The One Concept That Keeps Users From Leaving, According to Toss Leaders | PO SESSION

Hello, I'm Lee Seung-gun, team leader at Toss. Following the previous PO session, I've prepared a second session today. Before we start, I want to mention one thing: we're doing this for recruitment. We have a top-tier PO training process to create the best products. If you're a 5-year PM or PO, you'll be able to work with concepts and tools you've never even thought of at your previous job. So please come to Toss, let's create great products together and build a legacy in Korea. You can find the PO position at toss.im/career. Please apply. Today, I'll be talking about retention and activation.

POs and PMs probably all know the AARRR framework. The most important metrics are usually retention and activation. I'll cover some important concepts for improving these two metrics that aren't often found in books. Before we start, I'd like to address some questions from the last session. A server developer on my team asked a question about my first point from last time: when you force a specific action on power users, if the churn rate doesn't increase, it means they're converting to power users. I was surprised because I knew I'd misspoken while watching the edited video. That was incorrect. To be precise, when you force a specific action on users, the churn rate should improve. If it improves, CC increases. The statement that if it doesn't increase, they'll convert to power users isn't wrong, but it's a bit awkward. More accurately, if you force a specific action and the churn rate decreases, then all users are converting to power users and CC is increasing.

Many people also asked questions on YouTube. One person mentioned that even with downtime, traffic recovers, but with many competing services, wouldn't CC drop? Hasn't Toss experienced this? In our early days, we had several outages and were famous for them. I remember the nightmare of Toss being a trending search on Naver and having to restore servers. But in our case, we recovered almost completely with little change in CC. Even when we only had simple transfers, there were many alternatives. Despite having alternatives, we recovered without major issues. This has been validated at Toss, and I think most of your services will also see little change in CC.

Another question was: why is outflow a ratio and inflow an integer? The person already answered correctly. As I showed in the last lecture, churn is proportional to the amount of water in the pool, so churn rate is proportional to user count. New user inflow is an integer not proportional to user count. That's why they're expressed differently. Another question was: what if we add a time concept to CC? Since it doesn't change steadily like compound interest, shouldn't we consider acceleration? That's true. When MAU and CC have a large gap, MAU catches up quickly, but the growth rate slows due to diminishing returns. However, creating the gap between CC and MAU is often more important than how fast it closes. Knowing this practically may not be very useful.

Now let's get to today's presentation. The question is: what should we do after finding Product-Market Fit? The most important criterion for determining if you've found PMF is whether retention has plateaued. If users keep dropping off like the green line, you haven't found PMF. If it flattens out like the blue line, meaning users continue using it periodically, you have PMF. This means users keep coming back without advertising because the product has intrinsic value. Now, a very important hypothesis is resolved. The most important criterion separating startups from non-startups is the size of uncertainty. About what? Product, market, customer. These three things. As they become clearer, you move to the management stage. The startup stage is about low certainty in product, market, and customer. The process of moving from uncertainty to certainty is the lean startup concept. Eric Ries's Lean Startup is an immortal classic that increased global startup success rates by over three times.

A lot of this uncertainty has been resolved because finding PMF with a retention plateau means two things are solved. First, we know what structure our product should have. Second, we've found our customers: the people who decided to keep using it. Those who dropped off aren't our customers. So finding PMF means cohort retention is plateauing. Today's topic is what we should do with this retention plateau graph. After finding PMF, the first thing we should do is analyze this retention curve. Retention analysis and improvement is the first priority. AARRR should be calculated from the back: fix retention, then activation, then acquisition. But usually people do it backwards: run ads, fix the activation funnel because sign-ups aren't converting, then fix retention because users are leaving. That's the startup hell version. You should fix from the back: create a strong service where users don't leave, then make the user acquisition path smooth, then advertise.

How do we analyze this retention graph? Three ways. First, interview users who left. Do usability tests. Why did they leave? They clearly had a good experience at least once, so why don't they continue? Second, why do these users keep using it? Who are they? Why did they decide to keep using it? What's the key? What distinguishes them from those who left? This needs to be found through data analysis. These are our customers and the future customers we need to acquire. Why do we need these usability tests? They become the criteria for increasing CC later. Finding retention means we can now calculate CC. Calculating CC means we can think about how far we can reach. Eventually, when MAU hits the CC wall, growth stops. As I mentioned last time, when that happens, adding new service layers increases CC. Toss launched credit checks when transfers hit a limit to increase CC. The hint for what to do next with CC comes from usability tests.

We need to analyze what users we couldn't attract with our current service structure. What do they use instead? How can we bring them in? This analysis is necessary. Usability tests help us understand what use cases we're not satisfying and what features we need long-term to increase carrying capacity. A tip: when asking users, don't ask 'Why didn't you use Toss for the second or third transfer?' Negative questions aren't something humans can process well. Instead ask: 'What do you use for transfers now? Which bank? Why don't you use simple transfers?' This way you can identify the real cause. UX researchers say about five usability tests can reveal consistent answers about why users do or don't do something. But I usually do about twenty. I need that many to sense market opportunities and discover patterns. I did direct usability tests when planning Toss 3.0 and 4.0, gaining valuable insights about how to get non-users to use our app.

A disclaimer: these user tests don't directly help improve retention. Covering those use cases might increase retention, but developing new services takes time. What we need to do immediately is improve retention. These tests are for long-term planning when we reach carrying capacity. What we need to do urgently is analyze who our retained users are and why they use our service. When users who keep using it over time flatten out like this, we can say we've reached PMF. Now we can finally answer: who are our customers? This requires understanding personas through gender, age, region, industry, company size. But more importantly, where this retention plateau sits is crucial. I usually benchmark at 20%, 40%, and 70%. If retention plateaus around 20%, that's a borderline case. Below 20%, you can't build a company. At 20%, you can build a decent company, but it means only 2 out of 10 users stay. With 50 million people, that's only 10 million. MAU would stop at 2-3 million. Such services can't become very large businesses, maybe 200-300 billion won companies.

But if retention is around 40%, you can build a unicorn. If it's above 70%, you can change the world and innovate an industry. Such services become used by most of the population and gain the power to transform society, beyond just company valuation. For reference, when Facebook first launched, retention was about 68%. Instagram was similar initially. What about Toss? When we launched simple transfers, I recall it was about 68%. Very high numbers. We're a great company. It's lower now. So retention height determines your company's valuation. It gives you a sense of where MAU will land if everyone uses it. Finding retention is also difficult. Before PMF, you experiment daily, run out of cash, endure hunger. Then you find PMF! Retention exists! But it's only 20%. Then you might need to abandon this hard-won opportunity and aim for 40% or 70%. Because arriving at a small impact is the most bitter outcome. So remember that retention plateau height often determines company valuation.

Now we need to analyze who these users are. For Toss simple transfers, we found it was mostly people in their 20s and 30s. Specifically, it was mainly people in their 20s who were college students or new to the workforce, frequently splitting bills with friends. They transferred money a lot and felt frustrated with existing internet banking. So convenient transfers became more important. Understanding personas is necessary, but more importantly, we need fundamental data analysis on why some users stay and others don't. Because fundamentally, they're all the same people. Retained users often share common characteristics. This is called the Aha Moment. Today I want to focus on this. The key to improving retention is launching an Aha Moment movement within the company. When your company finds retention and spreads through marketing, it typically has 50-120 employees. Then everyone is no longer aligned. The company becomes complex with HR issues, hiring becomes difficult, opinions multiply. Product direction often gets lost at this stage. To prevent this, an Aha Moment movement is needed.

The Aha Moment is the moment when users experience the product's core value. Users who pass this moment continue using the service, while those who don't experience it stop. It's the singularity that makes users keep using the service. If 95% of users who perform this action retain, that's the situation. I drew this graph: the green circle represents retained users, and the specific action group shows that over 95% of users who took this action eventually stayed. This is the 'Aha! This service is amazing!' moment. When more users experience this Aha Moment, the retention plateau can rise. The Aha Moment is defined quantitatively. All company members can simply and consistently say 'Our Aha Moment is this.' It's the decisive moment when users stay with the service. For example, Toss simple transfers' early Aha Moment was transferring money at least twice within 4 days. Users who did this almost all stayed. Those who transferred less than twice or took more than 4 days didn't stay. It's quantitatively very clear. So our job is to make users transfer at least twice within 4 days.

A new employee joins as the 75th member, doesn't know anything about the company, it's a chaotic startup, they're adapting too. What's their job? To make users transfer at least twice within 4 days. They don't know anything else, but that's all they need to do. So you start this movement: anyone who makes users transfer at least twice within 4 days is doing well; anyone doing anything else is doing poorly. How much the company can focus on this Aha Moment largely determines future scale. But the Aha Moment isn't just expressed numerically. It should also make sense intuitively. If it's a simple transfer app, when would users keep using it? When they've transferred multiple times, before they forget. So it should also be qualitatively reasonable. I want to emphasize: Simplicity, not Science. This isn't about creating rigorous science. 'Why isn't 3 times in 4.5 days 100%?' Some users might transfer 4 times in 7 days. True, but that's not the point. When pursuing the Aha Moment to improve retention, the organization becomes too complex. Everyone needs a simple sentence they can follow blindly. It should be simple enough that even a 13-year-old middle schooler can understand.

When finding it, use both deductive and inductive reasoning. Deductive: since we're a simple transfer service, transferring multiple times should lead to the Aha Moment. Inductive: looking at users who transferred a lot, who stayed and who didn't. Both are correct. But I must emphasize: the Aha Moment isn't a silver bullet for retention improvement. The increase from optimizing it is only about 20-30%. The rest requires increasing carrying capacity through usability tests. It's not a silver bullet, but it's a very useful concept worth discussing. The Aha Moment was widely used abroad. The best at it was Facebook. Facebook's Aha Moment was: connecting with seven friends within the first 10 days. Then you retain, otherwise you don't. This proposition remained valid even when Facebook had hundreds of millions of users. Even with thousands of employees, everyone knew: 'What should you do? Connect seven friends in 10 days. Okay, got it.' Slack's founder said: when 2,000 messages are sent within a team, the team retains. Dropbox said: when at least one file is saved in a folder on one device. Zynga said: returning within a day of signing up. Twitter said: following 30 people. LinkedIn expressed it similarly but didn't disclose exact numbers.

When you use Twitter or Pinterest, after signing up, you're prompted to follow people. Why? Aha Moment. Is it really backed by data? Yes, it is. They found this causal relationship. Just do this one thing blindly and retention happens. Find that magical one-sentence rule. They found it and focused on it. Toss did the same. When you first signed up for Toss, you could make about five free 1-won transfers. Why five? Because the Aha Moment was twice within 4 days. No need to give more. Would giving 1,000 won improve retention? No, it was linked only to transfer frequency. But now it's gone. Why? Because Toss is no longer just a transfer app. So the whole world has been using this Aha Moment movement to overcome organizational complexity, focus on one number, and improve products by boosting retention. I remember when I first joined Facebook in its early days in Korea, I had almost no friends on it. Facebook asked me to input my NateOn ID and email. After connecting NateOn, my friends appeared and it asked if I wanted to connect with them. I realized Facebook had this clever strategy behind the scenes.

The Aha Moment takes this form: perform action XX at least ZZ times within YY days. Why? Naturally, if the service has value, this action will deliver that value. We're discussing the process after finding PMF. If you don't have PMF or retention, trying to find the Aha Moment is wrong. When users experience the core value of the service, like making a simple transfer, they receive value. When they receive this value multiple times, they retain. That's common sense. So the Aha Moment involves actions related to core value. If you don't have PMF, this isn't the time. Go back to experimenting until you find a retention plateau, eating ramen and staying up all night. The second element is YY days. Modern people deal with too many services and complex problems daily. Even with a great experience, it's hard to remember it for long. So before completely forgetting the initial impact, you need to use the service within a certain period to decide to keep using it. If too much time passes, you forget the service even existed. Usually, this date should be before you forget. To achieve this, companies often create fantastic initial app experiences.

ZZ times: ideally, if the service value is great, one use would be enough for retention, but that's almost never the case. Most services need two to ten uses for retention. If the service provides a shockingly good experience, ZZ could be much lower, but often multiple uses are needed. What if the Aha Moment is impossible to execute? For example, if you built Facebook and found that connecting with 40 people within 3 days leads to 95% retention. That's almost impossible. I'm not sure I even have 40 friends. If this happens, even with retention, you need to seriously reconsider the product. Such a difficult Aha Moment in such a short period would only discourage motivation. Statistically, even if 95% of users who achieve it retain, now isn't the time to find the Aha Moment. You need to reconsider action XX. You need to strengthen it. If the current service can only produce such an impossibly difficult Aha Moment, you need to increase XX's value. For example, for simple transfers, increase the transfer limit or the number of free transfers. If the Aha Moment difficulty is too high or you can't find an action that 95% of users do, you need to create a more attractive, more 'wow' action. You need to transform XX into XX' through product improvement.