Say goodbye to confusion! A guide to developing data analysis ideas

Data analysis is the core of modern business decision-making, but many novices often feel overwhelmed when faced with complex data. This article provides a clear guide for data analysis novices to develop their thinking and help everyone build data analysis capabilities in a systematic way.

A major problem that bothers novice students is that they have no idea of ​​data analysis when encountering problems. Common problems include:

1. Don’t know where to start

2. Not knowing how to conduct in-depth analysis

3. I don’t know what level of success is OK

Attention! Don’t think of all kinds of high-sounding models when it comes to data analysis ideas. In actual work, many business problems do not require textbooks such as "Statistics" and "Machine Learning". It only takes 5 steps to form a complete analysis idea based on the business:

Step 1: Identify business scenarios

"Which department's question am I answering?" is the first step in forming an idea. Data analysis is not fortune-telling. It is not like flipping a coin and getting the answer from heaven and earth. It needs to be specific to the needs of the department and colleagues. Common ones include:

Pay attention here! At work, people often ask, "Show me the sales data!" or "How are our users?" At this time, it is best to ask: which department or which leader needs to see it.

Because in the second step you will find that even in the same scenario, different departments focus on different indicators.

Step 2: Clearly analyze the indicators

After clarifying the business scenario, you can further determine the analysis indicators. The main indicators of most business scenarios are fixed. When doing analysis, there is often a main indicator + several secondary indicators/dimensions to jointly explain the problem.

For example, the main indicators of sales indicators are revenue and gross profit, but the finance/commodity/operation/supply departments have different focuses and need to use different indicators (as shown below):

In this way, selecting indicators + dimensions based on specific business needs can help observe data more effectively.

Some people lack good habits. When others need data, they always throw out whatever dimensions/indicators are in the database without any discrimination. Too many useless indicator dimensions piled in the report will only confuse their own thinking.

After sorting out the indicators and dimensions, you need to further clarify the criteria for judging the problem. Pay special attention here! Not all indicators have clear KPI assessments. You need to pay attention to the criteria that the business really cares about (as shown below):

Step 3: Locate the source of the problem

In the second step, if you find that the business is doing well, the next step is to analyze "where the good reasons are". If you find that the business is not doing well, the next step is to analyze "where the problems arise".

The third step is mainly about positioning, figuring out “where”.

To locate business problems, you can ask the following five questions in sequence (as shown below):

Take sales analysis as an example. If you find that sales this week are not up to target, then ask:

1. Is the deviation from the target small (-5%) or increasing?

2. Is this the first time in the past 12 months that the target has not been met? Or has it been met several times before?

3. Has the standard been failed to be met continuously in recent weeks? Has the trend of failure to meet the standard been expanded?

4. Are there problems with all products/all regions/all sales channels, or only with XX?

5. Did the sales department take any intervention action? What was the effect? ​​Did the marketing department do anything?

This fixed order can greatly clarify thinking and improve analysis efficiency. Because a small fluctuation of 1% or 3% in a short period of time is likely to be a "small problem". Before the report is finished, the indicator will rise again. If we look at the trend changes over a longer period of time, it will be easier to judge whether there are long-term and profound problems. It can also help us trace the source of the problem, making it easier to form hypotheses.

Step 4: Propose analytical hypotheses

Many people stop at the third step and just say "because product A is not selling well, the sales have not reached the target, so it is recommended to raise it." This kind of report is most likely to be criticized as "no analytical thinking, not in-depth enough."

Because "we have to do it high" is nonsense. What the business wants to know is: who will do it, how to do it, what to use, and how high it will be.

This step will stump many people. People often complain: I am not in business, how can I know how to improve performance? ? ? Attention! As a professional data analyst, you should not rely on business experience to make suggestions, but you can make suggestions through analysis. There are four common ideas:

Taking sales analysis as an example, if it is found that sales this week are not up to target and it is mainly caused by poor promotion effect of product A, then the corresponding four methods are:

1) Is there a product B of the same type/same price range as A, but with better promotion? Please refer to the operation method of B

2) A promotion is not good, is it because the ad report has few clicks, or the user conversion rate is not good after entering the store? If there are few clicks, change the promotional image; if there is no conversion rate, change the detail page

3) Keep the current approach and add more promotional expenses to A to see if this can work wonders.

4) I see that the competitor has a strategy that seems to be very effective. Let’s try it and see if it works.

It should be noted that both 3 and 4 require the cooperation of the business to verify the data. Otherwise, it is impossible to draw a direct conclusion based on historical data alone. Some companies are too conservative and dare not try, and insist on analyzing based on historical data. As a result, the business never accumulates experience, and not only the analysis ability cannot be improved, but the business itself cannot do well. Testing and historical data reasoning are equally important analytical methods.

Step 5: Verify the analysis assumptions

Note! Any assumptions about a single influencing factor are easy to verify (as shown below):

However, multiple influences intertwine and become troublesome. At this time:

1. It is necessary to use the MECE method to sort out the analysis logic and extract the truth

2. To quantify complex business behaviors, you need to label and find features

3. Sampling testing requires the use of statistical methods and experimental design

4. To make observational causal inference, a regression model is needed

Especially when there is a conflict of interest between departments, and each department blames each other, it is even more difficult to deal with (as shown below)

Therefore, when the analytical ability is weak, it is recommended to convert the problem into a single-factor hypothesis verification first, and test whether it is feasible one by one. However, many companies prefer simple and direct methods. For example, if they can verify that "price reduction is effective" or "changing promotional materials is effective", it is enough for the business to execute. Therefore, students should at least complete steps 1 to 4 + step 5 single-factor verification. We will share the complex methods separately later.