I did it! I used data analysis to improve user conversion rate

This article teaches you how to use data to analyze and draw optimization conclusions. Through a series of ideas and case analyses, it helps you gain something from your work in data analysis.

"Data analysis must identify specific business optimization points" is the requirement of many companies for data analysts, and it is also a problem that makes many students headache.

How do you draw an optimization conclusion from data indicators? Today, I will systematically explain how to do it based on a specific problem scenario.

Problem scenario:

An online education institution holds a free live broadcast every week, and all users can watch the live broadcast after making an appointment. The business side hopes to increase the user payment rate through live broadcast.

But after a period of implementation, the business side began to wonder: Do the viewing rate and number of viewers after live broadcast registration have any effect on the payment rate? After all, live broadcasting also has costs. Why do I always feel that the conversion rate does not seem to increase after more broadcasts? ?

Q: How to analyze this problem? Where are the optimization points for live broadcast business?

1. Common Mistakes

Many students are used to using whatever fields are in the database, without distinguishing the scenarios or labeling them, so naturally they can’t analyze anything. For example, in this case, it is very likely that the original data record is a live broadcast named XXX, with XX people signing up and XX people watching, and that’s it. Without thinking deeply, it is likely that you will tend to:

• Take the total number of daily live broadcast viewers & the total number of daily sales, do a correlation analysis, and see what the correlation coefficient is
• Take the number of viewers of each live broadcast & the number of people who purchased after the live broadcast, calculate the live broadcast conversion rate, and then draw a line graph
• Take the people who watched the live broadcast and the people who signed up but didn’t watch the live broadcast, divide them into two groups, and calculate the purchase rate

This way, we can calculate three numbers, but when we draw conclusions, we are easily challenged by the business:

• I found that the correlation coefficient between live broadcast viewers and sales is 0.76——So what! So what???
• I found that the conversion rate of live broadcast has been declining in the past three weeks - nonsense! I knew it...
• I found that the purchase rate of those who watched the live broadcast was 5% higher - of course it was higher, so what?

This is where common doubts come from. The reason why these conclusions are nonsense is that after the business side has seen it, they really don’t know what to do. The optimization suggestions that the business side expects to hear are: Can live broadcasting still be done! If it can be done, how many shows should be done? What topics should be discussed? What links should be posted? If it can’t be done, how can I solve the sales problem! This is useful advice.

So, how to solve the problem?

1. Understand the business scenario

To find business optimization points, the first step is of course to return to the business itself. From the perspective of the education business itself: it is not a very appropriate behavior to have all users start live broadcasts in the same pot. This is because the needs of different users are completely different:

• New registered users: They are unfamiliar with educational institutions and courses and need to build trust
• Users who have paid once: If you ask people to buy a new course after just two lessons, no one will buy it.
• Users who have paid n times: Users have already achieved learning results and progress, and the next promotion is also an advancement course

In short, different people have naturally different educational needs.

When many students see this, they will instinctively say: The effect will be better if we conduct live broadcasts among different groups of people!

You will also be criticized for this, because understanding the business scenario is just the beginning.

2. Analyze business pain points

When making suggestions, avoid making suggestions that sound reasonable but are actually brainless: "The indicator is too low, we need to raise it" or "Mixed broadcasting is not good, we recommend separating them."

Why? Because people are not fools. If they see that the indicators are low, they will definitely think about raising them. If they have the energy to do dispersed sessions, they will definitely think about doing them separately. There are usually hidden reasons behind practices that violate common sense . We need to further sort out and find the business pain points first.

On the surface, the current problem is that the conversion effect brought by live streaming is not obvious.

Looking deeper, the possible problem is: live broadcasts are all mixed together and there is a lack of classification guidance.

Looking deeper, why is there a pot of stew? The hidden reason behind it is probably:

• Some topics are of interest to everyone, such as career development, basic skills, etc. There is no need to divide them.
• Live streaming also has costs, it takes time and content production. But the pace of acquiring new people is not fixed, so if you do it for new people, scheduling is very troublesome.
• The learning status of old users was not separately counted for colleagues organizing the live broadcast, resulting in the inability to understand how many people are at each learning progress.
• After splitting the crowd, some groups may have very small numbers and the conversion rate may not be enough to support a separate live broadcast.
• Even if the business is split, it may not necessarily improve the conversion rate. There is currently no data to prove this.

In short, the so-called splitting may only look good, but there are many practical complications.

However, these specific pain points are a treasure for data analysis. The more specific the problem is, the easier it is to draw a conclusion; the more vague the problem is, the harder it is to draw a conclusion . With specific pain points, you can see how to use data to solve the problem.

3. Inductive analysis logic

Business pain points may be scattered, and solving them with data requires analytical logic. The simplest way to build logic is to start from big to small, from rough to fine, first eliminate obvious problems, and then pursue details.

In this case, from a data perspective, the above business pain points can be summarized into three categories of problems:

1. Is it true that the conversion rate of existing live broadcasts is low? Is it limited to certain entities, or is it all low?
2. For existing users, are there inherent differences in conversion rates? Which ones can be broken through by live streaming and which ones cannot?
3. Are all existing products suitable for live streaming conversion? Are there different scenarios with different unit prices?

These three questions can directly lead to specific optimization suggestions.

But please note that these three problems may be intertwined. For example, if a live broadcast fails to bring goods, it may be that the live broadcast itself is not good, or that users have no demand, or that the products do not match. At this time, it is necessary to construct analysis logic.

Judging from the title, the business side did not dwell on users and products, but started with live broadcasting. Therefore, when constructing the analysis logic, we should also start with live broadcasting, and first eliminate the problem that the live broadcast itself is not well organized (as shown in the figure below).

Secondly, in educational products, live broadcast topics are naturally related to the products to be sold, but not necessarily to the viewing users. Especially for novice users, they often cannot tell what they really want to learn and just watch it casually. Therefore, the second level can be divided into users, distinguishing between new registered users and old users (as shown in the figure below).

Now that the analysis logic is built, you can fill in the data, but it is still recommended to do some preparatory work.

4. Prepare data

In order to describe the business situation, it is often necessary to use a large number of labels, and it is very likely that these labels have not been prepared in advance. Therefore, preparation is required.

For example, in this case:

• Live broadcast labels (learning topic, instructor level, applicable groups, difficulty)
• User tags (new user/old user, source channel of new user, old user)
• Product tags (suitable for groups, price, learning topics)

These all need to be prepared one by one so that there will be clues for subsequent analysis.

Note: It is very likely that the business side needs to analyze the conclusions urgently, and the previous infrastructure is very poor, so there is no time to label them one by one. At this time, you should remind the business side: without labeling, it is impossible to analyze the problem in depth. It is recommended to at least label some particularly important ones first, otherwise you will always cram at the last minute and never make progress.

5. Output analysis conclusions

With all the above preparations, the last step is to fill in the data, which is a natural thing. Moreover, this kind of analysis can help you find the most obvious problem points and put forward very detailed optimization suggestions (as shown in the figure below. Note that due to space limitations, the figure below does not fully display the entire deduction logic. Interested students can complete it by themselves).

When building the analysis logic, the situation corresponding to each type of user is actually a specific business optimization point, but the data is the final judge . The more situations occur, the more problems will be solved. Moreover, when two factors are intertwined, we can choose the main problem to solve based on the amount of data. This is exactly where data analysis is useful. Otherwise, there will be too many problems to solve.

2. Summary

Therefore, we need to go deep into the business scenarios, analyze the problems, and make arguments layer by layer to get better optimization points. Note: As optimization suggestions, they are generally proposed from the perspective of filling gaps, but filling existing shortcomings does not mean that it is the optimal solution. There may be better ideas.

Author: Down-to-earth Teacher Chen

Source public account: Down-to-earth Teacher Chen (ID: 773891)