The Exchanges

Q Do you have to manage them in a different way? Maybe they're a little bit more larger egos. They like to feel like they're in control. Is there any different elements in how you manage them?

A Yeah, I mean, I think you have to let them cook. I mean, you, you want to give them enough band. If you, if you hired them and you trust them and they can execute, get out of their way, let them go experiment, right? I mean, you don't want to put guardrails over their ability to experiment and try different things. Guardrails is going to hurt everyone. It's going to make everyone think a little bit smaller. You remove those guardrails. Everyone's thinking big. One person at the company goes, takes a big swings, runs a TV ad with Jim Harbaugh. Everyone's now thinking at that level of, okay, how do I, how do we go bigger? And that's the kind of culture you want to create.

AI assessment note: “I think you have to let them cook. I mean, you want to give them enough band”

Q I love that. But you were at an MBA program, and a lot of people are at MBA programs who want to move into startups. You founded a startup. What happened there? What's your advice to others in MBA programs wanting to get into startups?

A Yeah, MBAs have such an interesting reputation in tech. Most people just absolutely hate them. I personally, even though I left a year early in the NBA, I had a great experience, right? Like I got there. I came to tinker with ideas. I never took an interview, started playing around with all my different ideas. You can get funding from the school. So it's essentially like a pre pre seed and you're just tinkering full time and investors start conversations start flowing and it became very clear. Like I'm going to learn a lot more. I'm going to have such a better experience. If I just go full steam ahead, On this idea. And, and that was it. I mean, it was, it was amazing.

AI assessment note: “You can get funding from the school. So it's essentially like a pre pre seed”

Q Why do you say TikTok's a crazy beast of a place? I agree with you. I don't know. I can share thoughts around that, but why do you say that?

A You know, the, the algorithm is so different than how we think about traditional algorithms, where it's built for virality. Uh, so if you're trying to do organic TikToks, you're comfortable with, you know, 10 failing, but one hitting seven million views. And like, that's almost like, it's, it's such a crazy power law. Um, On the, on the TikTok ad side, like, I think we are learning a lot around how AI companies need to be positioning themselves with new audiences that don't care about AI. Like you can't walk up and be like, well, we're an AI search engine and people, people are going to be like, oh my God, that's what I've been waiting for my entire life. Like you need to abstract away this whole AI piece and just go straight to the value prop of get answers immediately backed by sources.

AI assessment note: “the algorithm is so different than how we think about traditional algorithms”

Q So when you were brought in, did you have this team built out or did they say, hey, we want you to build it out yourself?

A It's a very flat organization. So there was nothing built out, but there were resources to pull from, right? You, you know, there, there are engineers that are ready to jump in on, okay, let's go hack onboarding or let's go really iterate on deeper retention or, or anything like that. But And it depends on when you are starting, but a lot of times it's, well, what even exists today? And your first few weeks stepping into growth is you're just turning over stones. You're trying to build the map of the world. Okay. What's happening in acquisition, what's happening in engagement, what's happening in monetization. And usually when you ask those questions, you don't get any answers and you just get a hundred more questions that you have to now go figure out. So then your job becomes, let me build the right infrastructure. So we can actually measure and understand the world. And that's where you start getting into engineering, logging, attribution. Do we have first party cookies? All of that.

AI assessment note: “there was nothing built out, but there were resources to pull from”

Q What has been the single biggest needle mover? Aaron said that you drove the numbers up by 10 to 15%. What has been the single biggest needle mover in doing that?

A I think there have been a couple. We've worked on a lot of different things. One is, can you makeshift audience? So we've done a lot of audience targeting. Around specific high quality cohorts. Two is, can you mix shift platform? I don't want you just on the web app, or I don't want you just on the mobile app. Can I get you across multiple devices? And that cross device conversion is really, really strong for retention because now you're using it on weekends. You're using it during the day. We're everywhere. We're integrating into your life. Those are the two really big ones. And then you never can just underestimate driving product improvements is always going to drive retention.

AI assessment note: “One is, can you makeshift audience? So we've done a lot of audience targeting.”

Q Why do you say TikTok's a crazy beast of a place? I agree with you. I don't know. I can share thoughts around that, but why do you say that?

A You know, the, the algorithm is so different than how we think about traditional algorithms, where it's built for virality. Uh, so if you're trying to do organic TikToks, you're comfortable with, you know, 10 failing, but one hitting seven million views. And like, that's almost like, it's, it's such a crazy power law. Um, On the, on the TikTok ad side, like, I think we are learning a lot around how AI companies need to be positioning themselves with new audiences that don't care about AI. Like you can't walk up and be like, well, we're an AI search engine and people, people are going to be like, oh my God, that's what I've been waiting for my entire life. Like you need to abstract away this whole AI piece and just go straight to the value prop of get answers immediately backed by sources.

AI assessment note: “the algorithm is so different than how we think about traditional algorithms”

Q bad idea. You know, I'm a VC for a reason. Uh, so, I, I, Arvind also said, uh, usual growth PMs would ask for marketing budgets that slam ads on TikTok and Instagram. He said that you first wanted to ensure that you had good CAT to LTVs before throwing money at acquiring users. What did you want to see in the CAT to LTVs that was important to you?

A I think there are a couple of pieces here. One is like, should we even be focusing on monetization right now? Right? Do we, do we want to think about paid acquisition as a lever that, that needs to bring in users and then we see that payback happen immediately? And my answer early on was no. I think we just need to go focus on top of funnel, new user growth and getting incredible sticky retention. And if we do that, guess what? We're going to be able to figure out how to monetize these users, whether it's through a consumer subscription, whether it's through ads, whatever it is, we're going to be able to figure out that monetization. So let's put down cocktail TV for now. And let's the only metric I'm going to just yell about all day is our activation rate and our retention rate and hold me to moving that.

AI assessment note: “So let's put down cocktail TV for now. And let's the only metric”

Q What did you not do that you wish you had done?

A This is an in-product one. So, I think, I don't think sharing is inherent to the perplexity user journey or user flow. Like, I don't think that we have an inherent viral loop where someone does some research and then shares a thread or screenshot to another person. But there's a lot of sharing happening on perplexity right now, right? Take an answer and put it on Twitter or send it to Harry. Whatever it is, we see a lot of traffic come in through sharing. The more I really, really want to get my hands on that experience and just drive sharing and find creative ways to promote sharing of this knowledge that you are curating as a user.

AI assessment note: “drive sharing and find creative ways to promote sharing of this knowledge”

Q What did you not do that you wish you had done?

A This is an in-product one. So, I think, I don't think sharing is inherent to the perplexity user journey or user flow. Like, I don't think that we have an inherent viral loop where someone does some research and then shares a thread or screenshot to another person. But there's a lot of sharing happening on perplexity right now, right? Take an answer and put it on Twitter or send it to Harry. Whatever it is, we see a lot of traffic come in through sharing. The more I really, really want to get my hands on that experience and just drive sharing and find creative ways to promote sharing of this knowledge that you are curating as a user.

AI assessment note: “really, really want to get my hands on that experience and just drive sharing”

Q Have you ever had an A-B test mislead you?

A A lot of times you see A-B tests with mixed results. So, and that, those are the trickiest ones to navigate. So let's say we're working on, okay, I want to get a user to hit their, you know, fifth query in the session, then I'm going to show them a sign in. Opportunity. Well, I'm showing it a sign-in opportunity. I'm now juicing activations. Great. But a lot of people saw that sign-in gate and they hated it. They were like, screw this. I'm out of here. And now I'm hurting query volume. I'm hurting searches happening in perplexity. So now I have two different metrics to, to hold on to. How do I value those two things down the road? That is where it gets really, really fun for some of us where it's like, okay, well, how, how do we value this? Will that activated user retain? And over time, those retained metrics, if I think about this on a 30 day timeframe or a 60 day timeframe will make up for that initial loss in query volume. I'm not going to go run that AB test for 60 days and then come back and tell the team, okay, we finally made a decision. You kind of have to make a decision right there or figure out how to optimize and like protect that downside.

AI assessment note: “A lot of times you see A-B tests with mixed results.”

Q What brand marketing thing do you not do today that you'd love to do? Sponsor a stadium, be on a shirt for a football team, be on an F-one car.

A I get a lot of these emails all the time. Um, You know, I mean, we're, we're doing a lot. What's cool about Perplexi is we don't have a marketing department. So we're, we're doing so many cool things and all of these are just someone on the team had a passionate, was really passionate about an idea that we believe has viral potential. And so if so, like go execute and make it happen. And the metric that matters in brand marketing for me is there's an expected number of impressions that anyone will tell you you're going to see if you go run a TV campaign or, or sponsor a jersey. And what really matters is what is the incremental impressions you're getting from that? Can that moment go viral where, okay, we may be paid for X amount of impressions, but we're getting a lot more on top of it because people are reposting about it because it has gone viral across social. That is where brand marketing can be really interesting.

AI assessment note: “the metric that matters in brand marketing for me is”

Q What brand marketing thing do you not do today that you'd love to do? Sponsor a stadium, be on a shirt for a football team, be on an F-one car.

A I get a lot of these emails all the time. Um, You know, I mean, we're, we're doing a lot. What's cool about Perplexi is we don't have a marketing department. So we're, we're doing so many cool things and all of these are just someone on the team had a passionate, was really passionate about an idea that we believe has viral potential. And so if so, like go execute and make it happen. And the metric that matters in brand marketing for me is there's an expected number of impressions that anyone will tell you you're going to see if you go run a TV campaign or, or sponsor a jersey. And what really matters is what is the incremental impressions you're getting from that? Can that moment go viral where, okay, we may be paid for X amount of impressions, but we're getting a lot more on top of it because people are reposting about it because it has gone viral across social. That is where brand marketing can be really interesting.

AI assessment note: “I get a lot of these emails all the time.”