How do I define a 'cohort' for retention analysis?

Cohorts are groups of new users acquired during a specific period. The most common method is to group users by their acquisition week or month. You can also further segment cohorts by dimensions like country, device, or acquisition channel for deeper insights.

What makes a good 'active user' action for measuring retention?

A good action should be directly correlated with the user getting real value from your product, not just a superficial interaction. Examples include posting a photo on a social network or completing a ride on Uber, filtering out users who merely open the app without engaging.

What is the most important factor in a good cohort retention curve?

The most critical factor is whether the cohort retention curves become flat over time. While a higher flat line is better, the primary goal is to stop losing all new users, indicating that a core group finds enduring value in the product, allowing for user accumulation.

What are common mistakes founders make when analyzing cohort retention?

Founders often make mistakes by choosing too large a time period (e.g., quarterly instead of weekly), selecting an 'action' that's too easy to game (like simply opening the app), or focusing on a single retention data point instead of the overall curve's shape.

How can I improve my startup's cohort retention rates?

Improving retention involves enhancing your product's core value, speeding up performance, simplifying flows, acquiring users who are a better fit for your product, optimizing the onboarding experience, and potentially leveraging network effects.

Should I trust built-in retention graphs in analytics tools immediately?

Be cautious. Analytics tools may not always measure retention exactly as you intend. It's recommended to build your own understanding by analyzing your raw logs first, then compare your findings with the tool's output to ensure accuracy.

What does the 'layer cake' retention chart represent?

The layer cake chart visualizes the total active user base over time, with each colored layer representing a different user cohort. A healthy layer cake shows thick layers from older cohorts, indicating that your product is successfully retaining users long-term and growing its user base.

Key Moments

How To Keep Your Users | Startup School

Y Combinator

Science & Technology7 min read30 min video

Aug 29, 2024|52,857 views|1,714|72

YC Y Combinator

Save to Pod

Key Moments

TL;DR

A product is only successful if its user retention curves flatten out eventually, indicating sustained value, not just temporary engagement. If your curves don't flatten by 20% retention, you likely haven't built something people truly want.

Key Insights

Cohort retention tracks the fraction of new users who continue to use a product over time, grouping users by their acquisition date (e.g., weekly or monthly).

The key metric for a successful product is not the absolute retention rate, but whether the retention curves eventually flatten, ideally around 20% or more.

A common pitfall is choosing too large a time period (e.g., quarterly instead of monthly) or too shallow an action (e.g., app open vs. core feature usage) for measuring retention, leading to artificially inflated, non-representative curves.

Google Photos achieved over a billion users because its weekly retention curves flattened around 20-40%, demonstrating long-term user value, even though 80% of users initially left.

Improving cohort retention involves enhancing the product, acquiring more suitable users (e.g., targeting users with existing memories for photo apps), and optimizing the initial user experience.

The ultimate goal is a 'layer cake' graph where the top line (total active users) grows due to stable, thick layers from older cohorts, signifying sustainable growth.

Defining success: The power of cohort retention

The core challenge for any startup is to create something people genuinely want. While making something people want is Y Combinator's famous motto, knowing if you've achieved it can be elusive. Cohort retention offers a quantitative answer by tracking how a specific group of new users, or 'cohort,' continues to engage with a product over time, rather than looking at the entire user base mixed together. This method provides a clearer view of user loyalty and product stickiness. The speaker learned this lesson the hard way, realizing his pitch to a VC was weak because he didn't understand cohort retention himself until much later in his startup journey. His experience with 'bump,' which reached 100 million users but failed as a business, and subsequently contributing to Google Photos which serves over a billion users, highlights the critical difference between acquiring users and retaining them.

Three pillars of cohort analysis

To effectively measure cohort retention, three elements must be precisely defined. Firstly, **cohorts** themselves need clear grouping criteria. The most common approach is by acquisition time, such as new users acquired each week or month. Advanced segmentation can further slice cohorts by country, acquisition channel, device, or other user characteristics. Secondly, an **action** must be chosen that signifies an 'active' user. This should be a core feature usage that delivers genuine value, not just an app open or site visit. For example, Instagram might track viewing three or more posts, Uber might track completing a ride, and Google Photos used tapping to view a photo full-screen as its key active user metric. Finally, the **time period** for measuring subsequent usage needs to align with the product's intended usage frequency – daily for social apps, weekly for utilities like Google Photos or Uber, and potentially quarterly or even semi-annually for infrequent use products like travel apps (e.g., Airbnb).

Visualizing retention: From triangle to line graphs

A common way to visualize cohort retention is the 'triangle chart.' This chart lists cohorts by rows (e.g., January users, February users) and subsequent time periods (e.g., February, March, April) in columns. It shows how many users from an initial cohort returned in each following period. Crucially, each user is counted only once per period, regardless of their activity frequency within that period. The chart is then normalized by dividing by the initial cohort size to show percentage retention over time. This normalized data can be graphed as line charts, where each line represents a cohort's retention curve. This visualization allows for observing trends across cohorts (rows) and performance over time (columns), revealing patterns of user engagement or churn.

The 'flattening curve' is the ultimate goal

The most critical insight in cohort retention is that the shape of the curve, not its absolute height, is paramount. A product is considered successful when its cohort retention curves eventually flatten out. This flattening indicates that the product consistently delivers value over the long term, securing a stable base of engaged users. While a higher flattening point is better, the mere fact that it *does* flatten is more important. If curves don't flatten, users are constantly churning, creating a treadmill effect where new user acquisition merely replaces lost ones. For Google Photos, a 20-40% weekly retention rate that flattened out gave the team immense confidence, ultimately leading to over a billion users. This flattening provides the foundation for sustainable growth, allowing the company to accumulate users over time rather than perpetually chasing new ones.

Common pitfalls that deceive retention metrics

Founders often fall into traps that make their retention numbers look better than they are. One major mistake is choosing too large a measurement time period. Using quarterly or semi-annual periods instead of monthly will naturally make curves appear flatter and higher, which can be tempting for investors and self-assurance but is misleading. Another pitfall is selecting too easy an 'active user' action. For instance, counting a mere app open or a notification click that doesn't lead to value. This can be gamed, especially using external triggers. A common and counterintuitive mistake is relying solely on payment as the active user metric. Users often stop using a product long before they cancel their subscription. Therefore, pairing payment with actual product usage is essential. The best action is one that represents a user getting real value, akin to observing a customer truly using and benefiting from the product.

Avoiding the 'single data point' fallacy and tool dependence

Another frequent error is focusing on a single retention data point (e.g., '80% week-over-week retention') without understanding the context of the entire curve or its trend. A high retention percentage in an early week might look good, but if the curve rapidly declines thereafter, it's a sign of underlying problems. Founders often can't articulate the numerator and denominator for their claimed metrics. Furthermore, while analytics tools are helpful, they can be misleading if not fully understood. Some tools might not separate cohorts correctly or may measure rolling retention instead of discrete period retention. It's recommended that founders first build their cohort retention curves manually or using simple tools like spreadsheets to develop an intuitive understanding before relying solely on dashboards. Refreshing these graphs weekly or bi-weekly is advised to catch issues quickly.

Strategies to flatten and improve retention curves

Improving product-market fit is central to flattening retention curves. This can involve enhancing product features, introducing new use cases, improving speed, or simplifying workflows. Meaningful product improvements will be reflected in flatter, higher retention curves over time, with newer cohorts performing better than older ones. Equally important is acquiring the *right* users. Sometimes, a product is great but marketed to an audience that doesn't align with its core value proposition (e.g., targeting very young users with a product focused on reminiscing over life memories). Understanding user acquisition channels and targeting appropriately can significantly boost retention. Slicing cohorts by different dimensions (country, user type, acquisition channel) can reveal which segments perform well and where to focus improvement efforts. Optimizing the first user experience and onboarding process is also crucial; getting users into a valuable state quickly is often the cheapest and easiest way to improve retention.

The ultimate goal: The 'layer cake' of sustainable growth

The most desirable outcome is not just flattening curves, but curves that begin to increase over time, alongside consistently improving newer cohorts. This indicates users are not only retained but becoming more engaged. The ultimate representation of a thriving startup is the 'layer cake' graph. This visualization shows total active users over time, with each layer representing a distinct acquisition cohort. A healthy 'layer cake' shows a top line (total active users) growing robustly, composed of substantial, well-retained layers from older cohorts. This signifies a large, stable, and growing user base built on sustained value, a strong indicator of a company on a path to significant scale, potentially a multi-billion dollar valuation. While retention curves don't dictate *what* to change, they are a vital compass showing *whether* your efforts are working, guiding founders to iterate towards making something people truly want.

Mentioned in This Episode

●Software & Apps

●Companies

●Concepts

●People Referenced

Cohort Retention Cheat Sheet

Practical takeaways from this episode

Do This

Define cohorts based on acquisition time (week/month) or other relevant dimensions.

Choose an 'active user' action that reflects genuine value derived from the product.

Select a measurement time period that aligns with the intended product usage frequency.

Focus on whether cohort curves flatten over time, not just the absolute retention numbers.

Regularly refresh cohort retention graphs (weekly/bi-weekly) to catch issues early.

Improve product features, user experience, and target acquisition channels for better retention.

Consider network effects if applicable to your product.

Combine quantitative cohort data with qualitative user feedback.

Avoid This

Do not mix users from different acquisition periods into one group.

Avoid picking an 'active user' action that is too shallow or easily gamed (e.g., just opening the app).

Do not choose a measurement time period that is arbitrarily large to inflate retention numbers.

Do not rely solely on a single data point; analyze the shape and trend of the entire curve.

Be cautious of third-party analytics tools; build your own understanding first.

Do not use 'just paying' as the sole indicator of an active user.

Do not ignore user acquisition strategy; target the right customer type for your product.

Common Questions

Cohort retention tracks how a group of users acquired at a specific time continues to use a product over time. It's crucial because it reveals whether you've made something people truly want, indicating long-term product-market fit and sustainable growth potential.

Topics

Technology & Innovation Business & Entrepreneurship Product-market Fit Customer Acquisition Growth Strategy User Engagement Startup Metrics User Retention

Mentioned in this video

Software & Apps

Google Photos

A photo management app developed by Google, which formed the basis of the app David Lee worked on after his acquisition. It serves over a billion users.

Bump

A mobile app co-founded by David Lee that allowed users to share contact information and photos by bumping phones. It reached over 100 million users but was not a successful business on its own.

Companies

Netflix

A streaming service mentioned as an example of a product where users might continue paying even after ceasing active usage, cautioning against using 'paying' as the sole measure of action.

YouTube

A video-sharing platform mentioned as an example of an entertainment app where daily usage is intended, suggesting daily cohort measurement.

Google

The tech giant that acquired David Lee's startup Bump and where he later worked on Google Photos.

TikTok

A social media platform mentioned as an example of an entertainment app where daily usage is intended, suggesting daily cohort measurement.

Instagram

A social media platform used as an example for defining an 'active user' action, suggesting 'viewed three or more posts'.

Airbnb

A travel booking platform mentioned as an example where less frequent usage might warrant quarterly or semiannual cohort measurement.

Uber

A ride-sharing company used as an example for defining an 'active user' action, suggesting 'completed a ride'.

Concepts

Gen Z

A demographic group targeted by a marketing campaign for Google Photos, which resulted in poor retention because the product's value proposition (reminiscing on memories) didn't align with the life stage of young users.

People

David Lee

The speaker and author, a group partner at Y Combinator, who previously co-founded the app Bump and worked on Google Photos.