4 questions about data-driven growth that you must know!

" Data-driven growth" began to be mentioned in China in 2015. As part of "Growth Hacking", it has gradually been accepted by product, operation , and data analysts in the Internet industry as the concept of Growth Hacking became popular.

Figure 1. Growth hacker search index (Image source: Baidu Index)

However, most of my friends have only heard of the term "data-driven growth" and lack a systematic understanding of its methods. The reasons for this are, first, a general lack of excellent data analysis tools among companies, and second, too few short and effective courses or articles. The author has summarized some common methods and processes through practical use in his work. Although they cannot be called perfect, they are enough for readers to achieve "basic learning" from scratch.

Friends who have not yet had an in-depth understanding of "data-driven growth" will definitely have this question: What is this thing? What's the use? How to use it? Section 1 of this article, "Basic Understanding", answers the questions of "why", "what" and "what's the use" through 4 sections.

Sections 2, 3, and 4 answer the question of “how to use it”. Section 2 introduces three usage examples, which makes it easier for readers to understand the subsequent method system. Section 3 introduces the method embodied in the examples in Section 2. Section 4 summarizes the general steps for building a data-driven growth model. The full text structure is shown in Figure 2.

Figure 2. Full text structure

To learn "data-driven growth", you must first have a correct macro understanding, and then learn the specific usage based on the macro knowledge framework. If readers already have a systematic understanding of growth, they can skip the first section which introduces the macro-understanding.

“Data-driven growth” sounds very profound, but in fact, the core content does not contain much. The reason why many famous experts use a whole book to talk about this skill is probably because publishing houses do not allow them to write only 10-20 pages in their books. The author is confident that after reading this article, you will have a comprehensive and systematic understanding of the core concepts and techniques of "data-driven growth" and will be able to start trying to use this skill at work.

1. Basic understanding

Correctly understanding growth

What is “growth”? It is generally believed that growth means increasing DAU, PV, and UV, and the best way to do this is to attract more traffic . However, the fact is: an increase in DAU/PV/UV that only “ attracts new customers ” but not “ retention ” is not growth!

This is like "drawing water from a bamboo basket". It seems that the water in the basket is increasing, but that is because the faucet is turned on too high. But the problem is: traffic costs money, and users leave before they click on the ad or make a purchase, so you can’t even recover the cost of acquiring the customer, let alone the word-of-mouth effect. Moreover, once your product makes users "disgusted" once, unless something unexpected happens, the users will not come back again. Ignoring retention can be said to be an approach of overdrawing the future! Figure 3 shows the suicidal attitude of “only attracting new customers but not retaining existing ones”.

Figure 3. The suicidal attitude of “attracting new customers but not retaining them”

PS: Some entrepreneurs use the method shown in Figure 3 to defraud investment , commonly known as the "To VC model."

Reconstruct your "data consciousness"

So how do we define growth? Personally, I think that indicators such as DAU and UV are “vanity indicators” and it is easy to go astray if you focus on them. Currently, the best explanation for "growth" is the " AARRR " model, which is also called the "Pirate Model" in some places, as shown in Figure 4.

Figure 4. AARRR model

① Obtain. It means obtaining “new visiting users” for the product from search engines , application markets and other channels . The goals of improvement should be: channel quality, quantity, proportion of new users, etc.

②Activation. Complete all pre-operations required to "experience the full product", such as registration, purchase, etc., and transform from a "visiting new user" to a "using user".

③Retention. The user recognizes the value that the product brings to him and continues to use the product. From "using user" to "active user".

④Cash out. Recover customer acquisition costs and make profits through methods such as click-through advertising, traffic sales, and service payments. The goals of improvement are: paid conversion rate , average order value, etc.

⑤Recommendation. Users are very satisfied with the value of the product and recommend it to others. Transformed from an "active user" to a "fan user".

These five core indicators together constitute growth. The five indicators have different emphases in different stages of the product life cycle . The exploration period focuses more on "activation" and "retention", the growth period focuses more on "acquisition" and "recommendation", and the stable period focuses more on "monetization". As shown in Figure 5.

Figure 5. Growth focus at each stage of the product life cycle

Whether you are a product manager or a product operator , the ultimate goal of everything you do must be growth. Therefore, everything we do must be aimed at improving one or more of these five indicators, and the corresponding data analysis should also revolve around these five aspects.

What data can bring to growth

" Conversion funnel" and "retention chart (table)" are two indispensable basic tools for analyzing growth data, and can be applied to every stage of the A AR RR model. Specifically, the "conversion funnel" can be used to measure channel quality, activation conversion rate, paid conversion rate, and recommendation conversion rate, and the "retention chart (table)" can be used to measure daily/weekly/monthly retention rate . As shown in Figure 6.

These two basic tools, combined with the "user segmentation" and "user detailed investigation" tools mentioned in the next section, allow us to discover the room for improvement and methods for improvement at each step in AARRR through data. This is the value that data brings to growth.

Figure 6. Use of data tools in the AARRR model

What tools are needed for data-driven growth

To do your work well, you must first sharpen your tools. Data-driven growth requires tools with specific functions. As can be seen from the previous section, the most commonly used data tools are the following five:

①Conversion funnel. As shown in Figure 7. It is used to quantify the conversion/loss of each step in a set of user operations for a certain function, as well as the usage rate of each function within the product.

②Retain the chart (table). As shown in Figure 8. It is used to analyze data such as 7-day retention, weekly retention after 1 month, and monthly retention after 1 year. It can be used to find room for improvement and methods for improving retention rate, verify the correctness of product optimization direction, etc.

③User grouping. Filter user groups through user behavior to achieve the purpose of marking important functions.

④Users check carefully. It can view all clicks and page browsing behaviors of a certain user, which is a powerful tool for conducting qualitative research.

⑤Source management. It is used to mark the source of users, and then analyze traffic quality indicators such as conversion rate, retention rate, and proportion of new users of each channel.

Figure 7. Conversion funnel diagram (Image source: GrowingIO)

Figure 8. Retention graph (Source: GrowingIO)

Among them, the first three functions are particularly important and indispensable. Figure 9 is the function menu of GrowingIO, Zhuge IO , and Sensors Data. It can be seen that each tool has these three core functions.

Figure 9. Main panels of GrowingIO, ZhugeIO, and Sensors Data

Whether one can skillfully use the three functions of "conversion funnel", "retention chart (table)" and "user segmentation" is an important criterion for measuring whether a product person has the basic skills of "data-driven growth". If you don’t have such a tool at hand, you can also use other alternatives, such as asking technical colleagues to guide the data or writing scripts yourself, but the efficiency will be much lower.

2. Actual combat cases

This section introduces three use cases. With cases as a foundation, readers can better understand the techniques and processes introduced later. The cases used in this section are extracted from Lessons 2, 4, and 14 of the GrowingIO open course. (I originally planned to use the author’s work background as a case study, but considering the issue of commercial confidentiality, I finally decided to use examples that are publicly available in society)

Case 1

Taking a music APP as an example, as shown on the left side of Figure 10, the retention of users who clicked "Like" more than 3 times within a period of time is shown by the red line, and the blue line represents the total users. It can be seen that the retention rate of users who click "Like" more than 3 times is higher than that of the overall users.

And then compare the retention difference between users who clicked “Like” more than 3 times and those who clicked less than 3 times? As shown in the right side of Figure 10, the bottom green line is the retention curve of users who clicked "Like" less than 3 times. It can be clearly seen that the retention rate of people who click "Like" less than 3 times is lower than that of the overall users.

The role of retention analysis is to guide how to optimize the product. Since I have found through the data that the retention rate will be high if the number of clicks on "Like" is greater than 3, then we can make a hypothesis: if users can be encouraged to click "Like" earlier, more customers will be retained.

Similarly, if users join an interest community, we can also see that their retention rate is improved relative to overall customers. Furthermore, if the user clicks "Like" more than 3 times and joins an interest community, the retention rate is higher than that of just clicking "Like" more than 3 times or just joining an interest community.

Figure 10. Retention chart of a music APP

Case 2

Taking an online travel website as an example, it is necessary to improve the conversion rate of the payment page, so a user who has reached the payment page but has not completed the payment is selected, and the "User Detailed Check" function is used to observe the user's behavior trajectory on the payment page in detail.

As shown in Figure 11 below, the leftmost figure shows the user's actions when he first entered the platform. The customer opened the page, browsed the travel product page, clicked to buy, and submitted the payment page, but exited directly without confirming the payment. The second time, the user came back, browsed travel products, selected another product, submitted the payment, and then exited from the payment page again. The user came in for the third time, browsed another travel product, submitted the payment, but ultimately did not complete the payment. This time the user completely exited the APP.

Through "user detailed inspection", it was found that users exited the payment page every time and then re-selected new travel products. Combined with the understanding of the business, the following assumptions are established: Customer selection of travel products is an iterative process, including travel time, hotel suites, transportation arrangements, attractions to visit, and so on. Customers are likely to change their choices again after submitting an order. If the payment page of the order cannot modify the order content or return to the previous page to modify the order, the user will eventually give up payment or exit directly, resulting in a low payment conversion rate.

Based on the above behavior, we can establish the hypothesis that "the payment page lacks product comparison function" and then verify or falsify this hypothesis. Specifically, it can be verified through interviews with seed users . If the development cost is very low, it can also be verified through online A/B testing.

Figure 11. Using User Scrutiny to find design flaws

Case 3

The conversion funnel of a certain function can be analyzed from the perspectives of region (analyzing the conversion situation in each region), platform (iOS, Android, web, etc.), behavior (receiving coupons, following a product, etc.), etc., as shown in Figure 12. By comparing the differences in conversion rates in various dimensions, you can find a lot of room for optimization. So you can take measures like: increase the delivery in certain regions or channels, increase the exposure of certain features, and distribute coupons to more people.

Figure 12. Using dimension comparison to discover room for conversion improvement

3. Introduction of skills

Tip 1: Find the magic number

The method in Example 1 is a typical "magic number". First, let's clarify the concept of magic number.

When new users use a certain function at a certain frequency over a certain period of time, they are more likely to stay and become loyal users. These magical numbers that can greatly improve user retention are called magic numbers.

This method originated from Internet companies in Silicon Valley. For example, Twitter found that if new users followed 30 friends within 30 days, they would easily continue to be active on the platform, otherwise the risk of churn would be very high; LinkedIn found that if new users added 5 contacts within a week, their retention rate and usage frequency would increase 3-5 times; Dropbox found that if new users used the Dropbox folder once, the possibility of becoming loyal users would be greatly increased.

However, it is not enough to just know the fact of the "magic number". We should also know the reason behind this fact. The benefits of understanding the principles are: ① If your product has many functional points, it will take a lot of energy to test them one by one. Understanding the principles can make the experiments more targeted. ② It is easy to confuse the causal relationship between usage behavior and retention improvement: Is the use of this function the "cause" of retention improvement, or the "result" of retention improvement?

The reason why the "magic number" phenomenon exists is that certain functions in the product can allow users to discover the value of the product more quickly. If the user value achieved by a product is 90 points, the user may only discover 60 points of it when they arrive at the product, and the remaining 30 points need to be gradually discovered by the user in the process of using the product. However, users’ patience is very limited. If you fail to make users realize the value of your product before their patience runs out, then you cannot prevent them from leaving. In Example 1, the two functions of "Like" and "Interest Community" allow users to discover the user value of the music APP more quickly.

Finding the "magic number" is equivalent to opening up the "Ren and Du meridians" of retention, achieving twice the result with half the effort. The same method is actually widely used in the products we use every day. For example, JD.com will issue points as rewards to users who shop more than three days a month, and Boss Direct Hire has placed the "message" entrance in the most prominent position within the app. As shown in Figure 13. This is all done to allow the user to trigger the "magic number".

Figure 13. Magic numbers in JD.com and Boss Zhipin

Tip 2: Find design flaws

The design flaws here include: bugs that testing colleagues failed to find and designs that make users uncomfortable. Discovering design flaws is a qualitative and then quantitative process, and its goal is usually to improve the conversion rate of a certain step in the conversion funnel.

A. In the qualitative phase, the goal is to find cases where user behavior is abnormal. First, identify which step of the conversion funnel you want to improve, and use the “user segmentation” function to mark the users who leave this step. Then, use the "user detailed check" function to find out what happened before and after the user left. Usually, you will find some phenomena such as "the user did not use it according to the original design intention" and "the user encountered a functional bug".

For example, in Example 2, the goal is to increase the conversion rate of the "payment" step. Through "user segmentation", users who "reach the payment page but do not confirm payment" are marked out, and then the behavior of these users is analyzed through "user detailed investigation". Finally, it is found that many of the users who "did not complete the payment" "return to reselect products" on the payment page.

B. In the quantitative phase, the goal is to estimate the number and percentage of users affected by the problems found in the qualitative phase. Because the cases discovered in the qualitative stage may be individual problems encountered by a single user or common problems encountered by a group of users, we need to consider the number of people affected and their proportion to evaluate whether optimization is needed here? How high is the priority? When performing quantitative calculations, first use “user segmentation” to define the users who need quantitative analysis, and then use the “conversion funnel” to evaluate the impact.

For example, in Example 2, we first use user segmentation to define users with the behavioral characteristic of "returning to the payment page and then reselecting products". This way we can know the number of users affected by this problem every day/week. Then put the users in this segment into the funnel to see what percentage of these users fail to convert every day/week at the “payment” step.

Tip 3: Estimate conversion upside

When improving conversion rate, we often encounter this problem: Is my conversion rate high or low? How much room for improvement is there? It is basically impossible for you to use the data of competitors as a reference, and there is no need to do so because your own conversion data contains a lot of information. For example, in Example 3, you can try to distribute coupons on a small scale, and then analyze how much the purchase conversion rate of users who received coupons increased compared to users who did not receive coupons. This way you will know how much you can increase your overall conversion rate through this strategy. Issuing coupons to new users to promote purchase conversions is very common in e-commerce and Internet finance . Figure 14 shows how Kaola.com and Aiqianjin.com issue coupons to new users.

Figure 14. Kaola.com and iQianjin distribute coupons to new users

4. General steps to build a data growth model

In the AARRR model, the most important things to pay attention to are "activation" and "retention". Although "acquisition" is also very important, "data-driven growth" only provides a means to analyze channel quality from the perspective of "activation rate" and "retention rate". Its core delivery strategy has not changed much compared to before; and the improvement methods of "monetization" and "recommendation" are similar to "activation" and will not be discussed separately.

Establishing a growth model consists of four steps: ① Define growth indicators. ②Find the magic number. ③Optimize core functions. ④Increase the number of people covered by core functions.

Step 1: Define overall product activation and retention metrics

The definitions of activation and retention should be clearly defined based on product characteristics. For example, e-commerce companies usually use "complete purchase" as the activation mark instead of just "complete registration". Similarly, product personnel also need to think clearly about whether to use "opening the APP" as the retention indicator or "browsing the product details page" as the retention indicator?

Step 2: Find the “magic number” of core indicators

After clarifying the definition of "retention", use the method in "Tip 1" to find the "magic number" and the product features that carry the "magic number".

Step 3: Optimize core functions

We should use limited resources to prioritize the core functions of products, which are the necessary functions in the "activation" process and the functions that carry the "magic number". Because if the "activation" related functions are not easy to use, users will leave directly, and if the functions that carry the "magic number" are not easy to use, the chance of the "magic number" being triggered will be greatly reduced.

For example, in Boss Direct Recruitment, "Register" and "Post Resume" are necessary functions for the activation process, while "IM Chat" and "Resume Submission" are likely to be functions that carry the "magic number".

For “core features”, follow “techniques 2” and “techniques 3” to maximize their conversion rates and make them more useful.

Step 4: Increase the number of people covered by the “magic number”

If you opened a distinctive restaurant, you would definitely try your best to serve your customers the best signature dishes. Because after tasting these signature dishes, customers are more likely to recognize the restaurant's cooking skills. Then customers will be more likely to come back to your restaurant next weekend. On the contrary, if a customer does not taste the signature dish during his first visit, he will mistakenly think that the taste of your restaurant is mediocre and will not come back a second time. The “magic number” is actually the product’s “signature dish”.

After optimizing the core functions, you need to do everything you can to get users to trigger the "magic number". This can be achieved through "user tasks", "material incentives", "pop-up prompts", " push push", "putting the function that carries the magic number in the most conspicuous position" and so on.

Sometimes, there is more than one magic number in your product. In order to maximize the potential for user retention, you need to test whether there is a "superposition effect" between different magic numbers. If there is an overlay effect, multiple magic numbers should be used in combination; if there is no overlay effect, the one with low implementation cost should be used as the preferred solution, and the one with high implementation cost should be used as the backup solution when the preferred solution is not triggered. For example, in Case 1, “click to like > 3 times” and “join the interest community” are two sets of magic numbers with superimposed effects, which can guide users to “click to like > 3 times” and “join the interest community” at the same time.

5. Postscript

"Data-driven growth" is an essential skill for product managers and operations

Driving product growth through data is a necessary skill for every PM, and it is best not to let a "data analyst" do it, because there is an important prerequisite for doing all of this - being very, very familiar with the product and users. For example, in Example 1, you need to be very clear about the user value of the product in order to specifically discover the functions that carry the magic number. Establishing reasonable hypotheses based on behavioral data also requires being very familiar with the users. For example, in Example 2, you need to know which operation entries are available for "Add to Cart", otherwise the conversion funnel data will be incomplete.

About Data-Driven Operations

Of course, the growth process cannot be achieved without the participation of operational work. However, the author has never been in charge of operations and has no say in operations. Therefore, this article only briefly introduces the common methods of "data-driven operations".

①Channel acquisition of new customers. Design delivery strategies based on the dimensions of quality, quantity, and price, and analyze the quality of the channel from the conversion funnel, retention chart (table), and percentage of new users.

②Precise operation. Users are classified according to their behavior, and then refined operations are carried out based on the characteristics of different groups. For example, users' behaviors on the forum include: visiting, browsing posts, replying, commenting, posting, forwarding, sharing, etc. We use "user grouping" to divide users into 4 categories: A browsing category, B commenting category, C dissemination category and D content production category, and then push different messages to different types of users. For example, we can infer which users have a higher possibility of paid monetization through their usage behavior and personal attribute information, and then give these users limited coupons.

③Activity operation . The effectiveness analysis of an activity should be linked to at least one of the AARRR models, rather than just looking at: how many people participated in the activity and how many UVs the activity brought to a certain function. For example, if the goal of Activity A is to improve retention, then we should also analyze how much the retention rate of users who participated in the activity increased compared to users who did not participate in the activity, whether the number of active days per week of these users increased before and after participating in the activity, etc.

The Limits of Data-Driven Growth

Nothing is possible without data, but data is not omnipotent!

For example, how did we get the first version of the “Collections” and “Interest Communities” functions in Example 1? Obviously it is not through data, because there was no data available before the first version; for example, why do some users go to the payment page, take a look, and leave without clicking anything? At this time, the users leave no data that can be analyzed.

This illustrates two limitations of data-driven approaches: 1. Data can hardly inspire major innovations. ②For some issues, there is no data available for analysis at all.

It can be seen that in addition to data-driven, product optimization must also rely on other driving forces. Regarding other forces that drive product growth optimization, I will write an article later titled "4 Product Evolution Driving Forces You Must Know" to introduce them, and it will be available to you soon.

If you have read this far, I believe you must be a strong-willed person. There is nothing in the world that a person with strong willpower cannot do, let alone mastering a not-so-complicated skill. What you need to do next is to continue to use this skill at work. Come on!

The author of this article @刘鑫洋 is compiled and published by (Qinggua Media). Please indicate the author information and source when reprinting!

Product promotion services: APP promotion services, advertising platform, Longyou Games