By Andrew Chen , a16z

Compiled by Tim, PANews

I've been looking at retention curve data for over 15 years.

I've seen thousands of retention curves, and it's one of the first metrics I ask for when evaluating startups. I've pored over thousands of repositories and analyzed retention curves broken down by various metrics. As a product builder, I've also seen this metric from another perspective. I've run hundreds of A/B tests and crafted countless versions of user guides and email notifications, trying to change the shape of retention curves.

A/B testing (also known as split testing or bucket testing) is a randomized experimental method used to compare two versions of a product (version A and version B). Its core purpose is to determine which version performs better in achieving the predetermined goals by collecting data and analyzing user behavior.

Judging from the results, there are some patterns here.

Just like the laws of physics, there are strangely certain patterns that keep emerging over time. Here are a few examples I’d like to share:

• You can’t improve poor retention. Adding more notifications won’t improve your retention curve. You can’t achieve good retention through A/B testing.
• Retention rates only decrease, never increase. And strangely enough, their rate of decay does follow a predictable half-life pattern. Early retention rates can predict later retention performance.
• Revenue retention expands, while usage retention shrinks. The good news is: while users may churn, those who stay sometimes spend more!
• Retention is highly dependent on your product category. It's a combination of nature and nurture. Unfortunately, you can't make your hotel booking app a daily user.
• As your user base expands and grows, retention becomes lower. The best quality users come from early and organic growth, while later acquired users perform the worst.
• User churn is asymmetric; it’s much easier to lose a user than to win them back.
• Measuring retention rate is extremely difficult. Seasonality does exist, newly released beta versions can skew the data, and system bugs can occasionally occur. While D365 is a true indicator, it shouldn't be the sole indicator.
• We’ve seen this repeatedly across multiple platforms and categories: viral growth with poor retention leads to ultimate failure.
• Excellent user retention is a miracle. When you actually see it, it's astonishing.

We will analyze these points one by one.

You can't fix poor user retention. You've seen it firsthand: You spend months developing a new product, then launch it. The initial retention numbers are abysmal. With the product months in development and the situation stagnant, how can you improve retention? Then, a flash of inspiration strikes: How about adding push notifications to remind users to come back? Or adding a bunch of new features? What about A/B testing your landing page to improve conversion rates?

I think we all know how this ends. Unfortunately, when a product's retention is poor, it's often extremely difficult to reverse, almost to the point of being completely wiped out. Sure, marginal improvements are possible. Let's say your day-two retention is 40%, and your goal is to get it to 50%. That's perfectly doable and worthwhile. But if your day-two retention is only 10%, it likely means you've built a product with fundamental market fit, and all the local optimizations like A/B testing and push notifications won't be enough to change the fundamental situation. With months of development time and sunk costs already in the balance, it's hard not to engage in desperate struggles. But I think in most cases, it's best to pivot now and make the right move.

This shift, aimed at improving user retention, requires a complete redesign of your app's homepage. If it's currently a newsfeed, perhaps you should shift to a structured, step-by-step process. If your product is centered around sharing, perhaps you should shift the focus to content creation and collections. You might need to describe your product positioning radically, even benchmarking it against competitors. This requires a massive, multi-dimensional transformation, the more radical the better, to reverse low user retention.

Retention rates decline, but they never rise. Retention curves often follow a very regular geometric pattern. For example, many curves I've observed show the following pattern: no matter what the retention rate is on day one, it drops by 50% on day seven; no matter what the retention rate is on day seven, it drops by another 50% on day 30. Over time, the retention rate may eventually approach zero, or if you're lucky, it may remain around 10% overall. This decay pattern is very predictable.

You've never seen a curve that starts high, then drops, and then rises again; it's impossible. In other words, if your early retention isn't great, your late retention is likely to be bad too. You have to start strong to finish strong.

There are a few notable exceptions to this rule that are worth pointing out:

• Some products are very hardcore (such as online poker). User retention for these products may be relatively low, but the users who do stay are often extremely loyal and spend a lot of money, which proves that this model can also be successful.
• For products with network effects (be they social networks, collaboration tools, or anything else with network effects), new users may be initially active, followed by a temporary dip in activity. However, if the product is able to leverage the growing number of users to reactivate older users, a retention curve with a slight upward trend is common. This is extremely rare, but when it happens, it's truly remarkable.
• Revenue retention expands, while usage retention shrinks. One of the best and most important properties of the retention curve is that it applies to both users and revenue. So far, we've been discussing user retention, but unfortunately, it tends to decline, which isn't ideal. Revenue retention, on the other hand, is interesting because users who stay tend to spend more money on your platform over time.
• This is one of the biggest advantages of B2B SaaS products. For example, if you look at user cohort data for Slack, you'll see that its retention curve follows a downward trend, just like any other product. Some embrace it, while others don't. But for those companies that invest the time to deploy Slack, the product naturally grows, and your revenue from these businesses increases over time. This phenomenon of revenue retention increasing, not decreasing, is truly remarkable, but unfortunately, it doesn't apply to most consumer products. It's this very characteristic that makes B2B products have a much smoother business model than consumer products.
• The model for consumer apps is more like Amazon. You might initially buy just books and music, but as the product's functionality expands, you'll gradually start using it to buy more and more items. Because of this, the total lifetime value of a user in the product is essentially unlimited. We've observed a similar phenomenon at Uber: while the user cohort decays over time, people's initial rides for airport transfers gradually expand to include restaurant trips and commuting. As a result, the user retention curve tends to decline, while the revenue retention curve continues to rise.
• Retention is closely tied to product category. I've written in the past about the nature vs. nurture of retention. The reality is that many products have natural use cases, like collaboration tools or programming software, which you might use every day at work, but capped at five active days out of seven. In contrast, a vulnerability alert system is expected to be used less frequently. The same is true for consumer products; people check news, messaging, and social apps daily, but don't typically use medical reference guides frequently. Some apps, like weather or banking, have high retention despite low frequency of use. Meanwhile, categories like games, while addictive and frequently used, often churn after a few weeks of content consumption.
• The nature-nurture factor is crucial because it reveals the reality that many new products struggle to break through. If you're building a social travel app, but people don't actually travel that frequently, building a product centered around connecting with friends will be challenging. A more sensible approach is to embrace the low-frequency nature of usage and increase monetization by controlling the transaction process, or to integrate high-frequency use cases like restaurants and nightlife, like Yelp, while retaining travel functionality. Going against the tide is difficult, and there's only so much we can do.

For this reason, if you want to build apps with extremely high retention and frequency of use, you'll likely need to develop in areas that users already consider core, everyday products. This means successful apps are likely to take time away from other everyday products, just like how my Google searches dropped significantly after I started using ChatGPT frequently, and how I gradually abandoned other social news apps when I started using Substack to read and write blogs.

As the user base expands, retention rates often decline instead of increase. Even if you're lucky enough to create a high-retention product, people often tend to extrapolate existing user behavior, monetization capabilities, and usage habits to the broader market, believing that simply multiplying a few good small data points with the core big data will naturally produce impressive macro results. However, the reality is that as the user base grows, problems begin to emerge. For example, as you begin expanding into Android users and international markets, acquiring more customers through channels like paid marketing, you'll quickly notice a decline across all key metrics.

The reason is that high-quality users tend to appear early. Those with the greatest monetization potential, the strongest willingness, the highest digital proficiency, and the most active online behavior often start using the product early on through recommendations from friends. As new users are acquired through other channels, the product may no longer fully meet their needs. For example, if you develop an iPhone app for college students in Western countries and expand it to Android users in emerging markets, various metrics will naturally decline due to incomplete feature adaptation. While continuous optimization and improvement can be achieved later, I can assure you that the results will never be comparable to those of the early user groups.

So, the question is: as user growth increases and user quality gradually declines, are they still valuable? Can the product continue to be profitable? More importantly, can the core group of high-value users who joined early on be retained?

No wonder these early adopters are often referred to as the "golden group."

User churn is asymmetric. It's incredibly easy to lose users; in fact, most products lose 90% or more of their users within the first 30 days. Meanwhile, winning back already lost users is incredibly difficult. This asymmetry between acquisition and churn is the core characteristic of churn. It often gets so bad that it's easier to acquire new users than to try to win them back.

For this reason, lifecycle marketing attempts to reawaken dormant users by offering discounts or offers are often costly and ineffective. A more effective approach is to engage existing active users in natural product usage scenarios to reawaken dormant users. For example, if a professional tries a new project management tool but doesn't stick with it, bombarding their inbox with reminder emails is unlikely to win them back. A more effective approach would be to have their colleagues invite them back to the tool to participate in new projects. However, this strategy is extremely difficult and complex to implement, and is typically only adopted by products with network effects (i.e., sharing and collaboration features).

Retention is incredibly tricky and difficult to measure. When people talk about retention, they tend to focus on the first day, week, and month, but rarely discuss what happens two years from now. This is because when developing a product, teams need a short enough timeframe and easily measurable metrics to make decisions based on. So, despite the importance of annualized user churn or long-term monetization, people often overlook these metrics and focus on the more immediate, easily measurable ones. However, this approach presents numerous problems.

Unfortunately, many product categories are subject to strong seasonal fluctuations. E-commerce, travel, healthcare, and online dating are all prime examples. Even the way businesses use business software is subject to cyclical fluctuations. Seasonality can cloud your judgment. You might see a monthly or quarterly dip, but is it because a newly released feature wasn't popular? Or are users simply behaving differently that quarter? When retention data is significantly lagging, it's difficult to effectively assess performance.

Likewise, factors like bugs, new tests, and marketing campaigns can skew the data, and you’ll end up constantly reviewing reports showing fluctuations in retention curves, but each number comes with a caveat as the team checks to see if a new Android release isn’t introducing an irrelevant comparison.

Insane user growth coupled with abysmal retention is a recipe for failure. Many new product developers focus excessively on new user signups while completely neglecting retention. After all, if all they want is a continuously climbing graph, why not simply increase top-of-funnel traffic to highlight rapid growth? Then, once they've raised tons of venture capital, they can slowly address retention later.

This phenomenon is common in the industry today: a product sees a surge in users through TikTok when a creator promotes their app to millions of followers or a single video generates a surge in revenue. This phenomenon continues despite suboptimal usage and churn.

The tech industry has run this experiment countless times. The conclusion is always the same: viral products with poor user retention will eventually die, because retention is a difficult problem to solve. When the novelty wears off, user acquisition slows, and you end up with both poor user acquisition and poor retention. The higher you climb, the harder you fall.

We've witnessed this phenomenon in numerous scenarios. In the early days of social networking, many products achieved growth by acquiring user email addresses and address books and frantically spamming them, ultimately directing users to inferior products. Sometimes, simply by getting users to subscribe to an annual service with low-quality ringtones, companies attempted to monetize and profit. However, it wasn't until the advent of Facebook, through user experience innovations like news feeds and real-name registration, that products with both high virality and strong user engagement were finally created. The same phenomenon occurs in the mobile app space, where apps that rely on forced SMS invitations suddenly become popular, but if the product lacks stickiness, the entire model quickly collapses.

High retention is pure magic. You might be feeling a little discouraged after reading this, and I know launching a project can be tough. But when a product truly works, the feeling is unparalleled. Seeing a product achieve a 50% 30-day retention rate (which I see every few years) is incredibly impressive. I've come to realize that these short-lived successes aren't due to a systematic A/B testing methodology or a rapid iteration process. What really matters is that little spark of magic. This magic comes from a breakthrough insight into a market or customer need that, while seemingly obvious in hindsight, allows the product to achieve exceptional retention simply by being the first to realize it. This applies today to video conferencing software, disappearing photo features, or magical AI that can respond to any topic. This magic isn't achieved through iteration and metrics-driven testing alone.

The real problem

You might have read all of this and still have a big question: Wait, so how do you actually achieve high retention? (If I could answer this question with a deterministic method, my job as a startup investor would be so much easier, right?)

But let’s do our best. There are some clues buried in my point above: ideas really do matter.

If you want a high-retention product, you need to choose a category that already has high retention.

You need to choose a product category where you already use existing products every day.

You’ll be building a product that directly competes with it.

If you win, then you stop using that product and switch to your own.

That's a tall order, but I think thinking it through is a good start.

Of course, if you're building a product that directly competes with existing offerings, you might be skeptical, "It's really hard to get users to switch." And that's true. So, you need to decide whether to take a healthy dose of market risk, but only a moderate one, by launching a new and unique product that redefines the core interaction model. However, this innovation is more likely to be the 20% improvement than the 80% disruption. Ideally, you want your users to be able to quickly and intuitively understand this innovation within the first minute of using your product.

At this point, you can't avoid one of the most common and difficult questions investors ask: "Why does it work now?" Your answer must point to a new industry trend, such as a general technology like large language models, or a societal shift like the oversaturation of social media, that makes your innovative idea opportune.

This allows you to quickly capture existing markets and is more likely to achieve strong user retention early on. Timing is crucial. If you mistime your entry into a niche market with insufficient product differentiation, you'll find yourself transforming your retention problem into a user acquisition problem. The difficulty with developing a new web browser is that once it succeeds, users will be incredibly sticky. However, people are already very satisfied with their existing browsers, and getting them to try a new product is inherently costly and complex.

That’s why I don’t blame those who come up with ideas like “Cursor for X” or “Figma for X”, just like the “Uber for X” concept in the past. They try to leverage existing markets and behavioral patterns to avoid huge market risks.

If you can accurately grasp the differentiated advantages, seize the market opportunity, meet the needs of massive users, and find the right core product positioning, then this model can indeed succeed.

How to open new markets?

The natural counterargument is that new markets are often more exciting than existing ones. Shouldn't the tech industry be about building entirely new things, not innovating 20% ​​on old ones? While that's certainly true, I think these products represent a tiny fraction of the market.

My counterargument to this is that most products actually inherit some kind of "old stuff," even if those predecessors are quickly forgotten.

Before Instagram, there was Hipstamatic. Early on, this app topped the App Store's paid photography category, demonstrating the enormous market potential of filters. Just as Google wasn't the first search engine (it was actually the tenth entrant, following platforms like Lycos, Excite, and Infoseek), these examples demonstrate both the strong user demand for search functionality and the challenges of early search engine commercialization. Tesla wasn't the pioneer of electric cars, nor was the iPhone the first smartphone. History has repeatedly proven that it's the tenth-generation innovator that truly determines market dynamics. This phenomenon is known as the "latecomer advantage," and I find this concept quite insightful.

Yet sometimes true innovation does happen. Uber was born by transforming an existing offline ride-hailing experience into an online application, not by building on a previously successful ride-hailing app (Lyft was just a weird bus booking service at the time). Or consider ChatGPT, which took OpenAI five years from conception to its full rise to prominence in version 3, without any existing blueprint to follow. These kinds of innovative journeys are remarkable and are the driving force behind the tech industry's success, as they create entirely new product categories at the cost of genuine risk.