Data tracking and user behavior analysis: Start with the end in mind and break it down layer by layer

Friends who are planning to read this article. This article is suitable for new product/operation/data people aged 0-3 years old. It was written about 3 months after I joined the company as a newcomer. I believe that the perspective of a newcomer will make my writing more understandable, and I hope it can help you.

User behavior analysis refers to analyzing a series of user behaviors to improve product design and operation strategies. As the Internet develops from extensive to intensive, user behavior analysis can help product design and operation strategy formulation shift from "feeling based on feelings" to "scientific and quantitative".

Simply put, if we compare the Internet C-end products to a large amusement park, the product manager and the front-end and back-end R&D are the architects and engineers of the amusement park, respectively, creating the basic framework of the amusement park. Operations are the activity planning of the amusement park, and various activities are held in the amusement park from time to time. User behavior analysis can quantitatively analyze the behavior of tourists to improve the architectural design and activity quality of the amusement park.

1. Basic idea: start from the end and split layer by layer

From an engineering perspective, if you want to analyze user behavior, you need to go through the following steps:

1. Collect user behavior data (embedding points + data storage)
2. Cleaning and processing this data
3. Analyze the data

In a large Internet company, data is often like a commodity on an assembly line, going through layers of steps from production to processing to final use, and ultimately delivered to different application departments. This makes it easy for different roles on the data chain to only see the trees and not the forest.

But data ultimately serves business goals, and understanding data analysis also requires a more complete idea. At this stage, I summarize it as: start with the end in mind and break it down layer by layer.

1. Start with the end in mind: On the one hand, from the perspective of data production, data analysis can only be carried out after data tracking is completed, and finally conclusions can be drawn. However, in reality, it is often necessary to start with the end in mind: first think about the ultimate purpose of data analysis, then determine the idea of ​​data analysis, and finally determine the tracking solution. On the other hand, in the process of data analysis, it is also necessary to think about the ultimate indicator first, and then infer the process indicator.

2. Split layer by layer: I will introduce in detail below that the process of data analysis is also a process of breaking down big goals into small goals.

2. Data Analysis: Indicators, Dimensions, and Calibers

The basic idea of ​​user behavior analysis is: determine the quantitative indicators to measure the success of the product -> think about which dimensions will affect the changes in the indicators -> determine a unified caliber and calculate the changes in different indicators under different dimensions.

The following is a further explanation of the indicators, dimensions and calibers.

1. Indicators

An indicator is a quantifiable business goal. For e-commerce, the ultimate indicator is GMV; for advertising, the ultimate indicator is conversion rate and ROI; for games, the ultimate indicator is DAU and ARPU value.

To achieve the final target, it can also be broken down into multiple process indicators. For example, for e-commerce, the link before the user pays can be broken down into: enter the app-> search (recommendation/live broadcast, etc.)-> enter the product details page-> add to the shopping cart-> checkout-> payment. Each link can set an indicator.

2. Dimensions

Dimension is a perspective for analyzing changes in indicators, and is also a factor that affects changes in indicators. Zhang Xiaolong, the father of WeChat, once said: Product design is about classification. Dimension can also be understood as a perspective for classification.

How to understand it? If DAU increases, is it the increase in male users or female users? User gender is a factor that affects DAU and is also a dimension that can be used to analyze DAU changes. Similarly, the price and category of goods are also dimensions used to analyze GMV.

3. Caliber

Caliber is a standard for calculating indicators.

On the Internet, there are often multiple ways to calculate the same indicator. For example, DAU can be calculated by removing duplicates using the user's device number or by removing duplicates using the user's unique account number. Each calculation method (or standard) is a caliber.

3. Data Collection: Events, Timing, Parameters

On the basis of clear data analysis objectives, we need to consider how to obtain the required data to meet the analysis objectives.

As we can see from the above, the idea of ​​data analysis is to determine the indicators first, then consider the dimensions, and finally unify the caliber. Data indicators need to be obtained by collecting statistics of corresponding user behavior events . The analysis dimension requires us to record the parameters related to the event while collecting the event. An important part of unifying the caliber is to clearly regulate the timing of the event when collecting the event.

Below is a further explanation of events, timing, and parameters.

1. Events

Each action of a user is called an event. Generally speaking, the actions of users on the Internet can be divided into three categories:

1. Browse: The user enters a page of the product

2. Exposure: An element/information on the product page is displayed to the user

3. Click: The user actively clicks on an interactive element on the page

When embedding points, each event will have a unique id (usually called event id) - such as clicking on the home page focus image (Home_FirstFocus_Click).

2. Timing

Timing is the exact moment when each behavior is recorded. Just as a caliber is needed when calculating indicators, timing is also a more precise definition of tracking events and a caliber of tracking.

1. Browsing time: before entering a certain page interface and after the interface returns

2. Exposure timing: When an element is exposed xx% and the user stays for xx seconds

3. Click timing: triggered immediately after the user clicks

3. Parameters

With the definition of events and opportunities, it is necessary to clarify what specific information needs to be collected when this event occurs.

Collecting user behavior is like writing a narrative, which requires recording the following: people, time, place, and specific events. In addition to the event ID, other information that needs to be recorded is abstracted as the parameters of this event. The parameters of an event are similar to the context and conditions for executing an event. Some people abstract the information that needs to be recorded for tracking points into 4W1H:

1. Who: Who is the user - that is, user ID, device number, etc.

2. When: When - the specific time when this happened

3. Where: Where is the location? There are two levels of understanding: one is the user’s actual IP address, and the other is the page where the event occurred.

4. What: What was done - the type of event (view, exposure, click, etc.)

5. How: How to do this - that is, the business attributes when doing this, such as the price and discount information of the product when clicking on the product

For each event (such as clicking on the homepage focus image), different people, at different times, and in different situations will perform this action, and each specific action recorded is called a log. In this way, every action of the user is recorded in detail.

4. A specific case: e-commerce data analysis and tracking ideas

With the above concepts, understanding data collection and user behavior analysis is still somewhat abstract. Let's start with a typical e-commerce case to understand how to analyze e-commerce user behavior step by step.

Step 1. Think about the user journey

A user entering an e-commerce app through affiliate advertising may have the following user journey:

Click on the affiliate ad -> enter the app -> product card is exposed -> click on the product card -> enter the product details page -> click to add to cart -> click checkout -> enter the checkout page -> click to pay -> enter the payment completion page

Step 2. Determine core indicators and important dimensions

The core indicator of whether an e-commerce product is successful is the final GMV, but achieving the GMV goal requires users to go through a series of user journeys and finally pay for a product. Therefore, we can set process indicators for each step in the user journey.

GMV = paid PV * average order value

Payment PV = Exposure PV * Exposure click rate (PV) * Click-to-cart rate (PV) * Cart settlement rate (PV) * Settlement payment rate (PV)

Exposure click rate (PV) = click PV/exposure PV

…

After breaking them down one by one, we can get all the non-composite indicators, which also correspond to the events that need to be collected.

In addition, e-commerce companies often conduct multi-dimensional analysis of indicators from the perspectives of people, goods, and places, which also corresponds to the parameters that need to be collected for each event.

Step 3. Design of embedding points

Based on the indicators to be calculated and the dimensions to be analyzed, define the events and parameters that need to be tracked, and determine the timing of reporting. We can use a table to clearly identify all the information that needs to be collected.

Step 4. Data collection and analysis

After understanding the user journey, determining the core indicators and analysis ideas, designing the embedding points and collecting the data, we finally started data analysis.

In this stage, we use the collected data to perform calculations based on the ideas we have just determined, and analyze the results.

1. Calculate the final data indicator GMV first to see whether GMV is rising or falling

2. Observe which intermediate indicators caused the final result

3. Think about the main dimensions that affect these intermediate indicators, and analyze which dimensions cause changes in the indicators.

4. Finally, a relatively scientific conclusion was obtained

There are still many details that need to be paid attention to in the data analysis process. This article only introduces the overall idea of ​​data tracking and analysis.

V. Conclusion

The so-called data-oriented approach is actually a way of thinking that starts from the end. It establishes the ultimate goal and breaks down the big goal into process indicators and different dimensions, which are quantified and presented with data.

In addition, although each product category has an ultimate North Star indicator (such as GMV for e-commerce), if we want to build a healthy and complete ecosystem, we need to build a good indicator system to measure the quality of a thing in many aspects. For example, for e-commerce platforms, the number of merchants' supply, user portraits, user stay time, etc. are all indicators that fully describe the health of a platform.

The same is true for individuals. What is our ultimate goal? What are the processes to achieve this goal? In addition to the ultimate goal, what other different goals do we have to make our lives more exciting? These are all questions worth pondering. Of course, life has more flexibility, just like the user path of a certain type of product may be completely overturned by a successful innovation. Only by thinking about products outside the existing data framework can we create greater possibilities.