Sincere summary: five steps to cultivate high-level data analysis capabilities

Introduction: This article briefly describes the methods to improve the team's data analysis capabilities and proposes a "five-step analysis method". It is hoped that it can help product managers solve the difficulties in the overall analysis capabilities of some teams and bring more vitality to the market.

A classmate asked: "The current data team's analytical ability is weak. I want to improve the analytical ability and combine it with the company's business to achieve quick results." It can be seen that the status of the running data machine is not only unsatisfactory to the individual, but also unbearable to the leader. But how can it be improved? Today, I will share with you the system.

Ways to improve your abilities

The essence of data analysis is analysis logic + business understanding + code writing. For students who have already started working, the most difficult part is analysis logic + business understanding. Because they cannot accurately understand business needs and cannot actively guide business thinking, they can only passively accept a large number of scattered data collection requirements with strange logic. Not only is their work efficiency low, but they will also be criticized by the business: "No insight, no value, I asked for this number but you can't even do it..."

However, this part is the most difficult to improve. First, not everyone is born with good logic. Second, each company has different businesses. Forcing a template will definitely cause problems. Therefore, the idea of ​​improving capabilities is not to look for a "one-shot model" all over the world, but to start with basic analysis templates and gradually iterate capabilities.

01 Step 1: Classify requirements by department and form a monitoring template

There are four common data analysis requirements:

1. Monitor business conditions
2. Analyze the cause of the problem
3. Predict business trends
4. Testing business ideas

Among these four categories, the most important thing to do is to form a monitoring template. Because business processes do not change frequently, monitoring indicators and classification dimensions are fixed within a certain period of time. This is conducive to the subsequent in-depth interpretation of data indicator changes, and can greatly reduce the number of irregular and temporary data collection, thereby buying more time for the team.

Different departments may have different business processes, such as sales, operations, products, supply, etc. It is recommended to design monitoring indicators for each process separately, so that the service business is more accurate. Experienced seniors in the team can guide newcomers to sort out the processes of each department and sort out scattered needs to see which ones can be integrated into regular data monitoring reports. This will not only make newcomers familiar with the business, but also improve their analytical logic ability.

02 Step 2: Understand the trend of indicators and discover the cycle pattern

After having monitoring indicators, many people will directly throw the indicators/dimensions to newcomers and say: "Data analysis is about comparison, you can see how to compare..." This is a very irresponsible and wrong approach. In the absence of analytical logic, the more comparisons you make, the more confused your thinking will be. Comparing eggplants with apples is the root cause of confusion.

So after you have monitoring indicators, don't rush. First, figure out the basic trends and patterns of the indicators. Especially the key indicators (KPI indicators) related to business assessment, such as sales, profits, number of new users, number of active users, etc. There are three types of patterns to pay attention to:

1. Natural cycle: whether the indicator is related to seasonal changes and holidays
2. Lifecycle: Trends in key business indicators from launch to de-launch
3. Cohort changes: Trends over N time periods after user registration

This process can help newcomers understand what "normal trends" and "regular changes" mean. It can greatly reduce the chances of newcomers making common sense mistakes, and at the same time, it can make newcomers more sensitive to real abnormal fluctuations. In addition, the indicator trend observation can be extended from the KPI indicator to other indicators, from the shallow to the deep, to avoid newcomers drowning in the sea of ​​data. The effect is very good.

03 Step 3: Disassemble the internal structure and discover the distribution pattern

After understanding the general trend, it is better not to let newcomers do random splits/comparisons, but to first understand the internal structure of the business. There are two types of internal business structures:

1. What parts does the whole consist of?
2. What are the components of the result?

For example, in sales, you can understand:

1. How many steps are there in the sales process and what data are recorded?
2. How many types of sales channels are there and how does each perform?
3. How many products are sold and what is the proportion of each product?

For example, supply, you can understand

1. There are several steps from raw materials to finished products
2. What resources are consumed at each step
3. What results does each step produce?

This process may be very long because different businesses have different degrees of digitization. If the degree of digitization is high, data can be seen directly. If the degree of digitization is low, you can only understand the business behavior first, and then slowly collect data or look at the overall indicator impact.

But it is very valuable to do so. Because it is the only way for newcomers to have a deeper understanding of the business. And after discovering data anomalies, the basic logic of tracing the anomalies is also to trace down along the internal structure of the indicator. This is a logical split. And this step does not require any technical content, and newcomers can do it themselves.

04 Step 4: Collect business actions and quantify proactive behaviors

After understanding the internal structure, it is better not to let new employees split or compare randomly. Instead, collect what the business does, and then take out the quantifiable parts to see the quantitative effect. For non-quantifiable actions, observe the overall indicator changes after the action occurs.

For example, to improve sales performance

1. If the marketing department is doing a promotion, they can use data to record which orders are promotional orders, observe the growth of promotional orders, and calculate the revenue of the activity.
2. If the sales department holds a sales training session, there may not be data to record how much each person has improved. At this point, the only option is to record which people/companies participated in the training and then see if the indicators have changed.

Doing so, on the one hand, can deepen newcomers' understanding of the business, and on the other hand, can help them understand the effects of various business actions from the results first. This will not only help them have more ideas when interpreting data changes, but also enable them to directly give some suggestions to the business based on the results, thus promoting data analysis from interpretation to guiding business development.

05 Step 5: Split the business problem and form analytical hypotheses

After quantifying business actions, do not split or compare randomly, but learn to make assumptions first and then find evidence. Proposing correct assumptions can not only help you form ideas faster, but also filter out various interference factors and reduce the burden of repeated data collection. Analyzing assumptions is also related to designing data tests. With clear assumptions, test sampling has a basis and it is easier to interpret test data.

The analytical assumptions come from three sources:

1. Make assumptions based on past rules, experience, and trends
2. Make assumptions based on business concerns
3. Based on the major problems found in the structural/hierarchical analysis, hypotheses are proposed

All three methods require the accumulation of the previous steps, so the hypothesis is put at the end. There may be many assumptions, and analysts need to sort out the assumption logic at this time. This is also an advanced job, which requires experienced students in the team to lead the team. For complex problems, it is best to sort out the assumption logic before handing it over to newcomers. For newcomers, being able to verify a single-dimensional hypothesis is already qualified. (As shown below)

summary

From the above, we can see that the improvement of analytical ability is centered around: understanding the business more and more and being more logical. This is why we repeatedly emphasize: do not disassemble or test randomly. Without logic, after disassembling, you will see that this is high and that is low, and various influencing factors are intertwined, and you cannot analyze the problem at all. Without logic, the testing process is random, there are 100 factors that affect the test results, and the subsequent disassembly is a pain in the ass, all of which will only make newcomers more confused.

Author: Down-to-earth Teacher Chen

Source: WeChat public account "Down-to-earth Teacher Chen"