Case Review | QQ Browser News New User Retention Growth Methodology

By reviewing a growth case I did last year - the growth of new user retention of QQ Browser Information, I summarized a growth methodology that is more suitable for local optimization of small projects.

Based on the established goals, we find potential opportunities through data analysis, design product solutions, and use rapid experiments and rapid iterations to come up with a relatively complete solution, which is then systematically promoted to a wider range of application scenarios.

This article will be divided into two parts:

1. QQ Browser News New User Retention Growth Case

2. Methodological summary and suggestions

1. QQ Browser News New User Retention Growth Case

1. Clarify growth goals

Information business is one of the core businesses of QQ Browser (hereinafter referred to as QB). With the decline of traffic dividends and the slowdown in the number of new QB users, the resources of new users are becoming increasingly scarce, and it is particularly important to attract new users.

In the past, QB had two options for new users:

1. Enter the novice protection period. Minimize disruption to new users;

2. Guide login. User assets are deposited by logging in.

Experiments have shown that the retention effect of login is very good. From the perspective of QB Information Business, how to improve the retention of new users based on previous plans is a proposition before me.

2. Data First

Many valuable opportunities can be derived from the analysis of data, which is our compass.

After exhaustively enumerating all the data indicators related to retention, we found some core influencing indicators for correlation analysis. We found that the number of new users’ points of interest has a significant positive correlation with retention, and there is an obvious inflection point.

Note: Points of interest refer to the machine’s understanding of user content consumption. Points of interest will only increase and not decrease.

The data has been anonymized

It can be seen that the more points of interest a user accumulates in QB, the higher the user's retention will be; when the number of points of interest is less than 50, the retention rate increases by about 3% for every 10 points of interest added by the user; and when the number of points of interest reaches 50, the retention rate will also be stable.

This logic is easy to understand. The more interests a user has, the more the machine can recommend content that the user likes, and the user will naturally like to come here to watch content more.

Therefore, we decided to focus on the interests of new users, and our proposition turned into how to increase the interests of new users.

3. Brain Explosion-Related Methods

With the help of some divergent tools, we brainstormed many methods. From the perspective of achieving costs and benefits, we finally extracted two means: importing external points of interest and users actively selecting interest preferences.

But the sad thing is that these two methods have been done by others before! The external introduction of interests is already online, and the effect is very good; the active selection of interest preferences was abandoned due to poor results of previous experiments. Sure enough, what I could think of was also thought of by others, and we were stuck here for a while.

Finally, I talked to a product that had previously actively selected interest points and found that there were two core problems with the interest selection solution:

1. The overall distribution funnel is very large, resulting in unclear experimental results;

2. There is no immediate benefit feedback after the user completes the selection, the consumption is not handled well, and the user experience is interrupted.

If I can solve these two problems, will there be a turning point? With this idea in mind, I decided to restart my interest and choose this option!

4. Propose new product solutions

I carefully studied the overall link for new user onboarding. In the new solution, I need to ensure two points: reducing the downward funnel and ensuring that the recommendation side responds to the user's interest selection results in real time.

The first point is to avoid the experimental effect being unclear due to a large funnel; the second point is to avoid the lack of corresponding content after the user completes the selection. As a result, it is impossible to perceive clear benefits of interest selection.

Based on the above two points, I proposed a new solution: 1. Choose the splash screen and terminal based on interest. The terminal's delivery efficiency is more than twice that of the background delivery. 2. After the user completes the selection of interests, the recommendation side needs to respond to these interests in real time, and perform weighted and independent recall to ensure that the user can perceive the recommended content in real time.

Overall Logic

This interest selection solution opens up the link between the function side and the recommendation side, and can respond to users' interest point selections in real time.

5. Experimental verification of feasibility

The above solutions are limited to speculation and analysis on the product side. Whether they are effective still needs to be verified by experiments. In order to verify the overall effectiveness of the product solution, we quickly produced an MVP and conducted a round of functional testing.

01 Functional Testing

We divide new users into three groups, using interest selection, login guidance, and non-interference to handle them respectively. Sure enough, fortunately, the retention data and related consumption data of interest selection performed significantly better than the other two groups.

02 Functional Testing

So far, we have verified the effectiveness of the interest selection function and we can increase investment to optimize it. The next goal is to find ways to optimize the funnel, increase the completion rate of interest selection, and improve the effectiveness of the experiment.

03 Path Testing

Because people of different genders have very different interests and preferences, and it is usually difficult to obtain the gender characteristics of new users, the initial design of the plan is to select gender first, and then select interest preferences in the second step. This raises a difficult question.

If gender selection is not performed, the lack of an important gender feature recommendation may affect the recommendation effect, and the corresponding points of interest need to include points of interest for both men and women, so there will be many options, which may affect user choices; but if gender selection is entered, the overall selection completion rate will be greatly reduced with one more page.

We set up another set of experiments to test the path for new users. The same new users entered the paths with gender selection and without gender selection respectively, and observed the selection completion rate, retention rate and subsequent consumption performance of the two groups of experiments. It was found that the relevant data of path B without gender selection was significantly better than the previous plan.

Path Testing

So we determined the optimal path for interest selection. The next step is fine-tuning the UI and copywriting.

04 UI ​​and copywriting testing

Different UI styles and copywriting have a great impact on the selection completion rate, so we set up four groups of experiments for the overall sorting UI, color UI, and key copywriting, as follows.

UI and copywriting testing

After layers of experiments, we finally decided that Plan 1 was the best option. The final data showed that after several rounds of optimization, the completion rate of the option was 20% higher than that of the previous option.

6. Systematic volume expansion

Finally, we fully launched the optimized solution. The data results after launch were better than those in the experimental period. The relevant retention increased by about 10%, which was higher than the initial expectations, and achieved the initial new user retention growth goal.

After verifying the effectiveness of this approach, we quickly extended this solution to more scenarios, adopted more diverse methods to collect users’ interests, and promoted it in multiple products.

Systematic promotion attempt

2. Methodological Summary and Suggestions

How to achieve growth in the local battlefield?

Use data as a compass, iterate quickly through experiments, and once a complete solution is formed, promote it systematically.

1. Lock in core indicators. There can only be one core indicator, and this core indicator must be able to directly reflect the goal you want to achieve;

2. Data comes first. Through data analysis, we can transform existing problems and find new ideas and opportunities.

3. Brain explosion methods and measures. If you can think of many methods that others have not tried, then try them one by one; if you can't find any, don't give up those methods that have not been successful before.

4. Propose your own product solution based on the goals. Product solutions need to return to the goal and the users themselves.

5. Conduct experiments. You can make priority decisions based on the difficulty of implementation and the benefits. Functional testing can determine whether there is room for new functions and whether more investment is needed. Path testing can help you find the path with the lowest user cost. UI testing and copywriting testing can always bring you unexpected funnel improvements.

6. Online systematization. When a method is proven to be effective, it can often lead to many other priority methods. If it can be scaled up systematically, the impact of this plan can be maximized.

My final thought is that there are many discussions on growth methodologies in the market, but most of them are about the growth of local single projects. This case is also the case, and the methods used are only applicable to local battlefields.

I have always been interested in overall growth. System growth is the most technically demanding because you first need to understand how a system works as a whole, find the key elements and key links, and know how to leverage the key links and elements.

This case was a relatively successful growth case last year. I hope that this year I can create a case of a higher dimension.

Author: Lin MuJoin

Source: Lin Mu Join