Perplexity's Head of Growth on the Search Engine's Biggest Growth Experiments | Bolt Recommends

线性资本·November 21, 2024

Growth driven entirely by word-of-mouth, no paid promotion, community channels are worth investing in

(This article is republished from Founder Park)

Perplexity recently raised another $500 million, bringing its valuation to $9 billion.

The company is also ramping up its commercialization efforts. Last week it announced it's experimenting with ads on its platform in the form of "suggested questions." This week came news that it's moving into e-commerce, rolling out shopping features for paid users that recommend products within search results and let users purchase without leaving the app.

As a flagship AI search product, how Perplexity approaches monetization and product growth is something many people are watching closely.

Recently, the popular podcast 20VC interviewed Perplexity's Head of Growth Raman Malik, an early member of the growth team at Lyft (the second-largest ride-hailing company in the US). Malik shared substantial insights on Perplexity's user growth, paid advertising, brand marketing, and hiring for growth teams. Founder Park has compiled and edited the podcast content.

Some highlights:

A well-defined A/B test tells you what to do next. It adjusts your direction and pace. A bad A/B test is poorly scoped and gives you nothing.
For Perplexity, if I can get a user to do three queries in their first session, I know I've found something important.
Most of our growth has come through word of mouth, and that's the best kind of growth.
Students are a demographic I really want to crack. We need to find the right partnership approach to reach student users at scale.
By month six, 45% retention is excellent. That's what we're shooting for, especially for a non-social product.
It's crucial that people understand Perplexity offers real-time search. We're searching the live internet to give users the freshest answers. That matters enormously.
For users who don't care about AI, you need to abstract away the whole AI search engine concept and just communicate the value: get answers instantly, with credible sources.
Community is an overlooked channel. Tapping into your core users and having them share the product's value with their small communities is incredibly powerful.
Our competitors are Google and ChatGPT. The search market is a massive target. If there's anywhere we want to establish ourselves, it's competing against the "big guys."
The best thing about recruiting founders is they're more accustomed to taking big risks and comfortable with failure.
For the team, if a new hire needs a detailed onboarding guide, we probably have a problem.
The biggest mistake founders make: assuming user pain points that don't actually exist.

01 A/B Testing Is Overrated

20VC: Is growth a game of micro-optimizations — like moving 90-day retention from 30% to 35% — or should you go straight for bold, potentially game-changing moves?

Raman Malik: I think micro-optimizations are actually severely underrated.

Anyone who's built a company growth model — you open an Excel spreadsheet with new users, retained users, reactivated users, and churned users. New users plus retained users plus reactivated users, minus churned users, gives you your weekly or monthly active users. If you assume a new user growth rate or retention curve, then experiment: what happens to weekly and monthly actives if I improve retention by 10%? That tiny 10% change in retention or new user activation rate lifts your entire active user base. That small micro-optimization improves the company's entire foundation.

But micro-optimizations also hit diminishing returns. At some point it's hard to squeeze out another 10-15% improvement without massive experimental effort. That's why my rule of thumb is to do several bold bets per quarter — a major new feature, or a large marketing campaign. It has to be high-risk, high-reward. I have to be able to stand in front of the company and say I'm doing this, willing to own the risk of failure, then narrow scope from there.

20VC: What should the success rate be on these big bets?

Raman Malik: It depends how much time you're willing to invest. I think 25% is pretty good for these major initiatives. So maybe once a year we'll have one campaign or new feature that really moves the needle or opens up a new audience for us. That's hugely important.

20VC: We talked about optimization, about extracting as much value as possible. One cool way to do that is through A/B testing — figure out what works, what doesn't, then double down. What role does A/B testing play in growth today? How do you think about it?

Raman Malik: My background is in data science, so I may be biased here.

A/B tests can be good or bad. A/B testing can be used to understand impact, to understand changes in each metric. But did we inadvertently degrade certain metrics? Do we truly understand what's incremental versus what's not? A well-defined A/B test tells you what to do next. It adjusts your direction and pace. A bad A/B test is poorly scoped, and you can't extract any information from it. It's a mess, the minimum detectable effect is too low. You run it and get nothing, just waste a week. The worst is running an A/B test just to show in a performance review that I moved some number, or to report to a manager. As companies get bigger, you see this — we run A/B tests just to be able to say "oh, I improved activation by 12%."

20VC: Do product managers really need A/B testing? Because what's working and what isn't is often obvious.

Raman Malik: I think if your goal as a PM is to optimize direction, to confirm we're doing the right thing, A/B testing gives you quick feedback. That said, when you start an A/B test you should have a pretty strong sense of what might happen. That's where intuition comes in — we're running this test, I think this will happen, we'll really move this metric, but we might see other metrics dip slightly, and that's what we need to watch for. If you don't have that intuition, you're flying blind.

20VC: Has A/B testing ever misled you?

Raman Malik: Sometimes you see inconsistent A/B test results. Those are the trickiest situations.

For example, we wanted to show a login prompt after a user's 5th query to boost activation. But a lot of people hate seeing a login requirement, they say "screw this" and leave. That hurt query volume, reduced searches on Perplexity. So now I have two different metrics to weigh. How do I evaluate the long-term value of these two things?

This is where it gets really interesting for some of us — how do we evaluate this? Will this activated user retain? Over time, if I think in a 30 or 60-day frame, will these retention metrics make up for the initial search volume loss? I'm not going to run that A/B test for 60 days and then come back to the team and say "okay, we finally have a decision." You have to make the call on the spot, or find a way to optimize and protect the downside.

02 Perplexity's Key Metric: Searches in the First Session

20VC: I think about Facebook — they wanted you to get 5 friends, then you had a higher likelihood of retention. Do you have a specific internal metric where it's like: "Hey, once they cross this threshold, we have a much higher likelihood of retention"?

Raman Malik: Yes, I call these milestone metrics.

We can look at any user retained at 30 days, 60 days, then look back at their first few sessions and ask: "Okay, what did they do differently? What was unique about their experience?" From this analysis we can build an actionable metric. It's not perfect because this isn't deep retention. But for Perplexity, if I can get a user to do three queries in their first session, I know I've found something important. Because this is something many quality users have done. The user has spent enough time in the product to now understand Perplexity's value, and they're more likely to come back and keep using it.

20VC: You said 2 to 3 queries in the first session. I had Alec Schultz, a growth veteran from Meta, on the show. He said the biggest problem in growth is that every goal or metric can be gamed. If you want to hit that target, we can stuff a bunch of suggested searches for you to click — technically hits the target, but doesn't actually mean you're getting what you want. How do you think about setting the right goals, given that every metric can be gamed?

Raman Malik: First is monitoring. Monitoring whether this metric is still correlated with long-term goals. If we see the correlation between the metric and the goal start to break, then okay, I have to be honest with myself about whether we're actually helping improve retention.

The other thing is taking a holistic approach. You can't just stare at one metric. I need to watch query count, but also first-session duration. Did the first session hit 10 minutes?

20VC: Why? Wouldn't you want it to be shorter, meaning time-to-value was incredibly fast?

Raman Malik: Because you got so much value early that now you're down the Perplexity rabbit hole. If one query sparks a chain of follow-up questions, that's fantastic. The user can go deep with Perplexity, learning and exploring. That's great for us.

03 Word of Mouth Drives 80% of User Growth

20VC: When it comes to user acquisition, what's the most important lesson you've learned since joining Perplexity? What's the composition of Perplexity's user sources?

Raman Malik: AI companies today get a ton of "curiosity traffic" — as long as a product looks "magical" or has a slick demo, users get drawn in and try it quickly. Most of our growth has come through word of mouth, and that's the best kind of growth. Even now, two years in, we're still enjoying the dividends from this organic traffic, which is really remarkable. The next question is: how do we maintain this "organic magic"?

20VC: What's the current user acquisition breakdown roughly? For instance, what percentage is word of mouth?

Raman Malik: Word of mouth accounts for roughly 80%, though the exact figure depends on which countries or international markets we're prioritizing. Beyond that, we also drive distribution through some large-scale partnerships.

External partnerships have been extremely effective for distribution. We could push Perplexity hard ourselves, but it works much better when someone else recommends it, or when Perplexity is bundled with an existing product — that drives more users to try it out. Our partnerships team is very strong.

20VC: But doesn't that have a big impact on unit economics?

Raman Malik: To some extent, yes, but it's part of acquisition cost. We're willing to trade a month of Pro membership. Because if the user doesn't use Pro, we haven't lost anything. But if they do, we have a chance to convert them into a loyal user. What matters most is whether we can ultimately get them to pay.

20VC: So what's the partnership you most want to land but haven't yet?

Raman Malik: We already have some fantastic partnerships, but I'm more focused on target audiences. I'm thinking about which groups we want to reach, and how to use partnerships to get to them. Students, for example, are an area I really want to crack — we need to find the right partnership approach to reach student users at scale.

20VC: Can you share any mistakes you've made on user acquisition, and what you learned from them?

Raman Malik: Too many. If you're not making mistakes, you're not learning fast enough. Early on, I was very interested in influencer marketing — I wanted to try promoting Perplexity through content creators. But I underestimated how difficult it is to work with these creators, including producing specific content and delivering precise messaging. Plus, Perplexity isn't an easy concept to explain. The words "AI search engine" aren't enough to move people. So working with a large number of small and mid-sized creators produced mediocre results and consumed a ton of time. Later, we shifted to deep partnerships with a select few creators, and that worked much better.

04 The Target Is 45% Retention

20VC: When you joined Perplexity, I spoke with Aravind (Perplexity's CEO) and he mentioned you quickly identified that the problem wasn't user acquisition, but user retention. Can you walk through how you diagnosed that, and what the context was?

Raman Malik: When I first joined, I mapped out the entire funnel end-to-end, from acquisition to monetization, trying to identify where we most needed improvement. Dimitri on the team and the business development team were already driving massive traffic through large partnership programs. But I found that the key lay in early activation rates and Week 1, Week 2 retention. While overall retention was already decent, improving these early metrics would expand the active user base. That became the focus of my work going forward.

20VC: What do you consider a good activation rate?

Raman Malik: It really depends on the company. With Perplexity, you can experience the product without registering. That's great because it allows users to try it quickly. Users can type in a question and immediately see value. But converting from non-logged-in visitor to registered user is a challenge. I want to push that conversion rate to 30% — get 30% of users to become registered users.

20VC: On retention, I think a lot of people don't really understand what counts as good. Can you explain?

Raman Malik: Consumer products are genuinely hard. The key is that the retention curve needs to flatten out. I think 45% retention by month six would be excellent. That's what we're aiming for, especially for a non-social product, because we don't have that "smile curve." So we need to work hard to create sustained value and increase usage frequency to hit retention above 45%.

20VC: Are you surprised by user churn?

Raman Malik: No, I actually think it's pretty good. Our retention is still quite high, which shows the product is genuinely delivering value.

20VC: Duolingo's retention is something like 50% after 12 months, right?

Raman Malik: Yes, 50% retention after 12 months — that's absolutely world-class. They've built an incredibly powerful user retention system.

20VC: If retention is a problem but our acquisition is performing well — say we're driving growth through partners — how do you solve for that?

Raman Malik: The methods for solving retention problems are actually limited.

First, you need to define key milestone metrics — what behaviors correlate with long-term retention. For example, we found that users completing three queries in their first session is a very important metric. So we focus on driving that behavior.

Second, you can optimize channel mix. Generally, users acquired through organic growth and referrals retain better than those acquired through paid channels. Since so much of Perplexity's growth comes from word of mouth and partnerships, we have limited room for optimization on the channel side.

Finally, you can target different audiences. Perplexity's user base is very broad — my grandmother uses it, doctors use it. So we can identify high-retention user segments, then specifically target those users in our marketing, thereby improving retention among new user cohorts.

20VC: Which user segments have the highest retention? Which have the lowest?

Raman Malik: A lot of trial-type users have lower retention — people who search things like "how many times does a person fart in a day" out of curiosity. Those users churn at higher rates. Users who genuinely want to fill knowledge gaps — people using Perplexity for work, students using it to find academic papers — those users retain very well. Because we're solving real problems for them, not just satisfying curiosity.

05 The Core Value Has Never Been Communicated Well

20VC: I saw Aravind recently mention that Perplexity's average query length is longer than Google's — something like nine words. How does this data impact your acquisition and retention strategy?

Raman Malik: It has significant impact. The more users type, the better we understand their needs, and the better answers we can provide. We can use this information to optimize search results and improve answer quality. So query length and information density are very valuable to us.

20VC: What optimizations do you think have moved the needle most on improving your metrics? Aaron mentioned you saw 10-15% improvements in your data.

Raman Malik: Mainly two areas:

One is user segment optimization. We focused on targeting and acquiring high-quality user segments.

Two is platform distribution optimization. I don't want users to only use the web or only use mobile — I want them to use Perplexity across devices. Cross-device conversion helps retention significantly, because users engage on weekends, weekdays, and Perplexity becomes more deeply embedded in their lives. Of course, product improvements are always the foundation for improving retention.

20VC: What strategy did you think would work well but actually didn't?

Raman Malik: Messaging optimization — like the onboarding when users first open Perplexity. We've been trying to help users quickly understand Perplexity's core value: "real-time updated data" and "sourced answers." This messaging is very important, so we keep experimenting with copy and user experience. But the results have been mediocre.

20VC: I like to play this game where I go on Shopify's website, find problems, and roast their page design live — and they actually changed it. Now, looking at Perplexity's description: "A free AI-powered answer engine that delivers reliable, accurate, and real-time answers." What do you think of that?

Raman Malik: I'd give it a 6 out of 10.

20VC: A bit wordy, right?

Raman Malik: Yeah, a bit wordy. But it's actually optimized for SEO.

20VC: But when I visit Perplexity's homepage, all I see is "Ask anything, explore the unknown."

Raman Malik: Right, so it's a balance. We need to consider how much information to show, the balance between functional and brand-oriented messaging. Our current direction is toward simplified language — "Ask anything," "Search like never before" — while highlighting Perplexity's real-time search capabilities. Getting users to understand that we provide real-time, updated answers is important.

20VC: Let's say this is your platform and you can choose whatever you want. What would you do?

Raman Malik: On the homepage, I'd want very simple language — something like "Ask away" or "What do you want to know?" We're moving toward "Search like never before." I think it's crucial that people understand Perplexity provides real-time search — we're searching the live internet in real-time to give users the latest answers. That point is essential.

06 Perplexity Is at the Point Where It Should Measure CAC/LTV

20VC: I heard Aravind mention that typically growth team PMs ask for big marketing budgets to run ads on TikTok and Instagram. But he said you wanted to ensure good CAC (customer acquisition cost) to LTV (lifetime value) ratios before scaling paid spend. What did you want to see from these ratios?

Raman Malik: There are a few key points here. First, should we be focused on profitability now? Or are we doing paid user acquisition and expecting to see immediate returns on investment? My answer: it's not time yet.

I believe we should focus on top-of-funnel new user growth and ensure extremely high user retention. If we can do that, then whether we monetize through subscriptions or advertising, we'll find the right approach. So for now, don't focus on CAC and LTV. The two metrics I care about every single day are activation rate and retention rate. Make sure those are growing.

20VC: So when do you think CAC and LTV should become more important metrics?

Raman Malik: It's about time now. Look, Perplexity has barely optimized anything to drive subscriptions.

20VC: How could you optimize that?

Raman Malik: For example, adding more usage limits. Free users might only get to use Pro Search or file upload three times a day, but if they really love those features, we can nudge them to upgrade to Pro for unlimited access. These limits and upsell tactics are all adjustable.

20VC: CAC and LTV are usually quite volatile. CAC changes significantly as your user base matures, and LTV takes a ton of data before you can spot any trends. If you were advising founders, how should they think about the importance of CAC and LTV?

Raman Malik: It depends on the company's stage. If you're at a point where you need to scale fast and you're confident about retention, or if your efforts to improve activation and early retention are hitting diminishing returns, then it's time to start optimizing for CAC and LTV. That's when we shift our focus from retention to CAC and LTV.

07 Don't Mention AI in TikTok Marketing

20VC: Since your user growth is mostly driven by flywheel effects — like Jensen Huang saying you're his favorite search tool — why do any paid advertising at all?

Raman Malik: We barely do any paid marketing, just occasional experiments. This kind of advertising can boost your top-of-funnel numbers, which looks great. But over the long term, the retention performance is terrible, and perhaps most importantly, it's not incremental.

I started my career on Lyft's paid acquisition team, where we spent millions on these channels. I remember vividly: our head of growth one day said, shut off all paid channels targeting new users. Paid marketing was a huge chunk of our user growth, but when we turned everything off, signups and installs barely moved — maybe down 5-10%. And that drop was almost entirely low-quality users. In other words, every dollar we spent acquiring those users was basically just cannibalizing organic traffic. Zero incremental effect.

20VC: So why do paid ads at all?

Raman Malik: Let's look at Lyft from another angle. On the rider side, paid acquisition had very low incrementality. But on the driver side, it was completely different. Acquiring drivers in the ride-sharing market is extremely difficult, and as a two-sided marketplace, you need supply-demand balance. So we spent heavily to get drivers, which balanced the market and improved overall efficiency. And the payback on that spend came much faster.

Also, if the unit economics work for an individual user, it can make sense to reach uncovered segments through paid promotion. Right now we're doing very limited paid ads, testing some marketing approaches on TikTok targeting 18-24 year-olds.

TikTok is a truly insane place, my god.

20VC: Why is TikTok an insane place? I agree with you, and I can share some thoughts later — YG said the same thing.

Raman Malik: You know, TikTok's algorithm is completely different from traditional algorithms we understand — it's designed for virality.

If you do native TikTok content, you might have 10 failures, but one hits 7 million views. So the ad effect on TikTok is this absolutely insane power law.

We're learning a ton, especially around how AI companies should position themselves with new audiences who aren't interested in AI. You can't say, "we're an AI search engine" and expect people to go, "oh my god, I've been waiting for this!" You need to abstract away the whole AI search engine concept and just deliver the value: get answers instantly, with reliable sources.

20VC: Yes, completely agree. I feel like AI actually has a bit of a negative effect in product marketing right now. My toothbrush claims to be AI-powered, though I have absolutely no idea how it does that, but that's what the packaging says. Anyway, the problem is, frankly, sometimes slapping the AI label on a product actually devalues it.

How are conversion rates on TikTok?

Raman Malik: They're decent. We're still in testing mode, but overall conversion is solid, top-of-funnel ad performance is good. Right now I'm waiting to see some retention data.

20VC: How much are you willing to spend on ads right now?

Raman Malik: Not much, actually. You don't need to spend a lot to learn a lot.

20VC: Specifically, $10k, $20k, $50k?

Raman Malik: $10k a week. That budget lets us learn a ton — especially what messaging works and what doesn't, or what use cases really resonate with people, whether it's travel planning, studying, or problem-solving at work. We can test many different directions because our product is so broad.

08 Brand Marketing Should Deliver Incremental Impressions

20VC: Do you like brand marketing?

Raman Malik: I believe people need to hear about your product 3 to 7 times before they're actually willing to try it. So I really believe in that. Measuring brand marketing value is hard, but just because we can't quantify it doesn't mean it's useless.

20VC: What's something in brand marketing you aren't doing today but would love to do? Like sponsoring a stadium, or putting your logo on a football jersey, or on an F1 car.

Raman Malik: I get those emails all the time (laughs). Actually we do a lot of things. We don't have a dedicated marketing department — a lot of the time it's team members who have a passionate idea, think it has viral potential, and just go execute it. For me, the most important metric in brand marketing is: you can estimate how many impressions you'll get from a TV ad or jersey sponsorship, but what really matters is how many incremental impressions you get from it. Like, will this ad go viral on social media, with people sharing and commenting, creating much bigger distribution effects. That's what's most interesting about brand marketing.

20VC: I completely agree. A friend of mine, a founder in our portfolio, started a recruiting platform a few years ago that brought banking employees from places like Goldman Sachs to startups. Then they put up a billboard across from Goldman Sachs' London headquarters that said "Goldman Sachs? Really? I bet your parents are proud." And every newspaper in London covered this startup vs. big bank story. That incremental effect, as you said, was massive.

Raman Malik: Right, especially when you're reaching completely new audiences. We did a TV ad with Jim Harbaugh, a famous American football coach. After it aired, lots of sports influencers and journalists shared and commented on it. This was a whole new audience segment, right? Sports fans sitting at bars watching games — they have tons of niche questions like "who won that game 12 years ago?" Perplexity is perfect for them. We're introducing ourselves to new audiences we could never otherwise reach, which is really great, very effective.

20VC: What's hardest about doing paid ads?

Raman Malik: There are so many different channels to test, and I don't want to waste time on channels where I won't learn much. I've had this experience before, like with newsletter sponsorships. I thought, newsletter audiences are interesting, let's try it. But unless you have a very clear product and can distill it into extremely concise ad copy — two lines of text max — that channel is really tough. So we tried it, but results weren't great, didn't get much insight from it, basically a waste of effort.

20VC: So you're willing to spend money as long as you learn something, even if results aren't great?

Raman Malik: As long as we're finding the right direction, we can spend to keep exploring. All we need to do is keep learning, figure out the right direction, then maintain the pace and keep pushing in that direction.

09 Social Channels Are Saturated; Community Is the Overlooked Channel

20VC: Last time we talked, you told me about capturing first position in a growth channel, or doing it better than anyone else. I think that's a really nice hedge, frankly, because being first doesn't always mean you win. Can you explain that thinking? I'd love to understand.

Raman Malik: Right, exactly — there's no middle ground right now. Everyone in growth is looking for alpha — finding undiscovered channels and capturing them before others. Dropbox did it through referral programs, coupons, and email. Pinterest through SEO. Duolingo through completely unconventional social media strategies. All these companies grabbed that concept early and then succeeded. So as a head of growth, you're constantly searching for that opportunity: what channels aren't being exploited at scale right now?

The problem is a lot of channels have become somewhat saturated. Facebook and Instagram ads are very expensive, Apple updates have made paid ads harder, organic social traffic is increasingly difficult to get, notifications are crowded — everything feels very crowded right now. There's no middle path. If you're not first, the only option to capture a channel is to do it better than everyone else. That's exactly what we need to do now.

20VC: What channels do you think are overlooked but deserve more attention?

Raman Malik: Community. This answer might be controversial. But I believe tapping into the potential of core users through community, letting them share the product's value with their small social circles — that approach is incredibly powerful.

20VC: It really is. How do you actually do it?

Raman Malik: Take students as an example. Many students use Perplexity and get tremendous value from it, but compared to ChatGPT, our awareness among students is clearly lacking. So the question becomes: how do we help those core student users, give them everything they need — marketing budget, swag, whatever — so they can drive Perplexity's growth on campus and build density there. We need to arm our "troops," provide these core users with all the tools they need to grow for us. This strategy is very powerful because these students are right there on campus, and campuses are small, so you can build density quickly. Once initial density forms, word-of-mouth effects start kicking in.

20VC: Who do you think is your biggest competitor?

Raman Malik: Obviously, Google and ChatGPT. Search is a massive market. If there's anywhere we want to establish ourselves, it's competing against those "big players."

20VC: Where can you do better? Of the channels you've already tried, which ones do you feel need improvement?

Raman Malik: Fortunately, we're not in too many channels right now. We're still primarily reliant on organic growth and partner-driven growth. I'd love to find more efficient ways to activate content creators and opinion leaders. It's not just about placing ads in their newsletters or podcasts — it's about finding genuinely creative ways to make collaborations truly "come alive."

20VC: I completely agree with you. Content has to do one of two things: either deliver knowledge or demonstrate deep understanding. If it does neither, it shouldn't be on any of our platforms. So much social media content now, like "excited to announce the launch of X product," is completely substance-free.

Raman Malik: Yes, that kind of content has zero value. I'd rather try to entertain people, even if it doesn't land perfectly, than put out meaningless stuff. Just introducing "this is Perplexity" — that approach never works.

20VC: What growth tactics have you tried that you now feel you shouldn't have?

Raman Malik: I'm still regretting those newsletter sponsorships.

20VC: Conversely, what haven't you done that you wish you had?

Raman Malik: This is a product-related question. I don't think "sharing" is woven into Perplexity's user flow. There's no built-in viral mechanism in our UX, like users doing research and then sharing results as posts or screenshots. Of course, plenty of users are sharing Perplexity right now — posting answers to Twitter or sending them to friends. And we've observed that traffic from sharing is quite substantial. I really want to optimize this sharing experience, finding innovative ways to encourage users to share knowledge they've curated themselves.

10 Build the Growth Team After Finding PMF

20VC: Going from founding your own startup to joining Perplexity — going from running your own company to joining a team is a completely different experience. Can you walk me through how that happened? Did Aravind (Perplexity's founder) come to you and say "Hey, we need a Head of Growth, you in?" or something like that?

Raman Malik: I spent three years grinding on my own startup. Entrepreneurship isn't easy at all. As a founder, month after month goes by, you keep pivoting, but things just aren't moving — that feeling really crushes your morale. So I started talking to different companies and quickly realized Perplexity was a company growing incredibly fast and doing really interesting things. I thought, I want to go there.

The transition from being a founder is genuinely different. First, you already know deep down how hard entrepreneurship is. Second, you're coming right off a failure, feeling pretty beaten up. So you need a complete mental reset. And this transition is actually about narrowing your scope. I was hired to own growth. You know, founders do all kinds of work, handle a hundred different things, steer the whole ship. But now you're focusing on one domain: owning growth. I narrowed my scope down to just doing this one thing well. That transition takes adjustment too.

20VC: You're Head of Growth now — that title can mean so many different things, it's quite vague. How do you specifically define the Head of Growth role?

Raman Malik: I see growth as a distinct function bridging product and marketing. There are growth product and growth marketing sides.

Growth product first — this requires product managers, engineers, designers, data scientists working together. But the core product you're focused on is the user funnel: acquisition, activation, retention, and monetization. Growth isn't just about feeding new users into the top of the funnel; it's about getting them to stick around, develop into core users, so you can ultimately monetize. You're focused on driving this entire user lifecycle process — that's growth product.

Growth marketing has the same goal as the growth product team: driving users through the entire funnel from acquisition to monetization. But the tools are different. Instead of approaching it through engineering, design, or data science, you're using marketing channels, lifecycle communications, community operations, marketing campaigns.

These need to work tightly together. When I work on onboarding experiences, I need the product team constantly optimizing each flow, while the marketing team thinks about copy, email campaigns, and so on. All of this has to move as an integrated whole to drive the funnel.

20VC: These are mature teams that require significant investment. As a company, when is the right time to build out a dedicated growth function?

Raman Malik: It depends on company type, development stage, founder composition, and other factors. The simplest answer is after product-market fit (PMF).

20VC: Growth product and growth marketing teams actually cost millions. If your annual revenue is only $2-3 million, even with PMF, you can't afford that. You probably need at least $25 million in annual revenue to support that team.

Raman Malik: But I think as soon as you see early signs of retention improving — say, if you're seeing 30% retention by month three or four — that means your "fire is lit," and it's worth investing resources to make it burn brighter. That doesn't have to mean immediately opening up marketing channels and spending big. You can start by improving that 30% retention to 40%, making it easier for users to convert.

20VC: So when you joined, was this team already built, or did they have you build it yourself?

Raman Malik: Nothing was built at the time. But the organization here is very flat, and it's easy to pull in resources — engineers who can jump in anytime to improve onboarding or deepen retention. In the first few weeks of owning growth, it was mainly about understanding: how's acquisition going, how's engagement, how's monetization. Usually when you ask these questions, you don't get any answers — instead, you generate even more questions that need figuring out. So your work becomes building the right infrastructure so we can actually measure and understand where things stand. That involves engineering logs, attribution analysis, first-party cookies, all kinds of things.

11 Founding Teams Should Hire More Former Founders

20VC: Let's talk about team building. In growth team hiring, what do you see as the first critical role?

Raman Malik: It depends on your needs. Do you want a marketer who can expand the top of the funnel, design experiments, and optimize onboarding flows? Or do you need a product person who can own marketing campaigns? I personally lean toward the latter.

I like spending time studying data, digging deep into potential problems, holding onto the data tightly. This still depends on the current team's and founders' composition. If you're happy with the top of funnel state, and the funnel itself is complex — involving multiple device platforms — then you should find an experienced product person to handle these things.

20VC: Are founders suitable hires? You're a founder yourself, and you previously worked at Rippling, which also had many former founders join. What's your experience?

Raman Malik: We have many founders on our team. The best thing about recruiting founders is they're more accustomed to taking big risks and accepting failure. If you chose the entrepreneurial path, you necessarily tolerate uncertainty more than average. Most people aren't willing to risk failure, let alone stand in front of the company and say "we're going to try this approach, but it might not work." So the more people like this you have on your team, the better.

Mike Maples had a great metaphor: a startup team should be like a "jazz band" — throw away the sheet music, don't rigidly follow the rules, but improvise freely together, working to make things come together. I think founders fit well in this kind of environment. However, as the company grows, you may need "sheet music" and more coordination. Some founders miss that improvisational feeling, and at that point they might jump to the next "jazz band."

20VC: Do you need to manage them differently? Maybe they have bigger egos, like being in control. What's special about managing them?

Raman Malik: I think you need to let them run with it. You hired them because you trust them and believe they can execute, so you should let them try and experiment. You shouldn't impose too many constraints on them, because constraints make everyone's thinking more limited. Remove these constraints, and people start having bigger ideas. For example, someone in the company proposes a bold idea, like shooting a TV commercial with Jim Harbaugh — that gets other people thinking in more ambitious directions too. That's the culture you want to create.

20VC: What did you mean by "taste" as a hiring criterion? What kind of people are you looking for? And what are you avoiding?

Raman Malik: "Taste" means having a deep understanding of what "great" looks like. I believe I'm not actually particularly good at any one thing, but I can observe what's happening in the world, understand what excellent performance looks like, and reverse-engineer to achieve it. That's the key.

12 If New Hires Need an Onboarding Guide, Your Team Has a Problem

20VC: If I don't know how to judge whether someone has this ability, what should I do? Do I need to show them some products and have them analyze? How do you determine it?

Raman Malik: I think the key lies in the questions they ask — this reveals the depth of their thinking. I'm very candid in interviews about the challenges we face, exploring them together with candidates. By observing how they think about these problems and what experience they bring, you can quickly tell whether they can efficiently reverse-engineer excellent solutions.

20VC: What questions do you always ask in interviews? I usually have some fixed questions too.

Raman Malik: I love hearing candidates share "war stories," especially from people with some experience. For example, "What difficulties did you encounter working at Lyft? What attempts completely failed? Let's walk through those failures together, even laugh about them."

20VC: I always ask about their first entrepreneurial experience. Great entrepreneurs usually start young. There's a strong correlation between the amount of early entrepreneurial activity and later success.

Raman Malik: Yes, because it's almost a personality trait. And what you're trying to evaluate is precisely this trait. It usually manifests very early on.

20VC: Do you give growth team candidates take-home assignments?

Raman Malik: Every role gets one. I realize it's a big ask, so I try to keep it reasonable.

20VC: Can you give an example of what that assignment looks like? How do you implement it?

Raman Malik: We usually present two or three different scenarios. They're fairly open-ended, so candidates can choose which aspects to dig into. The goal is for them to show us their thought process and all the factors they consider along the way.

20VC: Are these take-homes specific to Perplexity, or about other companies?

Raman Malik: Specific to Perplexity.

20VC: So do you give them an actual task? Some would argue that's unfair because you have more context about the company than they do.

Raman Malik: I think that's a fair point. However, we typically give a hypothetical situation about Perplexity. For example, "We're considering launching a certain feature — how would you design the rollout?" Through this hypothetical scenario, we can brainstorm with the candidate and get a feel for what it's like to work with this person, and how quickly they adapt and integrate into the team.

20VC: How quickly do you realize you've made a hiring mistake?

Raman Malik: Usually pretty fast. One rule of thumb I have is, if they need a detailed onboarding guide, we probably have a problem. If I need to give them a checklist of "20 things to do in your first 30 days," something's off. I'd rather someone figure it out on their own, quickly put together their own onboarding plan. I tend to give people enough trust and space to take ownership of their area, start improving it, and do great work. Of course, I'll share some perspective on how we should implement things or what to prioritize, but I want to see them adapt quickly and proactively drive things forward.

20VC: I get that. I've found the best employees usually perform from day one. They're already doing things on their first day, right?

Raman Malik: Exactly. For instance, they get their laptop on Friday, aren't supposed to start until Monday, but they've probably already looked through all the dashboards. Part of it is excitement, but you want to find people who are completely immersed in the problem.

20VC: What's the biggest hiring mistake you've made?

Raman Malik: This is something I'm still learning. I've been too slow to hire, and I'm working on getting faster.

20VC: Why has your hiring been so slow?

Raman Malik: Because I'm very focused on team. A lot of people say employees leave because of their manager, but I think they leave the team. If a team operates like a well-oiled machine, where everyone respects and enjoys working with each other, then leaving is hard because work is just too much fun. That's the environment I want to build. I want people across the company to think, "I wish I could work on the growth team for a few months — that's the best place to be!" Finding people who can fit into that culture isn't easy, and sometimes I overthink it. So I'm working on speeding up my evaluation and taking more hiring risks.

20VC: Well I have to caution you — most people on this show say their biggest mistake was hiring too fast, that they wish they could "let the fire burn a little longer," so I'm not sure. Maybe waiting for the perfect fit is the better choice.

Raman Malik: Yes, it's a trade-off. It really is difficult.

13 Founders' Biggest Mistake: Assuming User Pain Points

20VC: What's the most common, costly, and irreversible mistake you see founders make on growth?

Raman Malik: Assuming user pain points that don't actually exist. They really don't figure out what users need during that first use. This is the most common problem, and it sends you in a completely wrong direction.

20VC: Which growth channel is most "crowded" right now?

Raman Malik: Right now, especially after Apple's latest privacy changes, I don't know who can still scale Instagram or Facebook ads efficiently as a core growth channel.

20VC: If your friend were appointed head of growth and started tomorrow, what advice would you give them beforehand?

Raman Malik: Deeply understand every single number in all the dashboards. Get as much data as possible. When I started in a junior role at Lyft, my only goal was to know the numbers better than anyone else.

When you start finding correlations between these numbers, you begin to understand how the entire growth mechanism works. When you understand the numbers better than everyone, you get invited to all the important meetings because you're "the person who knows the numbers." This helps you build intuition about the growth mechanism, identify friction points in it, and make better decisions about where to focus next.

20VC: What's the biggest success lesson you learned from Lyft?

Raman Malik: Don't rely on paid advertising. Focus on retention and product iteration.

20VC: What was the worst thing Lyft did that made you determined never to repeat it?

Raman Malik: We overestimated the potential to optimize the rideshare product. I think we hit diminishing returns pretty quickly. Rideshare products four or five years ago aren't that different from what they are now.

So if you think you can open up new markets by continuously optimizing the product, that's one direction. But if you find you're wrong, you should pivot quickly to other areas, like food delivery. We underestimated how quickly marginal returns would diminish on rideshare product optimization.

20VC: So the key is knowing when you hit that point of diminishing returns?

Raman Malik: Exactly right. This is especially hard in AI because the world is changing so fast every day. But the key is to be honest with yourself about the true market size. Are we still seeing meaningful growth through product iteration?

20VC: What aspects of growth would you most want to change right now?

Raman Malik: Just figuring out what growth actually is. You put 10 growth people in a room and everyone gives a slightly different answer about what we do, but it really encompasses both growth product and growth marketing. I think the more people understand that, the better.

20VC: What belief have you changed your mind on in the past 12 months?

Raman Malik: When I joined Perplexity, I made my pros and cons list, and the biggest question at that time was: can the application layer capture and monetize value, or is it just a massive pass-through cost? But now, looking back, I absolutely love working at the application layer — it's been fantastic. So that really changed my view. I think the application layer is an amazing place, and a significant portion of future value will be created here.

20VC: It's fascinating how the entire ecosystem has repositioned itself, shifted its mindset, and adjusted around where value flows. I remember doing a show where everyone was asking, how do you create value on top of these models? It's just a wrapper. Almost nobody was bullish on that back then.

Raman Malik: Yes, eight or nine months ago, that question was still a big open one. It wasn't clear where value would be created, and today, we're gradually seeing some people achieve real success at the application layer, especially in high TAM (Total Addressable Market) areas, making a lot of money.

20VC: What's Aravind's greatest strength?

Raman Malik: He can rapidly switch from very high-level company strategy to very specific operational details, like how the user funnel should be set up. He can move quickly between these two levels, and it all happens within the same meeting.

20VC: What's his biggest weakness?

Raman Malik: I think Perplexity is his first startup, so there's a lot of intuition he needs to build up quickly through practice. Which is great — he's accumulating intuition on growth, partnerships, and so on very rapidly. But he really is learning by doing, because this is his experience as a first-time founder.

20VC: Last question: in the past 12 months, outside of Perplexity, what growth strategy has impressed you most?

Raman Malik: I really love products or features that drive viral spread between users. I think something like Spotify Wrapped* is a classic example — the moment it came out, everyone wanted to share it. This strategy is now being adopted by many AI startups, whether in image generation or other areas, creating something highly shareable and unique that people want to spread around, so others discover you. Growing quickly by having your users promote for you is the most effective approach.

*Spotify Wrapped: Spotify's year-in-review feature that shows users their most-played songs, artists, and podcasts from the past year

20VC: Why doesn't Perplexity do a Perplexity Wrapped? Like stats on how many searches you made, trending topics you searched, who you shared with most — that could be pretty fun.

Raman Malik: Who says we aren't?

📮 Further Reading

Linear Bolt

Bolt is Linear Capital's dedicated investment program for early-stage, globally oriented AI applications. It upholds Linear's investment philosophy, focusing on projects where technology drives transformative change, and aims to help founders find the shortest path to their goals — whether in speed of execution or investment mechanics. Bolt's commitment is to be lighter, faster, and more flexible. In the first half of 2024, Bolt invested in seven AI application projects including Final Round, XinGuang, Cathoven, Xbuddy, and Midreal.