It takes about 3 minutes to read this article. I wrote two basic things about product managers and how to keep them effective when doing data analysis at work. One of them is about “mind method” and the other is about “movement”.
01 “Make it right”
When I first came into contact with archery, I felt that my movements were okay, and I could hit the bullseye from time to time. All that was left was a lot of practice. But in fact, all things like my technical movements are bad, luck and other factors make me unaware of this. So I mistakenly thought “I don’t have a problem!” So over time, I found to my chagrin that my efforts did not make progress.
Also as a data product manager, I often think about whether we have ever mistakenly thought “I have no problem!” in data analysis.
In fact, I have seen a lot of invalid data analysis and related data requirements. They all have several similar “common points” to the above question:
- A lot of “vanity indicators” are cited, but nothing seems to be said, which makes it seem that there is no problem;
- Impose the results of the analysis on the pre-set conclusions to force no problems;
- Quoting a large number of wrong indicators, and then drawing more wrong conclusions, no problems are found;
- After the conclusion is reached, it cannot be acted upon and the problem cannot be solved.
There is no doubt that everyone knows that such analysis is wrong. But many times, we are doing similar things without realizing it.
In fact, even many veterans who have done data analysis for many years will make similar mistakes from time to time. Just as we need to judge the authenticity of the “product demand”, when we analyze it, its “purpose”, “method”, and “conclusion” also have differences in authenticity and some subtle actions. And if we are not paying attention, we will step on the “pit of data analysis”, and this is a common thing!
Therefore, it is possible to find and correct the problems of data analysis, so as to carry out effective analysis, from the “data sense”!
When we are doing data analysis, the first important thing is “data sense”. It is a way of thinking or even an intuition, which can help us quickly distinguish whether the analysis indicators, measurements, etc. are correct and effective in the scenario we want to analyze.
But this is a kind of ability that requires long-term correct practice, and often we don’t have so much time and energy to polish it. So as product managers, it is necessary for us to master data analysis? (It is a lie to say that data analysis is simple and does not need to be deliberately mastered.)
In fact, for the product manager, data analysis is not so “tasteless”, and there is no need to worry about how difficult it is. Before becoming a data product manager, I have done some business-oriented projects more or less. The purpose of writing this article is also to take this opportunity to use some of my own experience to explain as much as I can about “how to effectively do product data analysis”, and then through some “action improvement”, let our Improved analytical capabilities.
Due to space limitations, I tried to summarize a few dry goods in order to quickly generate a “data sense”:
- Good indicators are naturally comparative and computational (you can only look at indicators that can only be seen);
- The conclusion must be executable (those that cannot be done, put it aside first);
- The correct result must be referenced (otherwise it is easy to produce illusions);
- Average and total are not the only two methods for calculating results (mode, median, upper and lower quartiles, and variance are also important);
- For the same thing, there are two indicators that can be used, and only one of them must be selected (for new users, whether registration is considered new or placing an order is considered new).
- Indicators and before and after analysis, there must be a logical relationship or a direct influence relationship (can be justified before and after)
02 “Just do it”
The first thing is to set the analysis goals before the analysis begins. For each product line you are responsible for, this is the first thing.
In different stages of product development, we need to focus on different analysis goals. For example, before the user conversion is resolved, there is no need to pay too much attention to the retention and loss indicators of users. Because we did not ensure the stability of the core product form, most of the users are determined by whether to meet user needs. So at this time, the analysis of user retention and churn becomes a post-conclusion. Yes, but it is not the point, because they do not help us solve the problem at hand. (Early when the product was launched, I knew that our churn rate the next day was 80%, but it was just a conclusion. We cannot change anything through this indicator, because the conclusion cannot change the conclusion itself!)
Regarding the analysis goals, we usually have the following four types of goals for a product line to focus on one by one:
- Does the product design meet expectations, is the basic transformation of each link normal, and where is the problem at the moment? (We have known uncertain issues);
- User growth/retention/churn, is our user pool normal, and is there still room for operational improvement? (We have known and determined problems);
- The results of user interaction, such as browsing time, decision-making time and other behavior characteristics, how to use it? (Determining the problem that we don’t know yet);
- Do users have other potential needs, can they be tapped, and how much value does it have? (Unknown problem we don’t know).
The second thing is that in the analysis process, after we have clarified the analysis goals, we must build an effective, simple and clear analysis framework and indicator system.
In fact, most of the analysis process is reusable, and mastering a few ready-made methods can help us get started quickly. And most of our analysis is inseparable from these basic methods. About building a model, it is actually another larger content, which I will share separately in a later article. In this article, I can only talk about some basic ideas.
Apart from some common models such as AArrr and RFM, how can we effectively analyze the data? In fact, we can use some simple analysis model ideas to do it out of the box:
Causality before and after segmentation indicators:
This is a bit similar to a funnel analysis. When a target we are analyzing has a front or back logical node before and after it, then we should analyze the front node first, and then observe the back conclusion.
E.g:
New users start the App to the registration process, we need to analyze the registration conversion rate. Suppose we have a user registration product design, and completely rely on it to complete the registration conversion. Then for the analysis of the registration conversion rate, the analysis of the difference of the guiding strategy is the pre-analysis. At this time, if we only look at the conversion rate and conversion performance (such as the registration conversion rate, the number of registered people, and the duration of registration), and ignore the different factors in the user registration guidance strategy, we will not be able to find and solve the real problem in real time. At this time, we should pay attention to the implementation effect of different strategies (ABC strategy, whether it meets expectations, industry standards, whether there are large differences), and whether the user behavior goals after registration are healthy (whether there are a large number of “unconverted registered users” “, whether there are follow-up differences between different strategies), we can know whether our registration conversion rate is really normal, where the problem lies, and whether it is effective.
We segment the cause and effect, it does not digress, but disassembles it in detail. Find the conclusion through different results of the same caliber.
Analysis of influencing factors:
When we analyze a certain user goal, whether there are multiple parallel influencing factors. Then we should find out the degree of its influence on the target, and analyze which factors are wrong.
For example: when we analyze the conversions placed by users of our own apps, there are usually several parallel factors before users click on the details. For example, prices and discounts attract users’ clicks, title descriptions and pictures attract users’ clicks, ranking and exposure may affect clicks, and exposure scenarios may affect clicks. The above factors usually exist in parallel. At this time, we have to find the biggest impact factor (or find the same) through analysis, and then analyze the problem. Whether the price is sensitive, whether the description affects the decision-making, whether the ranking and exposure strategy is correct, and whether the scene has room for optimization.
Due to space limitations, I will not expand more content here, but only provide some inspiration.
03 End
In fact, the data analysis of product managers is also a practical process that requires correct goal-oriented and fixed methods, and contains many contents. I can’t finish a single piece of content at once, so I can disassemble this huge content in detail. (Big enough ambition)
As product managers, when we master some of the “actions” of data analysis, we can at least accomplish the following things:
- Through simple and clear indicator comparison, we can monitor the daily changes of the product to determine whether something is wrong;
- We can monitor through several core indicators, and we can quickly determine whether it meets product expectations;
- We can analyze the conversion rate, we can clearly know which link is the problem;
- Through behavior analysis, we can study whether the needs of users are hit;
- We can review the data through data, we can just quantify the value of the product and find the shortcomings;
And some important skills: burying points, SQl introduction, user behavior analysis, strategic transformation of business products, welcome to follow my follow-up articles.
