Finding Hidden Growth Opportunities in Your Product | Albert Cheng (Duolingo, Grammarly, Chess.com)

This video features a conversation with Albert Cheng, who has led growth at world-class consumer subscription services including Duolingo, Grammarly, and Chess.com. He introduces his "explore-exploit" framework for discovering and scaling new growth opportunities in products, along with a rapid experimentation methodology based on deep understanding of user psychology. The video covers a wide range of topics including Grammarly's monetization success story, ideal retention rates for subscription apps, the importance of resurrected users, the reverse free trial model, and the three key elements of successful gamification. Albert Cheng also shares his insights on how AI is impacting growth roles and team building, with particular emphasis on the power of brand and community, and key lessons from targeting 1,000 experiments per year.

1. Albert Cheng's Growth Leadership Journey: From Pianist to Global Growth Expert

The video begins by introducing Albert Cheng's experience leading growth at three of the world's most successful consumer subscription services: Duolingo, Grammarly, and Chess.com. He previously worked at YouTube developing streaming and gaming features for over 20 million users and has a unique growth approach that combines marketing, data, strategy, and product management. This conversation covers his "explore-exploit" growth framework, the biggest growth wins at Duolingo and Grammarly, how to accelerate growth using AI, the power of brand and community, and know-how for achieving the goal of 1,000 experiments per year.

Lenny asks Albert about his background in transitioning from pianist to one of the world's top growth experts. Albert explains that growing up as a child of immigrant parents, he learned piano and practiced consistently for 90 minutes every day. With perfect pitch, he could quickly pick up music. He once considered attending music school but chose to major in engineering. He says he only recently discovered the commonalities between music and growth.

"Both music and growth rely on consistent repetition. You have to get used to constantly making mistakes and learning from them through very fast feedback loops. Also, both have a structural foundation. Growth has growth models, metrics, experiments, channels, and so on, but every day you need creativity -- coming up with interesting solutions and hypotheses. Music is the same. There's music theory, scales, and other structural elements, but to create beautiful music, you need passion, emotion, and immersion."

2. The "Explore-Exploit" Framework for Discovering Growth Opportunities

Albert emphasizes that the "Explore-Exploit" framework is very effective for growth. He first encountered this concept through Nurmal, an engineering partner at Grammarly, and speculates that Brian Balfour introduced the concept in a Reforge class.

Explore mode: Like searching for the right mountain to climb. A phase of seeking new ideas and testing various hypotheses.
Exploit mode: A phase of concentrating resources to effectively climb the chosen mountain. It focuses on maximizing results based on insights gained through exploration.

Albert warns against leaning too far in either direction. Too much exploration can cause the team to lose direction and try random ideas, while too much exploitation can lead to growth plateaus and saturation. He prefers to apply this framework not only at the macro (big-picture) level but also at the micro level -- at the "insight" level.

2.1. Chess.com's "Explore-Exploit" Case Study: Improving the Game Review Feature

One of the important tasks Albert took on at Chess.com was encouraging users to learn chess and improve their skills. A PM named Dylan was responsible for learning features, and the most-used learning feature in the product was "Game Review."

Game Review is a feature where, after completing a chess game, a virtual coach tells you about your worst moves, best moves, and so on. Dylan's mission was to increase engagement and retention for this feature.
Explore phase: Dylan analyzed how users were utilizing Game Review and discovered an unexpected insight: "80% of people who use Game Review do so after a win." This was the exact opposite of what the team assumed when building the feature -- that people would use Game Review after losses to learn from their mistakes.
Exploit phase: Based on this insight, changes were made to the product experience.
- Instead of pointing out mistakes on the post-loss Game Review screen, the team shifted to first showing positive elements like "brilliant moves" and "best moves."
- The coach was designed to deliver encouraging messages like "Losing is part of learning. Keep going."
Results: These changes alone increased Game Review engagement by 25%, subscriptions by 20%, and significantly improved user retention.

Albert did not stop there -- he shared this insight across the entire company so that other product managers could find and apply positive patterns in their own products. For example, the puzzle team's PM began thinking about ways to provide positive experiences by changing puzzle success rates, modifying wording, or changing button colors. Albert explains that scaling the learnings from successful experiments across the entire company is the exploitation stage.

Albert notes that in his experience, experiment success rates are only about 30-50%, and many hypotheses can be wrong since consumer products are hard to predict. However, he emphasizes that failed experiments can also provide extremely valuable learnings. The key is for PMs to clearly communicate their hypotheses and discoveries so that other team members can rally around them, try diverse ideas, and increase success rates and impact.

3. Accelerating Growth Experiments with AI: Data Analysis and Prototyping

Albert emphasizes that AI greatly helps accelerate growth experiments and introduces several interesting cases from Chess.com.

3.1. Text-to-SQL Feature: Improving Data Accessibility

Albert has been using the text-to-SQL feature recently and explains that it is very powerful.

Chess.com has a data request Slack channel where various one-off questions come in, such as "How many subscribers do we have in South Africa?" or "How much time was spent playing puzzles last month?"
Previously, answering these ad hoc questions required a data analyst to spend time running queries. But with AI, this process can be greatly shortened.
Albert's team is training a Slack bot to provide first-pass answers to these questions. This enables the entire company to make much more data-driven decisions.
An interesting point is that the volume of questions itself increases significantly because people start asking the AI bot questions they were too shy to ask or didn't want to bother someone with. Albert adds that this is a phenomenon also seen with ChatGPT.

3.2. AI Prototyping Tools: Accelerating Idea Visualization

PMs are using various prototyping tools to flesh out their ideas.

Going from idea conception to final solution implementation requires a lot of manpower. But with AI, this process can be shortened.
Albert's team built AI prototypes of key product screens -- onboarding flows, home screens, chess boards -- using tools like V0 and Lovable.
With these foundational elements, other team members can add ideas and visualize them, making it much easier to discuss and quickly test ideas.

3.3. AI Stack and Workflow

The AI tools used by Albert's team are as follows:

PM: Primarily uses V0.
Designers: Prefer Figma Make.
Engineers: Use a combination of tools including Cursor, Claude Code, and GitHub Copilot.
Marketing: Uses AI tools for various purposes including translation, subtitles, and content adaptation.
Customer Support: Uses Intercom Fin.

Albert notes that while there are many of these tools, "they don't yet smoothly transition from experimentation to workflow." The tools used by different functions lack interoperability, so manual handoffs are still needed to ship ideas to production. However, he adds that this will improve over time, and they are investing in design system components and MCP to make the process smoother.

4. Grammarly's Monetization Success Story: The Premium Feature Preview Strategy

Grammarly CPO Noam Lovinsky asked Albert about Grammarly's biggest monetization success.

About Grammarly: An AI-powered writing assistant used as a Chrome extension or desktop client that overlays various suggestions on users' writing. It uses a freemium business model, with over 90% of users on the free service and the remainder as paid subscribers.
Problem identification: At the time Albert worked there, Grammarly's free users perceived it only as a spelling and grammar correction tool. This was because that was the main feature provided in the free version. However, Grammarly's paid version offers much more sophisticated features like tone improvement, clarity enhancement, and full sentence rewrites.
Data analysis: Previously, they were not properly tracking how often users encountered premium suggestions or how often they saw the payment screen. After building a system to track events, they discovered that the proportion of free users who "accepted all suggestions" was very low. Most users were selectively accepting only the suggestions they needed.
Strategy shift: Albert's team focused on this point and decided to completely change strategy.

"What if we extracted some premium suggestions and interspersed them throughout free users' writing, giving them a limited taste of what the paid service offers?"
Concerns and results: Initially there were concerns that giving away too many premium features for free would prevent people from subscribing. But the result was the exact opposite.

"Suddenly people began perceiving Grammarly as a much more powerful tool than before, and upgrade rates nearly doubled."

Albert advises from this experience that, especially for premium products, the free product should strive to reflect everything the paid product offers. While some premium features do incur costs, in the long run, showcasing the best features can yield greater results.

4.1. Freemium vs. Trial Models for Subscription Products

Albert explains why the freemium subscription model is effective.

Mission orientation: Many companies like Duolingo, Grammarly, and Chess.com have missions around education or information dissemination and want to make their products as widely available as possible. The lowest-friction way to achieve this is to provide a free product.
Word-of-mouth and network effects: Many products grow through word-of-mouth, and can build in-product network effects like Duolingo's social features or Grammarly's B2C-to-B expansion. Even free users can provide value by recommending the product to other colleagues or teams within a company.
Core value proposition: Albert advises that the core value proposition should be offered for free, while premium features should be provided in a preview format.

Regarding trial models, he explains:

Reverse Trial: For B2B features with lock-in effects, reverse trials can be very powerful. Users use the product without entering credit card information, then are prompted to convert to paid after a certain period. This is effective for products where users invest significant time, like CRM or content creation tools.
Consumer products: However, for many consumer products, reverse trials are difficult to implement. Albert says that standard free trials are generally more common. In Grammarly's case, they used a "limited trial" approach that provided a certain number of suggestions before prompting payment -- similar to a real-time reverse trial based on usage rather than time.

4.2. Retention Rates for Successful Consumer Subscription Products

What is the most important factor in successfully building a consumer subscription product?

"User retention is gold for consumer subscription companies. If you can't retain users, the burden of getting users to pay from day one becomes much greater. That is very difficult."

Albert points out that indie developers often think building an app and adding payment features is easy, but in reality, distribution, customer support, and growth are much harder.

New user retention: Albert says that a Day 1 (D1) retention rate of 30-40% is a decent level. Below that, it becomes difficult to build a sufficient daily active user (DAU) base.
Existing user retention: Even more important is existing user retention. Especially for products with high daily usage frequency like Duolingo or Chess.com, what matters most is how "sticky" established users with habitual patterns remain with the product. This retention drives long-term habit formation and product growth.
Grammarly's case: Since Grammarly is not a product users actively open every day, rather than existing user retention, initial activation, installation, and the "aha" moment are extremely important. This initial experience plays the role of keeping users engaged long-term.

5. The Importance of Resurrected Users

Albert emphasizes the importance of "resurrected users."

In Chess.com's case, about 80% of daily or weekly active users are existing users. The scale of new users and reactivated/resurrected users is nearly equal. In other words, as a product matures, bringing back people who used it in the past becomes just as important as acquiring new users.
Improving the reactivation experience: To bring back the hundreds of millions of dormant or occasional users accumulated in the product, investment in the product experience for "resurrected users" is necessary.
Duolingo's example: Duolingo effectively utilized social notifications. For instance, when a friend starts using Duolingo, a push notification is sent to bring the user back. For returning users whose previously learned language skills have declined, features like retaking a proficiency test guide them to the appropriate learning level. Such strategies can deliver high ROI for mature companies.

"Essentially, so many of your people have already tried in the past that to grow you need to resurrect people that have been there and so thinking through it's almost like a user experience for resurrected users."

6. Comparative Analysis of Growth Strategies: Duolingo, Grammarly, Chess.com

Albert explains that the three companies succeed in different ways and compares each company's unique growth strategy.

6.1. Duolingo: The "Green Machine" and Experimentation Culture

Characteristics: Duolingo has a very unique and consistent product development approach. They even created a playbook called the "Green Machine" that is deeply embedded throughout the company.
Talent profile: They hire many smart, energetic recent college graduates, provide them with amazing experimentation tools, and place great importance on the company's "clock speed" (pace of work).
Product experience: Duolingo's product experience changes multiple times a day for each user. This represents an incredible level of experimentation speed.
Strengths: Rigorous processes and clear specifications manage each stage of the product development cycle very efficiently.
Core driver: While providing a language learning product, the fundamental spirit lies in "motivation" and "habit building." Language learning is the first vehicle for exercising these motivation and habit formation capabilities.
Success factors: The structure is rigorous, but the ideas are very creative. Users are attracted through fun and creative elements like Super Bowl ads, memes, and gamification strategies.

6.2. Grammarly: Expanding from B2C to B2B, Product-Led Sales

Early days: Started as a paid product for students.
Expansion: Transitioned to a freemium model to reach more people, gradually focusing on professionals.
B2B transition: As many professionals used Grammarly, patterns of large-scale adoption were discovered in marketing teams, sales teams, and customer support teams. This led to adding a managed enterprise sales strategy.
Albert's role: Focused on growing consumer self-service revenue and active users while also working on demand generation and helping sales teams identify which teams, functions, and companies to approach. This was similar to product-led sales.
AI transformation: Was at the center of the generative AI transition, and recently acquired KOD and Superhuman, evolving into a productivity suite with rapid changes underway.
Differences in growth work: Unlike Duolingo, at Grammarly the core product experience itself drives repetitive activity. That is, the frequency and quality of suggestions provided daily determine existing user retention. Albert built a growth team but realized the core product team had a bigger impact on this metric and transferred the work to them.

6.3. Chess.com: Passionate Community and Focus on the Essence of "Chess"

Characteristics: Chess.com's biggest feature is that employees are extremely passionate about chess.
Talent profile: They hire chess lovers from around the world who play chess all day, watch streams, and fill Slack with chess-related conversations.
Core value: Unlike Duolingo with motivation and habit building, or Grammarly with application integration and AI technology, Chess.com is 100% about "chess" itself. The entire company overflows with the energy of people directly using the product and generating ideas.
Growth driver: Over the past five years, chess popularity has surged due to the pandemic, The Queen's Gambit, and YouTube/Twitch streamers. This demonstrates the importance of not just incremental improvements through growth experiments, but riding major external waves. Such waves can provide opportunities to quadruple registration rates overnight.

7. How AI Impacts Chess.com and Growth Roles

Albert explains the impact of AI on Chess.com and growth roles from two perspectives.

7.1. AI's Impact on the Chess.com Product

Chess and AI history: Chess and AI have been intertwined for nearly a century. IBM's Deep Blue defeating world champion Gary Kasparov in 1997 was a major shock regarding whether AI could surpass humans, but 30 years later, people still enjoy playing chess.
The role of chess engines: Chess.com has learned to augment the human chess experience with the power of chess engines like Stockfish (a powerful form of AI, distinct from LLMs). These engines are far more powerful than the world's best grandmasters (e.g., Grandmaster ELO 2800 vs. Stockfish ELO 3600).
"Game Review" feature: Chess.com makes this technology available to all users. Game Review runs a chess engine in the background for every move, provides evaluations, explains them in the user's language in an accessible and friendly way, and even supports audio. This is where LLMs (Large Language Models) are used to add "personality" to the user experience and deliver feedback.
Core principle: Albert emphasizes the importance of "using the customer as your compass" when applying technology. Rather than applying LLMs just because they are trending, the right technology should be applied to the right feature to deliver value to users.
LLM limitations: Interestingly, LLMs themselves are not good at playing chess. They hallucinate moves and excel more at pattern recognition than deeply analyzing specific chess problems. When creating chessboard images in ChatGPT, the number of squares is often wrong or the board is not properly configured.

7.2. AI's Impact on Growth Professionals' Work

Albert defines growth as "connecting users with the value of the product" and explains that AI can accelerate this work.

Accelerating the experiment cycle: AI can speed up various elements of the experiment cycle.
- Product Discovery: Unlike core product teams that develop products from a long-term perspective through user and market research, growth teams run many experiments in short cycles, with each experiment's results becoming inputs for the next idea.
- Writing analysis documents and generating ideas: Previously, writing experiment analysis documents manually, reading them to extract insights, and drafting new specifications took significant time. Now, tools like ChatGPT can summarize analysis documents and provide advice on new ideas, making the ideation and research cycle much faster.
- Accelerating prototyping: Thanks to AI prototyping tools, visualizing ideas has become much faster. While PMs are not deploying code directly, the time to flesh out ideas into forms team members can click through has been greatly reduced. Albert explains that this makes the "explore" phase much easier.

8. Tips and Best Practices for Running Large-Scale Experiments Successfully

Albert offers several pieces of advice for those looking to improve their team's experimentation capabilities.

8.1. Taking the First Step: "Start Somewhere"

According to an Atlassian report, 40% of product teams do not experiment at all.
Albert emphasizes that if you have a consumer product with some scale and usage frequency, you can collect sufficient data and should start experimenting.
He confesses that despite identifying patterns from working at many companies, his predictions are still often wrong. Consumer behavior is hard to predict, and after using a product for a long time, you tend to forget the general user's experience -- so don't miss opportunities by not experimenting.
He encourages starting quickly, like "crawl, walk, run" -- running A/B tests, leveraging third-party tools, or collaborating with engineers.
Preferred tools: Grammarly used Stat Sig; Duolingo and Chess.com use their own experimentation tools. For companies that experiment as frequently as Duolingo, building your own tools helps significantly, but he generally does not recommend building custom tools from the start.

8.2. Systems Matter More Than Individual Experiments

Growth model: It is important to first build a growth model that clearly understands how the company grows and which channels to use.
Product instrumentation: The product must be properly instrumented. Otherwise, experiment results can be distorted. Albert shares an embarrassing past experience where user retention was set up backwards -- positive results were actually negative -- and says "that will never happen again."

8.3. Duolingo's Virality Growth Case: Focusing on User Behavior

At Duolingo, Albert's team formed a virality team. While virality has many unpredictable elements, Duolingo is a product where sharing happens frequently.
Screenshot tracking: The team introduced screenshot tracking for a short period to identify where screenshots occur most frequently within the app.
Insights: Through this, they discovered moments users wanted to screenshot and share, such as "streak achievements," "fun challenges," and "reaching the top of the leaderboard."
Design enhancement: Illustrators and animators were deployed to create delightful experiences at these moments.
Results: By focusing on and enhancing the moments users naturally took screenshots, they were able to drive 5x or 10x growth. This shows that the strategy of leaning into where users already share organically, rather than forcing sharing against people's intuitions, is far more effective. This connects to the "explore-exploit" methodology.

9. The Three Pillars of Gamification: Motivation and Habit Formation

Albert emphasizes that both Duolingo and Chess.com are skilled at habit formation and motivation, and shares the three pillars of gamification model proposed by Jorge.

9.1. The Three Pillars of Gamification

Core Loop: The core cycle of actions users perform repeatedly.
- Example (Duolingo): Complete a lesson -> Earn rewards -> Extend streak -> Next-day push notification. Making this core loop very solid is essential.
Metagame: Elements that provide long-term goals and motivation.
- Example (Duolingo): Learning paths, leaderboards, achievements, challenges, and other elements that users can work toward over the long term.
Profile: A personalized space reflecting the time and effort users have invested in the product.
- Example: A profile built over time becomes a mirror showing the user's investment in the product.

Combining these three elements well can create a successful long-term learning journey.

9.2. Chess.com's Beginner Onboarding Experience

In the case of Chess.com, over 75% of new users identify themselves as chess beginners or newcomers.

Problem: Beginners find it hard to enjoy live games. Data shows that less than one-third of beginner users win their first game. Also, losing a game results in retention that is 10% lower than when winning. At scale, this 10% difference has a significant impact.
Common mobile game strategy: Many mobile games create simplified versions of the game for beginners, but in chess, changing the rules is difficult.
Chess.com's approach: Albert notes that when learning something new (a language, chess, etc.), it is easy to feel doubt and inadequacy at first, so it is important to create an experience that intentionally guides users through these difficulties.
Ideas being tested:
- For users who identify as chess beginners, instead of live games, offer a "more enjoyable chess learning experience."
- For the first 5 games, hide the rating so users don't see their rating plummeting.
- Provide various alternatives such as playing against a coach, playing with friends, or playing against bots.

10. A Counterintuitive Lesson on Team Building: "High Agency" Over Experience

Albert shares one of the most counterintuitive lessons in team building.

Traditional hiring approach: Companies typically try to hire people with experience from companies similar to theirs, based on traits specified in job descriptions (JDs).
Albert's discovery: Working at smaller startups like Duolingo, Albert realized that the top performers were not necessarily those with deep domain experience. Rather, people with "high agency," "fast clock speed (pace of work)," and "energy" excelled.
The double-edged nature of experience: Specific experience can even become a "crutch" that holds people back, he points out. Especially in an era of rapid change driven by AI, he emphasizes the need to intentionally shed learned habits from the past and adopt a "beginner's mind."
The future talent profile: Albert believes in investing in people who react quickly, move quickly, and learn quickly. He thinks these are the companies that will ultimately survive and thrive.
Characteristics of "high agency": Albert also explains how to identify high-agency people during the interview process.
- It often shows through aspects outside the formal interview process (e.g., the questions they ask, their experience deeply using the product).
- It can be sensed in how they prepare for the interview and the energy they bring to the conversation.
- He used to focus only on questions and evaluation criteria, but now he also pays attention to these "soft signals."

11. Choosing Company Size: Finding Your "Goldilocks Zone"

Based on his experience working at companies of various sizes, Albert explains that each person has a company size that suits them best. He himself found his own "Goldilocks zone" by moving from a large corporation to an ultra-small startup and then to a mid-size company.

Albert's "Goldilocks zone": He says he gets the most energy from mid-size companies where he can oversee the company's overall efforts while also diving deep into details -- collaborating with specific teams, analyzing experiment results, and checking things down to the pixel level. He enjoys a pace where work can be handled daily or weekly.
- The mid-size companies he describes are around 500 to 1,000 employees, typically 10 to 20 years old, stable, and profitable. These companies are still at important inflection points with many problems left to solve and provide dynamic environments.
Large corporations (e.g., Google):
- Pros: Experience dealing with enormous scale, learning best practices from colleagues, access to every kind of tool and capability.
- Cons: Felt frustration with the slow pace and difficulty of shipping products.
Ultra-small startups:
- Pros: Can move incredibly fast.
- Cons: Hardly anyone knows the company, so recruiting people one by one and acquiring users is extremely grueling -- he expresses that this "gave him gray hair." Even if you want to change the world, doing so from a small startup is realistically very difficult.

12. Failure Case: Chariot's "Solution-First" Product Development

Albert says he won't mention specific failures in growth since "you always fail in growth," and instead shares a failure from his core product development days.

Background: Albert served as product lead at Chariot, a shuttle service startup in San Francisco. Chariot was a 15-passenger commuter shuttle positioned between public transit and Uber. Users loved the reliable, fast, and affordable core service.
The failed idea: Introducing "Dynamic Routes": The Chariot team tried a dynamic routing system called "Chariot Direct" to increase shuttle utilization and innovate the service. Instead of running fixed routes, the idea was that drivers could deviate from their route to pick up passengers at their homes during downtime.
Three lessons: This attempt ultimately failed, but it taught Albert three important lessons:
1. A solution without a problem: It was solution-first thinking -- "Wouldn't it be cool if we did this?" rather than problem-first thinking -- "Who are our users, what problem are we solving, and why will this solution delight them?"
2. Insufficient stakeholder consideration: In marketplace-type businesses, you must consider multiple end users. The team focused too much on the passenger app and failed to adequately consider the impact on drivers and operations teams. When drivers became confused or dissatisfied, the entire product experience suffered.
3. Premature PR: A lot of press coverage was pursued before the service even launched. PR has its place and time, but promoting before validating customer demand is very risky. This creates significant sunk costs and pressure to succeed simply because the effort has begun.

"This was 10 years ago, but while building many products since then, I have continued to remember these three core lessons."

Albert adds that this failure experience led him to take the exact opposite approach afterward -- "experiment with everything before telling anyone."

Conclusion

The conversation with Albert Cheng provided deep insights and practical strategies for driving growth in consumer subscription products. His explanation of the explore-exploit framework for discovering new opportunities and scaling existing successes, Grammarly's counterintuitive monetization strategy, and the importance of resurrected users prompt a rethinking of conventional perspectives on product growth. Furthermore, the diverse success models -- Duolingo's rigorous yet creative experiment culture, Grammarly's product-led sales, and Chess.com's passionate community -- demonstrate that there is no single right answer. The way AI is transforming the growth professional's work by accelerating data analysis and prototyping offered forward-looking implications, and the team-building philosophy of "preferring high agency over experience" emphasized the capabilities needed in a rapidly changing era.

Ultimately, Albert argues that the essence of growth is "connecting users with the value of the product," and that a mindset of constantly experimenting, learning, and approaching things flexibly is what matters most. His approach of learning from failure, building systems, and keeping the customer at the center will serve as great inspiration for anyone thinking about product growth at any scale of company.