue. I think it's already delivering about fifteen billion of incremental EBIT, EBIT for Meta, right? Just in, in the form of more content recommendation means you have more time on Meta properties equals more ad inventory and better ad matching equals higher CPMs and higher CPMs just, you know, it's straight flow through to the bottom line. So it's, it's beautiful already for Meta. Now, uh, I look at the software companies and I just think it will take more time because you have to go out to your customer and say, look, I'm delivering you this value. Here's the data. And when your contract comes up again, we're going to raise your prices. It's just a hard conversation. But what you have seen, even since you recorded the pod with, uh, With David, Canva, which you mentioned, has raised prices on its enterprise plan by three X, three X. A lot of flow through, I think.

AI assessment note: “I think it's already delivering incremental ARPU.”

Q The question that I am perpetually stuck by is will the infusion of AI into products lead to an increase in average revenue per user, or will it just lead to a better customer experience?

A I think it's already delivering incremental ARPU. Uh, and I think it'll happen first in companies like Meta, like I said, because you have an auction marketplace, this gets repriced immediately. And so if I look at Meta, we estimate that it's not just delivering revenue. I think it's already delivering about fifteen billion of incremental EBIT, EBIT for Meta, right? Just in, in the form of more content recommendation means you have more time on Meta properties equals more ad inventory and better ad matching equals higher CPMs and higher CPMs just, you know, it's straight flow through to the bottom line. So it's, it's beautiful already for Meta. Now, uh, I look at the software companies and I just think it will take more time because you have to go out to your customer and say, look, I'm delivering you this value. Here's the data. And when your contract comes up again, we're going to raise your prices. It's just a hard conversation. But what you have seen, even since you recorded the pod with, uh, With David, Canva, which you mentioned, has raised prices on its enterprise plan by three X, three X. A lot of flow through, I think.

AI assessment note: “I think it's already delivering incremental ARPU.”

Q In 2016, there was a leadership transition, which I think was a tough moment. Doug told me they did not know whether to keep global equities and made you earn it. I was writing down vociferously. Talk to me about that. What happened there, Jeff?

A So you've probably listened to the Sequoia Crucible moments podcast where Rulof dives into these difficult moments in a company's journey that, you know, require a lot of fortitude, but ultimately set the company in a better direction. That was 2016 for SCG. So I'm, I'm not a founder, um, but I did help guide SCG through what I call a refounding moment. So Sequoia had hired an original portfolio manager for SCG. Um, you know, in, in, in, in, he was a smart, hardworking guy, but didn't really unlock the synergies that we should have within our ecosystem. So, um, really he, he had the same playbook at his prior hedge fund. And so in a lot of ways, he was trying to recreate his prior hedge fund, even investing in, in non-tech areas, for example, as opposed to building something special and unique to Sequoia. And so performance was good. It was not great. And so there's this crucible moment in 2016 where Sequoia decided to part ways with the original PM and actually consider shutting down the business entirely. So as the senior most, uh, partner remaining on the team, I was asked to come up with a business plan and convince the broader partnership why version two would be better. Uh, and for us, it was an incredibly unifying moment. Because we were really fighting for our survival, right? And so the team actually, we actually all left the office. We rented some co-working space dow…

AI assessment note: “I was asked to come up with a business plan and convince the broader partnership”

Q So I, I completely agree with you there. Like, I think there is still a tremendous, tremendous amount of room to run. Is Mag-Seven sustainable? It is carrying so much of this market. Is it sustainable?

A I think it is for some time. So when, when you think about AI today, Productization of AI is a function of owning the customer and owning the data. Right? We are not in a world where AI has created a new distribution methodology. And so if I think about some of the walled gardens, like a Meta, you own the customer experience, you own the data. And so your ability to productize and roll out AI features and functionality to Instagram users very seamlessly is incredibly powerful. Because Meta has this auction marketplace for ads, you also have the ability to reprice Your AI features very quickly, right? Everyone, all these merchants are getting a return on their ad spend, and if someone is getting better return on ad spend, the CPMs for these ads go up. So, I think there's a massive moat for some of these largest companies. Now, what I worry about Well, we're not, not worried about, but what I think will happen over time is this will start to disperse more broadly, but at least in the initial stages of AI, where your data and distribution are so important, I think the mag seven will continue to do quite well.

AI assessment note: “I think it is for some time.”

Q be a little bit different now, Brett. Like, how does that work? Do you know what I mean? It's like, so just help me understand. You decided you're going to do Sierra, and you're like, ah, what? Yeah, I mean, I guess it's kind of a question of, like, why fundraise? But then there's also a question of, like, how did you approach that? Now you could raise from anyone.

A Well, first, why did I fundraise? Um, I really believe in the importance of boards and having stakeholders and the accountability. Of, you know, having a board and investors and employees, and I want the employees coming to Sierra to know that Clay and I aren't doing this as a side hustle. You know, we want to build a generational company. And then similarly, I really value the advice. I've been a board member as well as an executive, and I really value the strategic advice I got. So when we started the company, I just called Peter Fenton, who I've worked with twice before. He's the only person I talked to. And, you know, that, that was our first board member. And with our subsequent round, similarly.

AI assessment note: “why did I fundraise? Um, I really believe in the importance of boards”

Q Now I would love to start. I heard that you worked in a series of restaurants to finance your education in the early days. Can you just talk to me about that? Where was it? How old were you? What did you learn from those early days?

A I was 17. I was lovestruck. I came to United States to chase a girl and I had no place to live. Well, that, that's the gist of it. So I went to a restaurant to beg for a job, and I succeeded, and I washed the dishes, and the bus tables, and I weighed down the tables, but I wasn't very good at it because my English was terrible. I persisted, and I continued to climb the ladder to become an assistant manager. So that was my career. It was a pretty good one. It was in Santa Ana, and I got shot at. But that trained me to be tough and resilient. But that was one of the early experience or experiences that I had.

AI assessment note: “It was in Santa Ana, and I got shot at.”

Q start though on a statement that you said before about consumer subscription, which is kind of a core focus for you in growth and one of my passion points, um, reason why I have a few friends. Um, but, uh, you've said before that, you know, in particular consumer subscription apps are easy to launch, but hard to scale. If we just deconstruct that, why are they easy to launch?

A Sure. So I think there are several advantages that consumer subscription businesses have. Relative to more complex business models, and that might include B to B SaaS. It might include marketplaces, uh, even certain social networks, right? Like simple on their surface, but there's a lot of complexity underneath. When you think about consumer subscriptions, and this certainly isn't true of all of them, but in many cases, they're, they're operating in relatively mature categories with customers who understand that the product they're bringing to market, uh, and, and they're just relatively easy to launch for a few reasons. Number one, most of them don't require sales teams. Um, number two, Most of them don't have to deal with complex two, two-sided marketplace dynamics. Um, they tend to have high gross margins, low marginal cost to serving additional subscribers. And then over the last 10 years, thanks to the app stores, they get all these advantages in terms of global distribution, payments, turnkey support tools. So they're very easy to get up and running. They, they tend to be able to get to market faster and with less capital than a lot of other tech companies. But then you run into a lot of challenges that make them hard to scale.

AI assessment note: “they're just relatively easy to launch for a few reasons. Number one, most of them don't require sales teams.”

Q Phil, can we just spend our way to dominance? Can we spend our way to growth?

A There are some, some rare exceptions where I think that has happened. Typically in marketplace businesses where the network effects are so strong once you get to the tipping point that you can have upside down unit economics for months or even years. And then, you know, once you Flip that tipping point. You pay it all back. And I think there was a period over the last decade where the funding environment and, um, VC preferences for, you know, massive user growth rates over profitability were such that a model like that could work and with the right business model and with the right network effects, it might be able to justify itself. But more often than not, certainly in the case of consumer subscription, that's, that's just not the case.

AI assessment note: “more often than not, certainly in the case of consumer subscription, that's, that's just not the case.”

Q just on the CACs alone, when you review the different companies that you've worked at and CACs and how they change over time, do CACs get higher as you saturate your core market and expand into ancillary markets that are maybe less direct, or do they get lower because you get brand recognition, word of mouth, and a lot of other ancillary benefits from just being bigger and more pronounced?

A Yeah, it's a great question, and I'll sort of divide the world up into two categories. There are the outliers, like, let's say, Duolingo, Tinder, Strava. You do have examples of companies where some combination of the quality of their product and the virality of their use case leads to this tipping point where they become so mainstream in the public consciousness that at least for a while, their cacks go down. Their blended cacks go down because you just have so many people in their target demographic talking about this product and sharing with I mean, ChatGPT is a great example of that right now, right? Like, it's so viral because it's such an amazing product, and so I don't know what their balance of paid versus organic acquisition is, but I have to believe that the vast majority of it is just viral because everybody's talking about AI right now, and ChatGPT is at the center of that conversation. But those are the outliers. For everyone else, CACs almost by definition will go up over time for a couple reasons. One is, and the simplest is, as you expand beyond your core, And as you tap out your highest intent early adopters, you're just going to have a harder and harder time converting eyeballs into subscribers, right? So like your cost per install will go up, your, uh, signup activation rate, trial start rate, trial conversion rate, like all of those metrics become harder and…

AI assessment note: “For everyone else, CACs almost by definition will go up over time”

Q On the CAC and the payback element. Yeah. CAC obviously is incredibly important because the higher the CAC, the longer the payback will be if, you know, the ARPU is the same. What is good versus great in terms of payback when you're looking at consumer subscription apps today?

A Six months is good. One month is great. You know, first, first session is exceptional, which is significantly lower than B to B, right? And B to B, you know, 12 months can be considered a good payback or even 18 months for an enterprise SaaS business. But we talked before about how consumer subscriptions don't have the benefit of Low churn rates, high net revenue retention, sometimes over a hundred percent net revenue retention over time. There's also the fact that RevenueCat recently had their state of subscriptions report for, ah, for twenty-twenty-four, and they reported that over 75% of trial starts actually happen in the first 24 hours after a user installs an app, in many cases on the first session. So when you look at both the shortness of consumer attention spans and how quickly most trials need to happen if they're ever gonna happen, Combined with the high churn rates and low NRR, um, dynamics of these businesses, it's really, really important to try to convert users quickly and to pay off that initial paid CAC, uh, as quickly as possible. So, so generally with the clients I'm advising, I'm saying, look, ideally we want payback within the first three months and ideally within the first month if we can get there.

AI assessment note: “Six months is good. One month is great.”

Q retention rates on Duolingo is like, 50% for 12 months, and that's considered great. How do you think about retention rates of good versus great after, after different time periods? And help me out as an investor here. What time period should I be looking at, Phil? Should it be 3060, 90. Should it be 31, 83, 65? What are the important time milestones and what is good for them?

A First of all, I like to break out retention rates on monthly versus annual subscribers because they look very different. Um, for monthly subscribers, I think any company that is retaining more than 50% of its subscribers for six months or more is doing a very, very good job because monthly subscribers just Tend to turn at much higher rates for annual subscribers. Typically look at the first two years of subscriber retention, because there's a lot of data that shows that after the second renewal period, the subscribers that you've retained for two consecutive years as annual subscribers are, are often going to retain for many years after that. And they become the foundation that you're building your business on long-term, because those are your highest intent users who are going to stick with you, uh, for the long haul. So, so generally I'm looking at. Two to three years on annual subscriber retention. The first six months on monthly subscriber retention and trying to get those retention rates as high as I can. And then there's the obvious thing you want to look at, which is just does the curve flatten out or in an ideal case, does it actually, is it a smile curve where it comes up? So Duolingo is an example of where even as highly touted as Duolingo is, they don't have fantastic monthly subscriber retention rates, but you actually get quite a bit of reactivation on Duolingo whe…

AI assessment note: “retaining more than 50% of its subscribers for six months or more is doing a very, very good job”

Q Do you buy notifications as a retention mechanism today? They are so overused. We have such notification overload. How do you think about that?

A Yeah, so I think it's a double-edged sword, and I hate to keep going back to Duolingo, but they are one of the biggest success stories in consumer subscription, so I will in this case. Luis Van An, the CEO there, has multiple times talked about their approach to notifications, and I've listened to a couple podcasts where he talks about how he got this warning, um, from one of the other, uh, founders that he looks up to That any time you increase the number of notifications or emails you send, in the short term, it's like the sugar high. It's going to lead to a short term pop in your metrics. But if you do that too many times, you kill the channel. Right? And so it's really important to make sure that every incremental notification you're sending earns its place. And that, to me, that means two things. One, it means that just having a statistically significant lift above baseline in a metric like DAU retention rate isn't good enough. It needs to be above, it needs to be a significant enough lift that it earns its place in the product experience. And then the second thing is, you know, moving from science to art again, it needs to fit within the holistic experience you're building for the user. And so if this notification is fundamentally making the user's experience a little bit better, and that could be because it's reminding them to come back and not lose their streak, or it c…

AI assessment note: “I think it's a double-edged sword... every incremental notification you're sending earns its place.”

Q What are the biggest mistakes you think startups make when it comes to the pricing and the packaging of their products in this way?

A I think there are a few things. So I'll, I'll separate out pricing and packaging. So on pricing, I think the biggest mistake, and it seems so obvious, is it's remarkable how many consumer subscription apps will set a price and then won't revisit it for years and years. And there are some, there are some famous examples of this, right? Like, I mean, Quizlet went many years without changing its pricing. Um, Strava, AllTrails, there, there are a number of companies that like launched their app, found a price that worked well enough, and then several years went by, they went back, they did a pricing study like Conjoint or Van Westendorp and or the AB tested the pricing and the product, and they found that there were significant opportunities to improve pricing. So on pricing, it's like, this is something that should be revisited, in my opinion, at least once a year. That doesn't mean you have to do a whole expensive pricing study, but like at least Do some quick analysis to make sure your price is still, um, optimal. Packaging I actually in many ways think is the opposite, where like, as I said before, less is more in consumer subscription, right? Consumers have short attention spans. Um, the more complexity you introduce in the paywall, the, the worse your conversion rates are going to be on the margins. And so if you're going to have multiple tiers, you should only be doing that …

AI assessment note: “I'll separate out pricing and packaging. So on pricing, I think the biggest mistake”

Q Can we take one, apply systems design in a way that you did, just so we can understand that in a more contextual manner?

A Yeah, so let's, the first job I had at Notion was actually, like, the three people reported to me were support, customer support, right? Now, typically, as the number of tickets or as the number of requests go up, um, The first thing that people typically do is they just add more people. They sort of like divide the tickets by productivity, and then they come up with some number of people. Uh, the product manager in me was like, well, we could build a better system. So we could build better tools so that each support person can do more every day. But also, you know, once you have enough of these, uh, customer tickets, then the next role you want to hire would be Something called a product ops, right? Product ops would take all this feedback and synthesize it and, and share it with the engineers. So one thing Ivan and I did in the, in the early years was, um, we created this like canonical database of about 200 types of requests that would come, which would map to like features, right? And we trained our support team to tag every request that come in With one of those two on attacks, you have to do that, right? That was part of the job. And every morning at six AM, we wrote a script, which would take all these tags from intercom and write it against these database tags in notion. And so essentially what. Every engineer at notion had was a real time view of what people were talki…

AI assessment note: “we created this like canonical database of about 200 types of requests”

Q the upside requirements change with each stage? And what I mean by that is, you know, when you are seed or series A, it's like, hey, I need this to be a fun return. When you get to growth, does it change to be like, mm, we'd like to see a three to five x pathway? How do you think about that, like, upside requirements as it changes across stage?

A When we started our first growth fund, Um, our, our hypothesis was probably three to five X over three to five years, right? Um, I'd just come out of, of, uh, a private equity firm, um, as had some of my colleagues. And so we had like a, we had that mentality. And when we looked at the performance of the fund, it turned out there's a power law. There were companies that were five, 10 X plus, um, There were very few companies that were actually three to five X over three to five years. And so we kind of accepted the fact that we're not value investors. We're invest venture investors at growth stages. We're driven by high conviction. And as a consequence, when we're building our investment cases that growth, we never, it's like, it's super dangerous. I think to say there's a safe two X, there's no such thing. And you alluded to this with, Um, the memory of all of the activity from 2021. There's no such thing as a safe two X. And so what we're looking for is definitely five X, um, plus upside, even at growth investments at late stages at high prices.

AI assessment note: “what we're looking for is definitely five X, um, plus upside, even at growth”

Q IQ scale just with that, but I'm thrilled that you like it. I do want to start there because you have Literally, I think the most impressive operating career that I've had on the show, definitely for a long time. So when we look at the operating career, what are the single biggest achievements to you that you say to founders as, hey, I did this in my operating career?

A Well, thank you. It was a 30 plus year operating career. So in that 30 years, there were some, some tough years and some good years, but the two things that kind of stand out, I spent Uh, almost nine years at a company called Data Domain, where I joined, uh, at zero dollars in revenue, and by the time I left, we'd gone through an IPO, took it to about a billion dollars, and then we got acquired by EMC in a hostile acquisition in 2009. Whole story about hostile acquisitions, if you ever want to get into that. Um, and then the next one is, I spent, uh, 10 years at ServiceNow, uh, joining Frank Slootman again, who I had the privilege of working with at Data Domain. Um, into that company, which was about eighty million or so in size when I joined, and when I left it was five billion in revenue, and, uh, felt like the job was well done, uh, to make that happen.

AI assessment note: “the two things that kind of stand out, I spent Uh, almost nine years”

Q too soon. They want to expand into SMB from enterprise too soon, and I always say, it is so much deeper and richer than you think, and we've only penetrated one percent of it. Like, let's absolutely double down where our strength is. How do you think about when is the right time to expand customer segmentation, and when is the right time to say, no, we're going to focus?

A What started happening at ServiceNow that we took advantage of was our customers started using our product in places that weren't It wasn't, ah, originally positioned for. Turns out that by 2015, we had hundreds of customers using us in new ways. And so what we started to do is just listen. And then we productized what they did and built full on customer service product. Full on security incident vulnerability products. Full on HR employee experience products. And because we just followed the customer, we had better success. Um, That was not easy, because we ended up creating business units. I had to come up with a different selling model to sell new technology, because the buying persona was outside of the core. I didn't do it until I was fairly penetrated and winning in ITSM. If I didn't feel like I was already winning there, it was not a distraction, then I could do it.

AI assessment note: “I didn't do it until I was fairly penetrated and winning in ITSM.”

Q Well, talking about the differentiator that being like being on the ground, being in person, when you launch a new city, can you just talk me through that city playbook rollout? Do you just pick random people and send them? How many do you pick? What's the organization? Can you just walk me through that?

A Yeah. So at Uber, we saw almost every city as its own startup, and that was an incredibly freeing thing. For people that were young in their twenties and thirties to kind of go into a market and create it from scratch. And the playbook was probably a 180 steps. We should probably open source it at some point. It's on an Asana checklist somewhere. But how we thought about it was we wanted the right team. And, you know, for us, the right team meant we were hiring for three roles. One was we'd send in a launcher, um, that would kind of pop around from city to city. Started out in L.A., then did Philadelphia, then did Atlanta, and sort of the profile of the launcher was, ah, MBA type, private equity, banking, and, you know, the launcher would be responsible for hiring a team. So in a general market, we would want a general manager who acted as like the CEO of the city, overseeing both driver and rider, and that person was, ah, very similar background to the launcher, banking, private equity, consulting, MBA. Stanford class of 20 12 was very good to us. And then The second role that we hired was an operations manager, and that was more like banker consultant types like me, um, that would oversee sort of the driver, uh, area of the operation and be responsible for growing that. And then we'd also work on, uh, getting a marketing manager that would oversee the rider side, DD, uh, part…

AI assessment note: “how we thought about it was we wanted the right team”

Q I think, I think for me, it's like input and output metrics. The amount of times it's like, hey, revenues are metric. And I'm like, no, number of rides per week is our metric. I don't care about the revenue. That is the output metric. Final one. What's the best growth strategy that you've seen in the last 12 months? And why have you been so impressed by that one?

A I'm a huge perplexity user. I think it's an amazing product. Um, huge fanboy to use your language. And, uh, you know, I think one of the things they've done recently that's really interesting is they're offering a free pro membership, um, to anybody with a LinkedIn premium account or an Uber one membership. So instead of spending 20 dollars a month or 250 dollars a year, you plug in your info and they'll give you the free usage. So they basically kind of locked in people for a year and, you know, if you believe that perplexity could be like the next Google search, you know, you're happy to pay that CAC or sort of that free membership and sort of The GPUs over the year to acquire those users and generate stickiness when a lot of tools are facing high term.

AI assessment note: “they're offering a free pro membership, um, to anybody with a LinkedIn premium account”

Q How did you retain the drivers when it was one out of 10, two out of 10? Because the hard thing is like the symmetry of timing, making sure it's aligned. How did you keep them when there was nothing coming?

A What I guess people call the chicken and egg problem or the cold start problem in marketplaces by doing a couple things. In the early days, we wanted to make sure that drivers, when they were sitting around, were kind of like paid for that time. So in the early days, we paid a driver 20 or 30 dollars an hour to sit there, right? And you know, this lasted probably 60 or 90 days into a market launch, and then after, you know, it became clear that the driver was making 20, more than 20 or 30 dollars an hour, We removed that guarantee and sort of let the marketplace float naturally. We also did things like we put drivers near places where we knew it would have high demand in cities. Um, and then we made it really easy on the demand side to refer your friends. So if you were riding in a car with somebody and they hadn't used Uber yet, by the time you got out of the car, you would have referred them and you each would have gotten like 10 dollars off your next ride.

AI assessment note: “we paid a driver 20 or 30 dollars an hour to sit there”

Q kind of core question that everyone's asking right now is, does more compute equal an increased level of performance, or have we reached a point where it is misaligned, and more compute will not create that significant spike in performance? Kevin Scott at Microsoft says, absolutely, we have a lot more room to run. Why are you skeptical? And Have we gotten to a stage of diminishing returns on compute?

A So if we look at what's happened historically, the way in which, uh, compute has improved model performance is with companies building bigger models, right? So I, in my, in my view, at least the biggest thing that changed between GPT, 3.5 and GPT four was the size of the model. And it was, you know, also trained with, uh, more data, presumably, although they haven't made the details of that public and more compute and so forth. So I think that's running out. I think we're not going to be, we're not going to have too many more cycles, possibly zero more cycles of a model that's, you know, almost an order of magnitude bigger in terms of the number of parameters than what came before, and thereby more powerful. And I think a reason for that is data becoming a bottleneck. These models are already trained on essentially all of the data that companies can get their hands on. While data is becoming a bottleneck, I think more compute still helps, but maybe not as much as it used to. And the reason for that is that, uh, perhaps ironically, more compute allows one to build smaller models with the same capability level. And that's actually the trend we've been seeing over the last year or so. As you know, you know, the models today have gotten somewhat smaller and cheaper than when GPT-IV initially came out, but with the same capability level. So I think that's probably going to continue.…

AI assessment note: “more compute still helps, but maybe not as much as it used to.”