Introduction: Data show that the transaction rate of product A and B is male > female, but the transaction rate of product A and product B is female > male. Do you believe?

Refined operation can help us adjust the product route and strategy under the guidance of data, so that the process of product improvement can be followed by rules. But sometimes, we run into data traps that look a little weird.

Case data

Let’s start with the problem description:

In product A and product B, the purchase turnover rate of men is male > female, but in the total, it is female > male. So, this is kind of confusing.

So,, should we reach out to men or women?

Our first reaction might be: Is the data inaccurate? And so we have this case…

Here we sincerely thank Tencent mobile analysis MTA users willing to grow up with us, to trace and analyze the underlying reasons of data.

Reasons for positioning

Let’s trace the source of the problem:

* The data here has been blurred and is not real

As can be seen from the data, in fact, in both product A and product B, the conversion rate of men is higher than that of women. However, since the conversion rate of product B is significantly lower than that of A, and A large number of men are guided to product B, the overall conversion rate is lower than that of A.

This is known in statistics as the Simpson’s Paradox.

The party that is dominant in group comparison is sometimes the party that is losing power in overall evaluation. The phenomenon was discussed in the early 20th century, but it wasn’t officially described until 1951, when E.H. Simpson published a paper describing it. The paradox was later named after him, Simpson’s Paradox.

Optimization scheme

Going back to the title here, the data tells us two things:

1. For both product A and product B, the purchase turnover rate of males is higher than that of females;

2. The average turnover rate of product B is significantly lower than that of product A.

So for launch decisions, we still prefer men. Under the same drainage cost, males have a higher conversion rate. For the split recommendation method of product A and product B, the customer unit price and profit rate of the two products may also be considered.

As can be seen from the figure above, product A belongs to the product with low unit price, low profit and high turnover rate, while product B belongs to the product with high unit price, high profit and low turnover rate. If we hope to introduce more users in the e-commerce development period, we may focus on promoting product A. If the platform has a certain scale and we hope to improve the per capita profit margin and reduce the customer acquisition cost of 100 yuan transaction volume, we may consider promoting product B more.

(Probably two products like this ↑)

All of this, of course, assumes that the cost of drainage is the same. And often in the consideration of practical problems, we will also need to consider the cost of delivery, delivery effect, transformation effect and other issues, this part in our first chapter also discussed, welcome to move on to discuss!

Case summary

In the actual combat of data operation, we may often encounter some data that makes us feel a little uncomfortable.

But behind these data, actually contain a lot of detail and energy. Therefore, it is very important to establish a detailed data analysis operation system, to understand the separation and integration of user groups, and to make the data available to you.

Review the data operation smile model we talked about in the last issue

In this issue, we talked about the operational value of the user group after breaking it down and analyzing in the process of locating the cause.

We can first define the event burying point to monitor the purchase event and report the parameters of the purchased goods. Then, through the user group setting, we screen out the people whose purchase parameters are equal to product A in the purchase event, so as to obtain the segmented user groups we want:

Then, through the calculation and analysis of the user group, we get the population characteristics of the group and start our data analysis work.

About us:

Tencent big data platform focuses on big data platform construction, data mining, data application, etc. Information sharing to promote industry exchanges. Through years of product construction, Tencent Big Data has successfully provided developers with Tencent Mobile Analysis (MTA), Tencent Mobile Push (carrier pigeon), Tencent Recommendation and other data products. Meanwhile, it cooperates with Tencent Cloud to launch big data processing suite (Shuzhi), providing reliable, safe and easy-to-use big data processing capabilities.