Growth Tips: Increase Transaction Rate by Over 900%

Let’s get into today’s topic, a sigh from the bottom of the valley.

Look at this curve. The first vertical line is in February 2017. We found that users have been lost and we are gradually losing competitiveness. Now we come to the second curve, April 2018. What does this curve mean? It was after I joined Bitauto that I successfully brought Bitauto to the bottom through a series of reverse operations. Zhao Benshan has a skit in which he says: "Wherever I go, the environment is bad." Is it the environment that is not good, or is it me who is not good enough?

I once suspected that I was the one who influenced the overall environment. Just kidding, actually during this period, we have tried a series of reflections, the main contents are: Why are the old Internet brands old? Is the technology old? Frequent crashes, giving users a bad experience? Or are our products old? Old content? Is there still a problem with the operating model? No growth mindset?

Everyone here is a new generation operator and may not be familiar with some old-fashioned operations and ideas. The automotive industry has been a part of the entire development history of the Internet. The previous generation did not pay enough attention to the concept of data driving business, or did not have this kind of thinking. How can we change this situation? From April to November 2018, we have been thinking about these issues for more than half a year.

After sighing, we did a very important thing at the bottom. We did not blindly spend money just because we lost competitiveness. While reflecting, we carried out a large number of tests. There is a saying that goes, "If the direction is wrong, efficient execution will lead to disaster." In order to avoid disaster, what we need to do is to use testing to know which path we should take. Ultimately, it was confirmed that we chose the right path during this period, which led to the current growth.

Today I also brought you three testing methods. Teacher Huang said in the book that growth is divided into three major schools. The first is the marketing school, which is the front-end traffic. The second is the experimental growth group, which continuously grows through experiments. The third is the technology mentioned by growth hackers. During that period, we have been testing in these three areas. Due to time constraints today, I will only bring you the first two. One is the traffic test, and the other is the product test.

1. Channel Evaluation Model

The boss often asks us which channels are good and which channels are bad. To answer these questions, can we only look at retention? Or just look at the commercial GMV? Actually, you have to look at both. How to analyze specifically? The channel evaluation model - AHP standardized model is used. The principle is very simple, first standardize, then multiply by the weight. In the current market, evaluation models are roughly divided into two categories: regression model and standardization model. Why choose a standardized model? Because it can support our business faster, whether focusing on quality or commercial value, we can change the weight at any time during different business strategy periods.

Channel quality assessment model flow chart

As shown above, the modeling process is divided into seven steps:

(1) Establishment of the model. Choosing a standardized score and multiplying it by a weight is a weighted standardization model.

(2) Storage standards. What kind of channels do we choose to enter the model, all channels or paid channels? Here we choose the payment channel.

(3) Algorithm establishment. There are many types of standardization, such as Z standardization and maximum-minimum standardization. Here we choose maximum-minimum standardization.

(4) Weight calculation. AHP is used here.

(5) Indicator screening. Indicator screening is very important for the model. Let us give an example here. The high school entrance examination has three subjects: Chinese, mathematics and English. It is very reasonable to use the total score of these three subjects to evaluate whether a student has studied well or not, because these three subjects are independent of each other and have very low correlation. However, if we use the total statistical scores of advanced mathematics, linear algebra and mathematical analysis to evaluate a student, it can only reflect how good the student is at mathematics, not the overall score. The same is true for the model. There will be some indicators with very strong correlation in the model, which requires us to eliminate or integrate these indicators to ensure that the indicators are independent of each other and can represent the business.

(6) BI display.

(7) Operational strategy.

Model skeleton diagram

AHP is a method of transforming qualitative analysis into quantitative analysis. Suppose we have five indicators, and we compare quantity and behavior. If they are equally important, it is 1; if quantity is slightly more important than business, it is 2; and if quantity is obviously more important than cost, it is 5. Fill these numbers into the yellow matrix, and the corresponding weights will appear. We multiply the standardized score by the weight based on the weight, and that is the entire model. Finally, we fill in the skeleton and the model is completed.

BI display and strategy map

Next is BI presentation and strategy. Why is BI presentation so important? Generally, bosses spend less than 10% of their time making a decision, but contribute 90% of the value. We spend 90% of our time on data processing and model building at the underlying level, but only contribute 10% of the value. BI display plays a connecting role, allowing the boss to understand the business status and make decisions in the shortest time. This is why BI presentation is very important.

Combined with the first channel rating details dashboard, fill in all the model skeletons, which include weights, absolute values ​​of specific data, and total scores. The boss can find many problems and channel characteristics through this dashboard, as shown in the following figure:

Algorithm Page Diagram

For example, information flow channels have been a bonus in the past year, and in our automotive industry, it is an absolute bonus. Because information flow channels led by short videos have changed the practices of app stores. Through a short video, the product form can be fully introduced in 15-20 seconds, which is something that app stores do not have. This is also why it has become a bonus in the automotive field.

Moreover, we found that as long as the information flow channels choose the right materials, the commercial value they bring to us will be higher than other types of channels. Let’s talk about some typical app stores’ attributes. Its user stickiness and retention are very high, and the user's UGC behavior will be better. This is the experience we can summarize through the channel dashboard, and we can formulate reasonable delivery strategies based on these conclusions.

What do the natural and paid portraits on the right mean? It represents the comparison between performance delivery and brand delivery. I don’t know if you have ever seen the advertisement about buying a car without losing out. After the advertisement is placed, how to check the effect is to use this radar chart.

Comparison chart of paid and organic users

Red represents brand investment, which is natural volume. The blue color represents the effectiveness of delivery, which is equivalent to buying volume. This chart can clearly show the differences between these two groups of people and carry out refined operations based on these two groups of people.

The overall channel trend chart on the left. The total channel score is used to monitor whether the channels are optimally matched. Channel optimization can be achieved by adjusting the materials and the ratio of different channels, such as the ratio of Android and IOS, the ratio of information flow and application stores, etc.

Channel overall trend chart

This is the end of the channel evaluation model. If you want to learn more about this model, you can check out the article "How to use less money to bring better quality volume?" | Channel quality assessment model》.

2. Growth Model

The experimental growth model is the most mentioned in "Growth Hacker". The experimental growth model is divided into four steps: discovering problems, proposing ideas, testing the matrix, and reviewing and analyzing.

Experimental Growth Flowchart

Next, we need to roll out these four steps, accumulate small successes through each experiment, and eventually achieve big victories. This is the most basic principle of growth experiments.

Find the problem

The first step is to identify the problem. What problem did we find in the picture of the valley bottom we just saw? We found that users were churned, so what we need to do is to increase users . So how to achieve user growth? If retention grows, we will be able to achieve user growth. So how do we achieve retention growth? What is our growth method and growth formula? This brings us to our second step: coming up with ideas.

Propose an idea

The second step is to come up with ideas. There are five major methods on the market that can improve retention.

The first is the channel. Are our delivery materials reasonable? Are the users we acquire the same as our target users? The channel evaluation model just mentioned can screen channel traffic, control user quality through traffic entrances, and improve retention.

Second, we test products constantly, for example, to allow users to access the aha moment, which is our core function, more quickly; the product's icon shape, color, size, position, etc.; these can all be tested experimentally to determine whether the product is optimized.

The third is activities, which keep users logged in continuously through tasks or activities, which is actually equivalent to spending money to buy retention.

The fourth is service. If you want to increase retention, you have to improve the service. This is the first priority. For the Yiche APP, our services are information, videos, articles, reviews, Q&A, etc. In addition, we also have our car wash business, and services for car owners such as asking for the lowest price are all our services.

The fifth is through unconventional means of operation, such as applying red dots. For example, an unread WeChat message will display a small red circle in the upper right corner of the APP. There will be a circle 1 or circle 2. Users with obsessive-compulsive disorder will want to click on it, and clicking on it means retention. However, this type of retention is not recommended for everyone.

Apply Red Dot

Because the users who click in this way have lower penetration rate and later performance than the first four options, we recommend that you choose from the first four methods to improve retention. Today I will introduce to you the second method, how to grow from the product side through experimental growth models.

Test Matrix

The third step is to test the matrix. First, a growth formula will be developed for retention, as follows:

Retention gap value = reached retention - unreached retention

Penetration rate = number of people reached/DAU

Multiplying the two and taking the absolute value is our growth formula. It is easy to understand that the difference between use and non-use multiplied by the radiation range, the higher the value, the more significant the retention growth effect will be when we change this function.

After determining the growth formula, the next step is to screen the experimental groups. Taking our team as an example, our team includes technology, products, and data, and the maximum number of experimental groups we can withstand in a week is 10. Then we will sort the growth coefficients of all product improvement ideas in descending order and select the top 10 for experiments.

The last one is the growth method. According to the formula, if you want to grow, you need to make changes in two directions, as shown below:

The first is to start with increasing the retention gap value, as shown in the following figure:

Test Matrix

It can be seen that in the previous version, the retention rate of users who have not registered is 41%, and that of users who have registered is 39%. The retention gap value is -2%. If this is swapped and users are not forced to register, their retention rate can grow positively. This is the first point, the increase in the retention gap value.

The second is penetration rate. How to understand permeability? Suppose this page is a third-level page, and the shortest path for users to reach this page is three steps, 810,000 out of 1 million people can access this page, and its penetration rate is 81%. Can we adjust the entry depth of this page so that the second page is directly presented to users after opening it? This will increase the penetration rate and allow pages with high retention to penetrate more users, thereby achieving growth. Testing is about achieving growth through these two ways of thinking.

Experimental review chart

The fourth step is to review the experiment, which is the most important part.

We identify problems, come up with ideas, and conduct tests, and this is where the success of the tests is determined. Usually there are three results when reviewing a market: positive, negative, and no result.

These three results are all very valuable information. Even if there are no results, don’t underestimate this conclusion. At least you know that such operations have no impact on growth, which is also a very important conclusion. Here I have picked out two classic cases from the testing of Bitauto.com. The first one is positive, as shown below:

Classic Case 1

That is, Group A and Group B are tested. Group A is the vehicle model page, which is also the tool attribute page. Everyone will enter this page when choosing a car. Group B is the information flow page, which will recommend some related articles to users. Through these two pages, we found that the information flow page is not as effective as the car model page. The car model page has at least one percentage point higher retention rate and transaction rate than the information page.

After analysis, we know that the car model page has tool attributes and is the aha moment (core function) of our product. Users come here more to buy, view, and use cars. Therefore, the closer users are to the aha moment (core function), the higher the retention rate and transaction will be. This is a positive conclusion.

Next, let’s talk about the opposite conclusion. Whether registered or not, Yiche.com still insists on being registered. Requires user registration. Why? Although sampling surveys have shown that the retention rate of unregistered users is higher, it is not difficult to imagine that if they register, many users will be waiting. Just by leaving their mobile phone numbers or WeChat accounts, many users will jump away directly, and there is a high probability that they will not come back, which is a loss for us. So why do we persist?

This is because we found that although the retention rate has decreased, the conversion rate has increased, and the value of a registered user in his or her lifetime is 20 times that of an unregistered user. This also teaches us a lesson. Sometimes, don’t blindly pursue vanity indicators and lose core indicators. To improve retention, you need to consider the overall situation. This is also an important ability of a data analyst - a global perspective.

Classic Case 2

There are two major factions in the market: registration and non-registration. Tencent News is the leader. It does not require users to register and you can use it at will. Another one is Bilibili. Bilibili not only asks you to register, but also gives you a hundred questions. However, we have found that the stickiness and UGC behavior of Bilibili users are very good. After a series of data verifications, Yiche finally achieved data-driven business and chose to retain registration as a key behavior.

Regarding this small registration threshold, I have a comment. If you play games, you will know that this year's TGA2019 Game of the Year is "Sekiro". The producer of "Sekiro", Hidetaka Miyazaki, once said: Many things in society now have become fast-food in order to cater to users. This is actually a kind of degeneration. Everything has a threshold, even a game.

I would also like to take this opportunity to say that APPs also need to have certain thresholds, but the threshold should not be set too high. A slight threshold will be enough. In this way, users can cherish our APPs more, and at the same time we can provide users with more accurate and high-quality services. This is a two-way choice. These two examples are also very valuable information we obtained during the experimental testing phase.

Finally, I want to tell everyone that no matter where your APP is in its life cycle, whether it is in the trough period, growth period, or stable period, don’t forget to experiment, change yourself, experiment again, and grow again. Only crazy testing can trigger user growth.

This is what I want to share with you. I wish you all crazy testing and exponential growth in 2020.

Author: Jiang Di

Source: Jiang Di