”
This article is based on shao Zhen’s speech at GMTC 2016. Reply keywords [Shao Zhen], get the complete PPT download link.
Introduction of the old driver
Shao Zhen, former Employee of Google Search, Square full stack engineer and Mobile Growth Tech Lead, focuses on the business Growth of Square App and the specific application of Growth methodology in Mobile development.
I used to work at Google, I came to Square around 2013, and I’ve been working on Growth ever since. This is probably the least technical part of a technical forum. When you think about Growth, people think it’s amazing. Teacher Fan Bing once wrote a book called growth Hacker, the title of which is very confusing: hackers are like superman, sitting alone in a dark room. In fact, Growth is so real that all of Silicon Valley is talking about it. Not only are they talking about it, but they are doing it.
As far as I know, Growth is also a very popular concept in China. First of all, every company is talking; Everybody thinks he knows what Growth is doing; Growth has a little concept called viral marketing that everyone talks about. Second, Growth’s charm is why so many people come to the lecture even though no one knows me.
We’ll talk about our Growth
The first part is “Growth again.” Why talk again? The word “Growth” has become very popular, but behind the scenes, I see a lot of misconceptions about it.
Review images
So the first question is, what is Growth? My short and accurate definition of Growth is that Growth is “the effective Growth of the business.” There are a few key words: implementation, business, effectiveness, and finally growth. Growth hacks, I translate that as, not “Growth hacking,” but “Growth method.” The word “fa” in Chinese contains both “tool” and “thought”. Growth is a relatively new idea, and Silicon Valley has been talking about it, doing it, groping from the bottom up. It was only in the last two years that someone refined the concept of Growth as a methodology.
So why is it so popular in China now? Because now Chinese companies are feeling the pressure. Growth can improve efficiency, whether it is engineering efficiency, or efficiency in using money and acquiring users. This increase in efficiency is very important to the company. In the case of particularly good market growth, greater value will be gained by using resources to realize product features; Growth matters when the market is cold, competition is stronger, and everyone has to calculate the cost of acquiring customers.
What are the difficulties for Growth in China and silicon Valley in general? This is due to the nature of Growth itself. In a small company, Growth is about adapting to the market, which means that what value the company provides to users ultimately determines how many of them stay on the platform and how often.
There’s a famous article on Medium called “Why Startups Shouldn’t Have a Growth Team?” Why a small company should not have a Growth group. It smacks of clickbait, but it has merit. Since the whole company is a Growth team, the CEO should take the lead in Growth. In the early days of a company, Growth is used to get close to the essence of the product and find its market foothold.
In the later stage of development, when the market demand is stable and what the company does and how it makes money is stable, Growth can move to a more advanced position to spread value to users and make users’ behavior converge to the value you want to provide. At this time, Growth becomes a way to show the company’s value to users.
Growth is good at how companies put limited resources, whether money, engineers, product managers, etc., into an infinite opportunity space. Growth finds the point at which the maximum benefit can be obtained at the minimum cost. But it’s hard to do.
Growth represents the crossover approach of engineers. Most of the Growth in Silicon Valley is driven by engineers, who are not ordinary engineers, but need good knowledge of product and operation, design and data. Few people are competent.
Domestically, I think there are some additional challenges. There are some misconceptions about Growth: buy traffic is Growth itself; Attach importance to non-technical operation; Product iteration is particularly fast, fast to the data burial point, the whole data collection is not done; Domestic entrepreneurs have the habit of making decisions from the top down, and the boss has the final say. There’s also the challenge of not having the tools in place. Silicon Valley has reached a stage where the startup environment is more mature, and many of the tools have been around for years. But it’s hard to find a particularly reliable tool in the country, considering the background of the tool and whether it belongs to a competitor. These are difficulties caused by missing data and inadequate tools.
Let me break down the so-called “growth method” and say it in a simple, straightforward way, as witty as possible.
The first is the core topic that Growth cares about. We split Growth into four parts: acquisition, conversion, monetization, and retention. The AARRR model is slightly different from some of the technology sharing we’ve seen before, but it’s essentially the same. AARRR gives the illusion that Growth is a funnel, that it tapers off. Must monetization be generated by loyal users? Not necessarily. Must recommendations be generated by paying users? Not necessarily. Can new users do it? You can.
Let’s look at a real life version of Growth. In reality, Growth focuses on these core topics, but these topics are actually intertwined and very fragmented. If there’s one place you can really think of it as a strict funnel model, it’s only user conversion: start with “potential users know your brand” and end with “turning potential users into loyal users.” Each step is strictly dependent on the success of the previous one.
What is the basic unit of Growth?
First observe patterns of product or user behavior. For example, this pattern might mean that the user is lost very quickly on a particular page. Based on the observed pattern, design a solution, which assumes a solution, and design a metric, which data tells me clearly whether the solution is good or bad, and how good it is.
After the scheme design is finished, we need to do engineering implementation. If the experiment is successful, it should be applied and monitored, because short-term results are not indicative of long-term benefits. Finally, if there are more powerful data analysts or product personnel, they will further explore the enlightenment brought by this experiment.
The picture above is a metaphor. It’s like developing a drug, you study the disease, you develop the drug, you do clinical trials, and then finally, you have to monitor the drug, and then you go back to theory. Making medicine in America is very, very complicated. Medicine is used on people, and people are the foundation of the country. The whole system of medicine is people-centered. The same is true for Growth. User experience is sensitive stuff, so be very careful, or once your users leave, it’s very difficult to call them back.
In this process, some tool support is required. For example, A/B Test is A very important tool when conducting experiments.
There are several more steps that are important from the operator’s point of view. First of all, the people who do it need to understand their product, all the people who do Growth, need to be statistically literate, need to understand the users, and understand how difficult it is to implement the project.
In fact, the operation of the Growth group will organize it into an experimental pipeline, which will track the experimental ideas, which have been started and which have been completed in some way. The influence of the completed experiments will be calculated, and the influence of the successful experiments will be accumulated, which can roughly measure the success of a Growth group. That’s a very rough statement. There are experiments behind any Growth, and each experiment is very small, maybe hundreds of experiments per quarter. While the benefits from one experiment are small, the sums add up.
One of the things we talk about in Growth is: I have lots and lots of ideas, which one do I do first, which one do I do later? At this time, I will draw this map in my mind to think about each experiment, how big the opportunity is, and how big the project cost is. If engineers are familiar with engineering, they should have a very clear estimate of the cost of the project. If they are familiar with the data, they will also have some estimate of the opportunity. In fact, they are calculating the cost performance.
Review images
There is an extra dimension to this diagram. The circles in the diagram are large and small, indicating that the same idea at a point will have a series of ideas, can produce a series of experiments. We tested one of these experiments, which might suggest that more gold can be found in the same pit.
Here are some of the lessons from the Growth practice, which are rather piecemeal. First, the most important part of Growth is retention. No one is a fool, and at some stage they may be fooled, but they will ultimately stay on your platform, depending on how valuable your platform is.
Review images
In terms of organizational structure, a minimum Growth unit should have at least engineering, product, data, design, and operations. Not every part has a person, but every part has someone who can cover it. It’s driven by engineers, because engineers are the people on the team who know the product the best. He knows whether or not you are triggering the features of the product, he knows exactly when he is writing the code the weaknesses of the performance, he knows where the best burying points are, and even if the product is being modified, how much the engineering cost is, the engineer must have first-hand information, and any actual changes must be implemented by the engineer. Being driven by engineers is actually the least costly way to communicate.
From a company’s perspective, Growth requires a top-down strategy: the company needs to protect this idea structurally. If you ask a technical engineer to pursue technical difficulty, and his job is measured by this thing, he’s not going to put the impact on the business at the heart of it, and he’s forced to think about things that run counter to Growth’s thinking.
In the end, rapid iteration is not essential, but it can be done very smoothly. No one knows until the final figures come out. For example, you want to experiment with the ability to support anonymous users, and you want to see if users retain more people without registering. So, do you go for App wide anonymous support right now, do you go for a product that doesn’t require login at all, or do you do a very small experiment, such as delaying a step or two to ask for login, to see how much of an impact the fact that you don’t need to log in has on users?
Growth methodology
Growth is a methodology, and it’s beautiful, but it doesn’t work for every company. You need to at least agree with, or agree with, a few ideas in your heart.
The first idea is whether you think the success of a product is determined by whether users like it or not. For example, if you do government projects or market scarce functions, users may have to use them. Therefore, Growth only plays a very small role in your user behavior, so you don’t need to consider Growth.
Second, do you agree that everything is ultimately measurable? For example, can you quantify product likeability, can you quantify customer acquisition strategy?
Finally, is Growth just the icing on the cake? Because it takes a lot of energy to do this, Growth in Silicon Valley often hangs under the director. It takes people, resources and coordination to dig out the big hole of Growth. It may go well at the beginning, but you soon need to consider whether it is worth doing. From silicon Valley now, the vast majority of companies have dug out this hole, which is a certain reason.
At an abstract level, the methodology of Growth is how you think about Growth and how you do Growth. Now, I don’t think I’m good enough to summarize Growth’s methodology, so I’d rather quote a passage from a master in this field, whose name is Zhuangzi.
“The section has a space, and the blade has no thickness, with no thickness into the space, the slackness of the knife will have room”. There are “simple fixes, but big ones,” like “cracks in joints.” If the blade is not thick, Growth is like a sharp knife. To “use the thickness to enter the space”, that is, to use the most powerful resources to do these “joints” with low cost and large impact. In this way, “after 19 years, the blade is like a newly developed knife”.
When you see a service intersection, or a user experience or performance sensitive point, you have to be careful. This is called a “family”, “blood cross point”. You should pay attention to, to have the heart of fear, “see its difficult, fear ran to quit”. So what do you do? Zhuangzi tells us, “look until you see, act late, and cut very little”, that is to say, you should look first and do it later. Each cut is very small.
In the Growth methodology, this is very typical, you observe, you use the data to guide you, you figure it out, and then you do it, and each experiment is very precise, very small, but it always adds up to something. At this moment, Zhuang Zi vividly described what he would look like after he had solved a big problem. “Standing with a sword in hand, looking around and hesitating full of thoughts”, he hurriedly wrote a report and sent it to the whole company. He was very happy. Finally, “good knife and hide”, that is, to clean up the experimental code.
Opportunities and challenges for Mobile Growth
In Mobile, the concept of Growth can also be used. It’s just that the platform is different and there are specific challenges and opportunities. It’s hard to do cross-linking, it’s hard to do attributes, it’s hard to iterate quickly on Mobile because of release cycles, it’s hard to do A/B tests, and there are download barriers. This is a challenge unique to Mobile.
At the same time, there are some unique opportunities within Mobile. A mobile phone is a personal device, which opens a Square and basically knows who you are. You can track many of a user’s actions on the platform over time. From a product design point of view, there is no need to log in, which is also good.
Second, the device is the entrance, turn on the phone, the App is already there. No longer subject to the search engine clamp, only need to be used by users. Thirdly, the device is the channel. App opens the communication channel for service users when it is on the phone. This channel is resident and you can find your users through it, which doesn’t exist on the Web.
User Onboarding top-level idea
User Onboarding is very important. After obtaining the App, the User is ready to download the App, which is the starting point of User Onboarding. If you are familiar with this graph, you have a base of Growth. This is the most typical Funnel Analysis, where the volume of users is used as an indicator and then the flow is broken down.
As you can see, each big red column is equivalent to one user loss. If 100% of the users are at the starting point, maybe only half of them will download the App, and a small number of users will be lost after the login page. The registration step requires the user to provide identity information. After gathering the information, let him experience the core features of the product. Then use it a second time, use it a third time, and pay for it.
In this process, the number of users is decreasing. One of the most basic Growth methods is to look at how much your users are decreasing at each step, optimize accordingly, and bring some users back.
How willing are users to download apps? Do some ASO on the download page, for example, to test what kind of images and titles are more attractive. You can make some very precise assessments of how much better one is than the other, and what’s better? The size of the App itself also affects the willingness to download.
Another example is to support anonymity to reduce user churn during registration. Maybe the user doesn’t want to continue because they need to enter email. Consider canceling or postponing registration, or access to a third party, such as Weibo or wechat, to ease the difficulty of login. As you gather information, turn what you manually enter into choices and guess what you can guess. By Aha Moment, maybe next time he will forget the App and lose users, at which time he will push some information. When it comes time to actually pay, the user will give up because the payment process isn’t good, or the user thinks it’s too much, so offer more flexibility.
This is the most typical way to do funnel analysis in Onboarding, to find some optimization methods.
Funnel analysis is a very basic and useful tool. But it has its problems. The first is to create an antithetic mindset: no matter how good your product is, users will always leave. Just as the number of users will fall no matter how good your landing page is, the number of users will fall no matter how good your Aha Moment is. Unlike users, product designers look at funnel analysis in a statistical way, and users are not retained in a statistical way.
Second, when the big problem is pretty much solved, you focus on solving these small run-off problems, which may be largely due to performance problems, and Growth slowly becomes performance optimization. Third, the user doesn’t want to pay, not necessarily because the payment page isn’t good, but probably because he doesn’t think it’s worth the price, and that’s a problem you can’t solve with funnel analysis.
One way I suggest is to look at the process from the user’s point of view. Here, the horizontal axis is still the process, and the vertical axis is the user’s willingness to use the App, which is roughly described by a number. First of all, let’s say the willingness is 100, at this point, the user needs to spend his willingness to download. The user comes to the login page, and if the login page provides good information and materials, the user has more expectations for the future, or wants to know more about the product, his willingness may increase.
The user will still be healthy during the sign-up and activation process, but if you do a good Aha Moment and turn him on again, you can reactivate him to be more willing to pay later. If the product experience is well done, each use behavior can accumulate willingness and ultimately support a realization. A good product, in this line of thinking, is always thinking about what the user really thinks.
Another set of optimization methods can be derived from this idea: whether some positive improvements can be made. Some optimizations can be made on the login page first. For example, make it clear exactly what the App is worth to you, including the referrer’s profile picture, providing context and turning the social credit of the recommendation into the App’s own appeal. If the 2B product needs to collect the ID number, phone number or enterprise information, the user may not be ready in this process, then you can tell him on the login page, “Need to prepare these information before registration”, to avoid the user stuck in the half of the process and lost.
Some guidance and explanation can be given throughout the activation process. Why you need a phone number, for example, is not to sell personal information to the next buyer, but to find you quickly if your business goes wrong. Emphasize and explain the “value to the user.” This “value” is often considered when you design the product, so you should also consider how to present it when designing the process. For example, when doing a Moment, emphasize that the user feels valued; Make some interesting animations, or even substantial rewards, to increase user interest. Aha Moment is an important Moment that can increase the interest of many users.
Once the first and most critical feature is done, guide the rest of the new features and push them out at the right time so they’re still interested in using them. Request access carefully to avoid making users feel uncomfortable. Before charging, what is the function of charging, what value can be obtained through charging, what benefits other users have obtained after charging, and what is helpful to their enterprises can be gradually shown to users in advance. This is all in preparation for charging.
Review images
There are two key points of User Onboarding: one is to transfer value so that users can see where the value is; The other point is to remove pain points, and you need to remove pain points that users encounter in the process. If these two points can be done correctly, it is a very good Onboarding experience. Third, when Onboarding, create a channel for users to allow push notifications, for example, in order to guide them to use more new features later.
Mobile Growth key technologies
Here are a few key technologies for Mobile Growth. Growth is a very complex process from the top down, and the techniques used are very complex. Here are some of the most important and critical parts.
Review images
Here’s a summary from the Mobile Growth Tech Stack. The authors summarize the Tech trends of the year in such detail that only very large companies will use many of them, but it’s a good summary.
A/B testing system was established
Review images
The first one is A/B test. In the past, repairing a computer had a concept, the computer is broken, exactly where is broken, pull out the memory, replace the memory, if the system is fixed, then the memory is broken. You can take the same approach when analyzing the product flow. Design A mechanism called A packer that tells the product: Let this user perform action A and let that user perform action B. Through data analysis, it can be seen that the downstream behavior of users who implement A is better than that of users who implement B. This is the basic A/B test.
Review images
The communication “A or B” between the pacer and the product is the most basic part of the A/B test system. Add A few more “details”, A/B Testing system that can be used at scale. The green components in the picture are called “details”. The grouping tells the products how to group, and the grouping results and downstream behavior are “buried” through the Log, which needs to be processed. After Log processing, on the one hand, it is used to guide the next experimental flow allocation, on the other hand, it is used to calculate indicators.
This indicator may be presented to the experimental designer, product manager, or engineer as a chart on the panel. Data people may need some additional tools to analyze metrics in depth. The experimental configuration tool is the entrance to the configuration experiment. This system not only determines how the pacers are grouped, but also determines which indexes need to be calculated in the end. It also has some relations with traffic allocation. When QA is testing A product, it may be necessary to accurately control whether the experimental behavior is A or B during A test.
This is a more complex multi-party system, and it is difficult to write them all right at once. Fortunately, most of the time you don’t need to worry too much about the technical details, and there are many third-party tools that can help you complete the process.
For A/B test system, more is not technically how to write, but how to choose and use. If you want to design A system, or evaluate A/B test system, you need to consider two things first, the first is the core, the second is optimization.
The core part is “what does it take to be the MVP of an A/B test?” Optimization refers to the expansion direction. Avoid hurting the user experience, this is the most basic requirement. Sending A request before deciding whether to choose A or B, which takes 5 seconds to return at the latest, and the entire interface stops, certainly hurts the experience. For example, if the back end accidentally changes the implementation and sends back an unstructured response, the client reports an error saying, “Illegal experimental grouping data,” which of course hurts the user experience. A system failure, can be rolled back to the correct product behavior, elegant processing, so that users do not know, this is the most basic requirements.
Second, statistics are valid. There are specific statistical requirements for how to design an experiment or how to evaluate the results of an experiment. For example, whether the real control variables were achieved, whether the design of the experiment was naturally biased against group A or GROUP B, whether there was subjective misinterpretation of the statistical results, whether there were cross-effects between experiments, whether the experiment had A novelty effect or A start period, etc. Experiments involve A lot of statistical knowledge in order to correctly design and analyze, A/B test system should calculate, judge and display results based on correct statistical assumptions.
There are a lot of details in the “optimization” of this system, such as performance considerations, scalability considerations and so on.
Trends and usage of deep-link
Deep-link is very popular in China. For example, there is a scene where I search a news keyword on Baidu and the entry is provided by a news App. When I click the entry, a news App is opened and the specific news page is pushed to me, which is deep-link. In other words, you can jump in from outside the App and find a page of Deep content inside the App. This is called a deep-link.
In some complex cases, there is a variant called Deferred deep-link that can achieve a similar effect without the App being pre-installed. It’s not technically easy to do, but it can still be done. As you can see from the picture, the effect is good, eliminating some steps, each of which originally had a loss of users.
The implementation details of deep-link are not detailed here, but many third-party tools can help you implement it. In principle, it is relatively simple. Through a short link, the parameters are stored in the third-party server. When using this link, the third-party will be asked and the parameters will be retrieved through fingerprint matching. The principle is not particularly complicated, but it is difficult to implement.
In addition, what I want to talk about most is: what is the impact of deep-link on product design and Growth?
First, it solves the dilemma of cross-links and attributes. Especially on iOS, you go to the App Store and you lose all of your parameters, but Deferred deep-link will help you get them back.
Second, it changes the reality that apps are only available on the App Store. With deep-link, traditional search engines or content portals can also become portals for apps, which is very beneficial for content-oriented apps to spread virally.
Finally, entry diversification and business slicing. The entrance to the App is no longer limited to downloading, clicking on the App icon, registering, and using. Users coming in from a specific content or sub-feature expect a quick flow to the content they are interested in. Whether or not to optimize these new entries individually can be a new problem and opportunity.
This change also represents a trend towards “business process slicing”. This trend tells us that putting a big, comprehensive App in front of users may not be a good choice. Facebook’s main App is 100+MB, which basically contains all the features, but future apps will probably try to avoid this as downloading such an App is a burden for users.
An App has several different business entrances, and each business pair should have a relatively independent module, or even a sub-app. Or as Google’s Instant App advocates, only a small section of App functions can be loaded, and simple functions can be realized in this section, and then the whole full-function App can be promoted. Such a move could become a trend.
This trend is not only an external slicing, but also an internal slicing. Even if you don’t care how you access it externally, you can slice it inside your App. A big feature of mobile apps is that the screen is small and there are many partitions. The early stages of designing a product always want the process to be clear, focused, and uninterrupted. But later, when you have more functionality, you want more flexibility between processes.
A compromise might be to set up each business process starting point as a deep-link target, so that jumps can be expressed in a concise URL that makes jumping easier without giving away too much implementation detail. This internal flexibility allows for flexible Growth.
The content in Growth is dynamic
Dynamics is a big topic, but there’s only one thing we’re going to talk about here, and that’s how Growth sees dynamics.
Dynamic in Growth’s eyes is not a large and complete system, such as React Native. Growth’s biggest concern is doing something with minimal cost. Let’s look at a function switch, on/off, which is dynamic; Dynamic content filled with static template, is dynamic; Dynamically adjusted components to build pages that are dynamic; The process of a business module can be captured from the server side, which is also dynamic. When considering dynamics, there is a tradeoff about how a specific use case will be implemented. Why React Native if you only need to change the title?
A high degree of mobility sacrifices testability and increases the amount of late resources needed for use. Function switches do not require late resources. Content templates require the participation of at least one author. Specific assembles the page to need the designer to design. If we discuss dynamic processes, both UX and PM will jump in and ask: is this process scientific and consistent with the philosophy of our product? React Native’s dynamic architecture may require the entire development team to use. Since Growth is about cost efficiency, dynamic is also about efficiency, doing what you want to do at the least cost.
conclusion
Growth is real, widely used in Silicon Valley, and has now come to China. There are many objective reasons why it is not used in China. But conditions are now ripe, such as the need for efficiency caused by fierce competition. Eventually you have to go with Growth. The only caveat is, don’t rush to make a big, all-in-one system. Go with third-party tools. Because doing the most with the least cost is the core of Growth.
This Thursday at 8:30 PM, the seventh live broadcast is hot online, please scan the code to sign up!
del
(Click on the title)
An article to understand the status of open source database: scheme, design ideas and application scenarios
4 years, the story of a techie who became a CTO (Part 3 of series)
BAT Declassification – What is Driving technology Development?