6 steps of data operation, from methodology to cases to get you started!

Analyzing operational data helps us further perform differentiated operations for users.

1. Data Operation

1. Purpose

Analyzing operational data helps us further perform differentiated operations for users.

2. Core

Analyze what aspects the problem includes, and optimize the points that account for a large proportion and where you can exert your efforts.

2. Data Analysis Process

1. Split work items

Operations is a job that involves many trivial matters. Operations personnel must be able to split their own work items and analyze specific operational data in a targeted manner according to the characteristics of different work items in order to achieve twice the result with half the effort.

So how do you split work items? It can be broken down according to the user groups it is facing, which is usually associated with user classification, and the work can be split into those targeting all ordinary users, active users, paying users, and so on. You can also split your work into projects. You can also determine the goals for different stages in chronological order and split the work items according to your own goals.

2. Establish an indicator system

After splitting the work items, each work item has different indicators. We need to further split and refine the operational data indicators according to the characteristics of the work items, and then judge the operational problems and continuously optimize the operational plans through the analysis of each indicator.

The splitting dimension can be based on the data inclusion structure or the sub-items contained in each work item.

Taking user operation as an example, user operation includes attracting new users, promoting activation, retaining users, and converting them to paid users. As for attracting new users, the key indicators include the scale and growth rate of registered users; channel quality - what are the registration channels, and what is the registration conversion rate of the channels; registration process quality - the number of users who have completed registration, and statistics on users' dropout points in the registration process; registered user behavior tracking - statistics on users' behavior after completing registration.

3. Refine the analysis objectives

Refining analysis goals means determining data points that can be optimized based on operational goals. This step lays the foundation for the subsequent data extraction, processing and analysis.

To give a simple example, if you have just finished an event and want to know what areas can be optimized when you hold the same or similar event next time, the points you need to pay attention to include not only the final participation effect, but also: what are the channels for event promotion, what is the participation path of each channel, how many people participate in each step of the path, what is the conversion rate, etc. Once you have a clear goal for your analysis, you can then determine which data points you want to extract.

4. Extract and process data

Extracting data involves the issue of data tracking. In the early stages of product design, operations personnel must plan key operational points, list tracking points and submit them to developers to avoid the situation where they want to view certain data but have no data record information during later operations.

In addition, the extracted data must undergo a series of processing before entering the analysis stage.

So what does common data processing include?

First of all, we need to clean the data we get, which is the process of dealing with duplicates, missing items, contradictory items, and abnormal peaks or troughs in the data. There are many ways to remove duplicate items, which I will not go into detail here. The most common way to deal with missing data is to fill it with an average value, which can be the arithmetic mean of all data or the average value over a period of time. Contradictory items refer to erroneous data. For example, the data should have been all single-digit numbers, but more than one digit appeared in the extracted data, or an email address appeared in the name field, etc. At this time, you need to check whether it is an error during data extraction or data entry. If it is an error during extraction and the error has a great impact on the result analysis, it should be promptly fed back to the relevant person in charge.

Pay special attention to the peaks and troughs in the collected data, as these are often the key to problem analysis. Generally speaking, the reasons for peaks or troughs in data include additional promotion opportunities, system failures, statistical bugs, etc.

Secondly, the data needs to be further processed. Because the extracted data may not be suitable for direct analysis, some functions and tools are often used at this time, such as VLOOKUP function and PivotTable.

After the above cleaning and processing steps, data that can be used for preliminary analysis are obtained. Further processing of these data is required in order to conduct in-depth analysis.

5. Data analysis summary

(1) Data analysis methods

Common data analysis methods include comparative analysis, structural analysis, average analysis, weight analysis, DuPont analysis, etc.

1) Comparative analysis

It refers to making comparisons along different dimensions to explore changes in data and discover the patterns or insights contained therein.

The dimensions of comparison include: comparison with expected goals, comparison in different time periods, comparison with peers, comparison with results before operation, comparison between different users, comparison between different operations, etc.

Next, we will use an example to explain how to create user profiles by comparing data from different time periods and adjust operational strategies based on the user profiles.

The line chart above reflects the changing pattern of the number of daily active users of a certain product. From early April to early July 2016, it basically changed in a cycle of one week. The points with larger data were mainly on weekends, so we can infer that the main users of this product are students. Moreover, the daily active data dropped slightly in June, but rose after mid-July, and the changes basically coincided with the time of students' final exams and holidays, further confirming the user portrait.

2) Structural analysis method

The comparative analysis between each part of the analyzed population and the whole is often expressed by the structural relative index (= (part/total) * 100%). The larger the value is, the greater the weight of the part in the whole, the greater its importance, and the greater its impact on the whole.

3) Average analysis method

It reflects the general level of an indicator under certain conditions and is mostly used to measure the health of a business.

For example, a certain product has three sales channels, A, B, and C. If you want to know which of these three sales channels contributes the most to revenue, you can calculate the average sales of these three channels. At this time, it should be noted that the "average" in the average analysis method has a prerequisite, and it must be based on whether the data used to calculate the average are valid. For example, if the sales data of channel A suddenly drops to 0 one day, this is very abnormal. At this time, we need to find out where the problem lies. If it is because some sudden failure occurred in channel A that day, then this data should be excluded and then the average value should be calculated.

Does that mean a higher average means a healthier business?

uncertain. For example, A sells down jackets and B sells short skirts. In summer, A's average sales volume is lower than B's, but this does not mean that A's business is worse than B's.

The average analysis method is only meaningful when the businesses and situations of both parties are relatively similar, or as we often say, comparable.

4) Weight analysis method

Convert multiple indicators into one indicator that can reflect the overall situation for analysis and evaluation. The specific approach is to determine the weight of each indicator, and then summarize the processed indicators to calculate the comprehensive evaluation index. It is often used to analyze subclasses that are in a parallel relationship.

As shown in the figure, a product has three promotion channels - A, B and C. These three channels are further divided into conversions through recommendations for purchasing maternal and infant products, conversions through participation in related offline activities , and conversions from public platforms . If you want to measure the quality of channels A, B, and C, you can set a certain weight for each segmented channel, define the formula corresponding to the indicator "channel quality" (for example: channel quality = the number of recommended conversions after purchasing maternal and child products * 60% + the number of conversions through offline activities * 30% + the number of conversions through official accounts * 10%), and compare the quality of the three channels by taking the weighted sum.

What is the basis for setting weights? One is based on the importance of each segment indicator, and the other comes from past operating results. Let’s take the product just now as an example. Assuming that the product is related to maternal and child products, then based on previous operating experience , users who are attracted through recommendations after purchasing maternal and child products have a higher probability of being converted into active users in the future. Therefore, the weight of this channel can be set high accordingly. The user churn rate directed through the official account is extremely high, so its weight can be relatively low.

5) DuPont analysis

The DuPont analysis is a comprehensive analysis method created and first adopted by the American DuPont Company. By utilizing the internal connections among various indicators, you can conduct a comprehensive analysis and evaluation of your operating conditions and benefits.

As shown in the figure, assuming that the recent revenue has decreased after the product update, the boss asked us to analyze the reasons and what adjustments can be made, then we can split the revenue - Revenue = Number of paying users * AR PU (average revenue per user). Next, we will split the number of paying users: number of paying users = number of active users * payment penetration rate. It is observed that the paid penetration rate has hardly changed, while the number of active users has decreased, further breaking down the number of active users. The number of active users = active users among new users + active users among old users. If the number of active users among old users increases and the number of active users among new users decreases, it can be further split. Then analyze: new users = number of people covered by promotion * conversion rate. When the conversion rate remains basically unchanged, segment the promotion channels. According to the data, channel one has decreased while channel two has increased. Continue to segment further until the indicators can no longer be segmented. Analyze the segmented indicators to see which ones have a greater impact on the final revenue, what are the reasons for the changes, whether they can be improved through human adjustment plans, etc.

(2) Causes of data fluctuations

Common reasons for data changes: time, promotion and reach, operational activities, related characteristics, user attributes and composition, failures, and industry trends.

I won’t go into detail about the first three, but will talk about the last few elements here. The so-called correlation characteristics are actually the elements just split out through the DuPont analysis method, while user attributes and constituent elements refer to the fact that for different users, the daily activity, payment and other data of the same product or activity will change. The impact of industry trends on operational data: Let’s take an example that was very popular last year - O2O . When the concept of O2O was particularly popular last year, a large amount of capital was poured into this market. Stimulated by various subsidies, the number of users surged. Now that the market is more mature, the growth in the number of users has been relatively slow.

(3) Summary

After analyzing so much data, the final conclusion needs to be reported to the boss. So what does the summary include? Generally speaking, it is necessary to explain where the problem occurs and where optimization and improvement can be carried out.

When presenting conclusions, charts and PPT are often used. PPT is not the focus of this article, so I will not go into details here. So what are the things we need to pay attention to about the chart?

First, you need to choose a suitable chart. For example, if you want to see the proportion of different projects in the total projects, you can use a pie chart. If you want to see the trend of data changes, you can use a bar chart or column chart when there are only a few projects. If there are a lot of data items, you can use a line chart.

Secondly, the chart should be complete and should include: title, axes and units, legend (, footnotes, data source), etc.

In addition, each picture reflects one point of view, and the title should directly state the problem reflected by the data. For example, if you have analyzed the time periods when users of a product are active, you should not write "user active time periods" in the title. Instead, you should write the conclusion reflected by the chart - "user activity is high in a certain time period, and low in a certain time period." This way it is clear at a glance, and your boss can quickly understand the core information that your chart is trying to convey.

6. Feedback and application

A careful observation reveals that the above data analysis process actually forms a closed loop. After the summary report, we need to apply the conclusions into practice, continue to observe changes in data and continuously optimize our operating strategies.