Raman Malik on Perplexity’s Growth: Retention, Partnerships and Experiments

Raman Malik, Perplexity's head of growth, joins Harry Stebbings to discuss the company's acquisition, retention and marketing experiments. His descriptions of Perplexity's performance and product reflect his role at the company and are not independently verified in the interview.

Malik says he entered an MBA program to experiment with ideas, left early and spent three years building a startup that eventually shut down. He values the program as an expensive but useful bridge into certain industries, while recognizing that the founder-to-employee transition required a narrower focus and an adjustment of ego.

He defines growth as a bridge between product and marketing. A growth-product team works with engineering, design and data on acquisition, activation, retention and monetization; growth marketing pursues the same user journey through channels, community, campaigns and lifecycle communication.

When to build a dedicated function depends on stage, company and founders. Malik suggests early evidence of retention can justify investment in the funnel before scaling paid channels. His example of 30% retention after three or four months is a heuristic, not a universal threshold or a claim about Perplexity's result.

At Perplexity he initially mapped what was happening across acquisition, engagement and monetization. Missing answers created more questions, so his early work included measurement infrastructure and attribution. He had access to engineers but no fully built growth team.

Small improvements can compound through a growth model: better activation or retention changes the level of the active user base. But incremental optimization eventually has diminishing returns, so Malik also wants a few high-risk, high-reward experiments each quarter. He says one major success a year could be valuable; his proposed success rate is a rule of thumb.

A well-scoped A/B test can identify incremental impact, unintended metric changes and what to work on next. A poorly scoped test may provide no usable answer. Malik warns against running tests merely to produce an impressive performance-review number.

Mixed results require judgment. A sign-in prompt after several queries might increase activated accounts while driving away other users and reducing search volume. The team must weigh possible long-term retention against immediate loss without necessarily waiting 60 days for every decision.

Malik calls early behaviors correlated with later retention milestone metrics. At Perplexity, three queries in a first session is one such signal: many users who later return did it. He cautions that the correlation is imperfect and does not prove forcing three queries will cause retention.

Metrics can be gamed. Recommended searches could raise query counts without providing real value, so he checks whether the correlation with later retention persists and looks at multiple measures, including time spent. Stebbings notes a short session can also mean a user found an answer quickly; Malik says longer exploration can be valuable in other cases.

Malik says most Perplexity growth at the time was organic word of mouth, estimating about 80%, with the mix varying by market. He attributes some discovery to public curiosity about AI and says partnerships also play a large distribution role. These are company estimates rather than audited channel attribution.

He says promotions with partners can encourage trial because another trusted product or person introduces Perplexity. Free access to paid features has an acquisition cost, and the business must eventually convert some users into paying subscribers. His comments about variable costs are a company perspective, not a full unit-economics calculation.

Influencer marketing was harder than expected. Managing many creators individually, aligning their content and explaining the product took significant time and results fell flat. He prefers developing deeper relationships with fewer creators and wants to reach student audiences through relevant partnerships.

Malik says the initial opportunity was less about acquiring traffic than improving early activation and week-one or week-two retention. He describes existing consumer retention as healthy, but believed lifting it would raise the number of active users over time.

Users can try Perplexity without signing in, making conversion from logged-out visitor to activated account a distinct challenge. Malik mentions a goal around 30% reaching milestone behaviors and signing in; he presents this as an ambition, not a reported achieved conversion rate.

For consumer products, he wants the retention curve to stabilize. He calls 45% retention at month six a very strong target and cites a Duolingo comparison. These are his benchmarks and recollections; definitions and cohorts differ, and the interview does not establish Perplexity's six-month retention.

To improve retention, he looks at early milestone behavior, channel mix, target audience and product quality. He believes organic and referral users often retain better than paid or partnership cohorts, though results depend on the actual users and context.

He says curiosity-driven visitors may be less likely to return than people using Perplexity to close a knowledge gap at work or find sources for study. That is his account of observed segments, not a quantified causal comparison.

Stebbings asks about a reported 10–15% improvement. Malik points to reaching stronger-retaining audiences, encouraging use across devices and improving the product. He does not specify the underlying metric or experiment in enough detail to attribute a precise causal gain.

One difficult area is explaining the product immediately on landing: users need to understand live information and sources without being overwhelmed by copy. Stebbings critiques a wordy description; Malik says the team is balancing search optimization, brand and functional clarity, and would favor simpler language.

Early on, Malik prioritized acquisition, activation and sticky retention over immediately optimizing customer acquisition cost against lifetime value. He expected the company to explore subscriptions or advertising later. He says CAC-to-LTV becomes more important as retention improvements diminish and channels are scaled; LTV itself needs enough cohort data.

At the time, he says Perplexity barely used paid acquisition. Paid campaigns can make top-of-funnel numbers look good while capturing users who would have arrived organically. He stresses measuring incrementality and retention, not assuming every attributed install is a new user.

He recalls a Lyft experiment in which the team turned off paid acquisition for new passengers and installs and signups fell only about 5–10%, by his recollection. He interprets the result as substantial cannibalization of organic demand. This is a retrospective account, not evidence that every paid channel or market is ineffective.

Lyft still used paid acquisition for drivers because a marketplace needs enough supply as well as demand. Malik says paid channels can also make sense for under-represented audiences if unit economics support them. At Perplexity he describes small TikTok tests aimed at 18–24-year-olds.

He says those tests helped the team learn to explain the user benefit—quick answers with sources—rather than simply advertise an 'AI search engine.' Top-of-funnel conversion looked solid in the early tests, but he was still waiting for retention data; he mentions spending around $10,000 a week for learning.

Malik believes repeated exposure can help people try a product, though he acknowledges brand marketing is hard to measure. For a sponsorship or television campaign, he looks beyond purchased impressions to additional sharing and attention. His three-to-seven-exposure rule is a belief, not an established law in the interview.

A television ad featuring football coach Jim Harbaugh reached a sports audience beyond Perplexity's usual users, he says, and sports commentators shared it. By contrast, newsletter sponsorships disappointed him: a brief placement made a complex product difficult to explain, and he felt the experiment produced few useful learnings.

His growth shorthand is to find an underused channel early or execute much better than competitors. He points to referral programs, email, search optimization and social media as historical examples. Stebbings notes that being first alone is not a guarantee; Malik stresses the need to find an edge in crowded channels.

He sees community as an underappreciated channel. For students already getting value from Perplexity, he proposes providing support and materials so they can share it on campus and build local density. This is a proposed distribution strategy, not proof that every campus program works.

Google and ChatGPT are the obvious competitors, he says. Perplexity remains largely organic and partnership-driven, and he wants better ways to work with creators than isolated sponsorship placements. Both speakers favor content that teaches or entertains over generic announcements.

He regrets the newsletter experiments and wishes the team had done more to improve sharing. Users already share answers or screenshots, but he does not consider sharing an inherent viral loop yet. He wants to make the experience easier without assuming it will automatically create growth.

The first growth hire depends on the team's gap: a marketer who can experiment in product, or a product person able to run campaigns. If acquisition is healthy but onboarding and cross-device behavior are complex, Malik favors someone experienced in building funnels.

He values former founders' willingness to take big swings and accept failure, comparing an early team to a jazz band improvising together. As a company grows, coordination becomes more important and some people may prefer the freedom of a smaller team. He recommends giving trusted hires room to experiment.

For hiring, he defines taste as recognizing what good looks like. He discusses actual challenges with candidates and listens to the depth of their questions. He also asks about difficult experiences and previous failures rather than relying only on polished success stories.

Perplexity uses scoped take-home exercises with hypothetical scenarios related to the company. Malik wants to see how candidates reason through a launch and collaborate in a discussion; Stebbings notes the concern that candidates have less internal context, a trade-off Malik acknowledges.

He wants new hires to investigate and identify opportunities independently rather than need a detailed 30-day checklist. Malik says he can be slow to hire because he cares about team fit and wants to become faster; Stebbings cautions that hiring too quickly is also a common mistake.

In quick fire, Malik identifies assuming user pain without understanding it as a costly growth mistake. He advises a new growth leader to learn the numbers deeply, understand how they relate and use that knowledge to find friction in the funnel.

From Lyft, he took a lesson to focus on retention and product rather than automatically spend on ads. He also thinks the team overestimated how much further ride-sharing product improvements could unlock growth, and should have recognized diminishing returns sooner. This is his retrospective assessment.

Malik says his view of the AI application layer changed: he initially questioned whether it could capture value above model costs, but now sees meaningful opportunity. As an employee of an application company, he has a direct stake in that thesis.

He praises Perplexity CEO Aravind Srinivas's ability to move between company strategy and detailed funnel questions, while saying the first-time founder is still building some intuition. These are his assessments of his own CEO.

He admires shareable product features such as Spotify Wrapped and thinks users sharing distinctive outputs can drive awareness. Stebbings proposes a Perplexity version; Malik hints at a possibility but gives no confirmed launch details. They close by thanking each other.