As AI continues to reshape the way people interact with digital content, tech companies are racing to deploy machine learning at scale. Few product managers, however, have the experience to turn these innovations into features that reach billions of users while still feeling personal. Nisarg Shah, a product manager in AI/ML at Meta, is one of them.
From launching co-watching experiences at Amazon to surfacing real-time cultural trends on Instagram, Nisarg has built products that combine deep analytics, human insight, and technical rigor. In this interview, he shares how he spots ideas with “cultural potential,” validates them with data, and leads teams of engineers and data scientists to bring them to life.
Nisarg, you’re known for launching products from scratch that resonate with massive audiences. How do you identify when an idea has real “cultural potential” and the ability to scale to millions of users?
Cultural potential lies at the intersection of deep user insight, societal trends, and technological feasibility. I start by analyzing behavioral data — patterns in how users engage with existing products, what they share, and where they spend their time. At Amazon, we knew people watched content with their loved ones as a way to bond with each other, which led to creation of Watch Party. At Meta, I look for signals in social interactions — such as how users remix or amplify content on Instagram. An idea can scale when it taps into a universal human need, like connection or self-expression, and can be powered by AI to personalize at scale. For example, Instagram Trends was born from observing how users were organically curating viral moments. We used machine learning to surface these in real time, making the platform a cultural pulse. The key is to validate the idea with data, test it with a small cohort, and ensure the technology can handle exponential growth without losing its soul.
Tell us how the idea for Prime Video Watch Party came about. What patterns in user behavior or shifts in cultural context inspired this feature?
The idea for Prime Video Watch Party came from a mix of user behavior and cultural shifts. In 2020, we saw a surge in virtual socializing as people craved connection during lockdowns. Effectively, watching content together was one of the most common ways people bonded with their loved ones. Data showed that users were screen-sharing movies on third-party platforms or coordinating watch times via group chats. This was a clear signal: people wanted shared viewing experiences, but the tools were clunky. We also saw spikes in engagement with interactive, community-driven content on social platforms. Based on these insights, we hypothesized that a native, real-time co-watching feature could bridge the gap.
AI helped us model user preferences to suggest watchable content for groups. At the same time, cultural trends toward remote bonding gave us confidence in its potential. I believe Watch Party was more than a feature, it became a response to a societal need for connection, amplified by technology.
What were the biggest challenges in launching Watch Party? How did you balance ease of use, real-time content synchronization, and rights management?
Launching Watch Party meant solving three core challenges: real-time synchronization, user experience, and rights management. Real-time sync was the toughest. We needed to keep everyone watching together across devices under varying network conditions. In practice, we relied on Amazon’s expertise in low-latency streaming and AI-driven buffering algorithms to keep everyone in sync. To keep it easy to use, we iterated on the interface, used A/B tests and ML-driven feedback loops to simplify joining and hosting. Rights management was trickier.
We built the system in a way that honored the existing content rights while still providing a superior co-watching experience. That required building new technology to keep everyone in sync on their own player. Balancing these areas meant prioritizing user delight, making the experience feel seamless while we handled the complexity behind the scenes. Data showed that most users found it intuitive and a key way to stay connected with their loved ones, which validated our approach.
Fast-forward to your current work at Meta: how did the concept for Instagram Trends come to life? Was the primary goal engagement, freshness, or structuring viral content?
Instagram Trends started when we saw users organically curating viral moments like hashtag challenges and meme waves. Our data showed that these moments actually drove disproportionate engagement but were often buried in the noise. The goal was a mix of freshness and engagement: we wanted to surface real-time cultural pulses while keeping users hooked. Using advanced ML models, we analyzed content velocity: how fast a post, sound, or hashtag spreads.
Then we paired it with sentiment analysis to prioritize authentic trends over spam. The challenge was to structure virality without losing its organic feel. We tested prototypes with creators to make sure Trends felt like a natural extension of the platform, not a forced algorithm. This gave everyone on Instagram the opportunity to reach millions of people, regardless of their account size, which was previously impossible. It also helped creators and brands reach their audiences more authentically and be part of cultural moments relevant to people on the platform.
What’s your framework for building 0→1 products? Do you start with a user pain point, a bold idea, or a technological insight?
My 0→1 framework starts with the user and is powered by data and technology. First, I identify a clear pain point through qualitative feedback and quantitative signals: for example, users struggling to curate their feed on e-commerce or social media sites.
Then I explore bold ideas that address that pain, using AI to simulate outcomes and predict adoption. For example, with Watch Party, we started from the pain of disconnected viewing. We used ML to model group dynamics. Next, I validate with rapid prototypes. I rely on A/B tests and user cohorts to refine.
Technological insights, such as advances in real-time synchronization or recommendation algorithms, often shape the approach. The key is to stay hypothesis-driven: test quickly, learn cheaply, and scale what works. I believe this mix of empathy, data, and technology ensures products solve real problems while pushing boundaries.
In your view, what’s the difference between simple innovation and a product that becomes part of cultural behavior?
Simple innovation solves a problem, but cultural products redefine how people live. In my view, the difference is the emotional resonance and scalability. Simple innovations, such as a better search bar or algorithm, improve efficiency but do not change behavior. Cultural products, including Instagram Trends and Watch Party, tap into universal desires like self-expression and connection, and become habits. They are powered by AI to personalize at scale and still feel intimate.
For example, Trends did not just add virality. It changed how people connect with one another within a shared context, with machine learning curating the experience in real time. Cultural products also require ecosystem alignment across partnerships, creator buy-in, and technology that scales globally. Data shows that cultural products drive 10x engagement over incremental ones. They do more than solve problems. They inspire.
Working on high-impact AI/ML projects at scale can be intense—how do you stay curious and avoid burnout while tackling such complex challenges?
Staying curious starts with focusing on the user impact. I believe every model we train or feature we ship could redefine how millions connect or create. I keep curiosity alive by exploring cross-disciplinary fields, such as behavioral psychology and emerging technology, to spark new ideas. To prevent burnout, I use structured prioritization.
I use data-driven frameworks to focus on high-impact problems, and I avoid overoptimizing low-value tasks. I also block time for deep thinking: uninterrupted sessions to wrestle with complex problems, often using AI tools to simulate scenarios and reduce cognitive load. Finally, I stay grounded by mentoring junior PMs. Their fresh perspectives actually recharge me. In the long run, balancing intensity with purpose keeps me sharp and sustainable.
