Source: Down-to-earth teacher Chen
"If you look at the recent sales figures, what do you find?" The annoying part is: most of the time, the daily data is just a little fluctuating. If you directly report the conclusion of "3% month-on-month growth", you will be deducted with the hat of "I also know this, I need to analyze it in depth!". So what should we do?
There is a standard order for interpreting data, which is divided into 7 steps: looking at the numbers, finding the rules, setting standards, looking at the structure, clarifying the hypothesis, checking the authenticity, and drawing conclusions. Let's not be in a hurry, let's talk about it step by step:
Step 1: Look at the numbers
This is the most basic, year-on-year, month-on-month, absolute value, up and down...... The daily reports are all these things. But these things are not welcome. First, as long as a person knows the situation at a glance, there is no need to write anything, and second, these things have no business meaning, and talking about them is the same as not talking about them, so they must go deeper.
Step 2: Find a pattern
If you want to go deeper, you can drag the data time a little longer to see if there are any natural patterns. There is no technical content in this step, and you can just connect the daily reports to see them, but it is very effective! Because many conventional data fluctuations have periodic laws. By mastering the rules, you can avoid fuss, false positives, and false positives. You can also keenly observe the real problem (as shown below).
Step 3: Establish standards
If you want to go deeper, you have to find a judgment standard. Data + standard = judgment, with good or bad judgment, we can continue to think: why is it good/why is it bad. The best criterion is to have a KPI value on the head, so that a direct comparison of the KPI completion rate can be concluded.
However, there are some non-core indicators that do not have KPI requirements, so you have to find other standards. For example, the scenario dismantling method is used to find out the relationship between non-core indicators and KPI indicators, and the numerical range of non-core indicators when the KPI is met, so that judgment standards can also be formed and judgments can be made.
Step 4: Look at the structure
With a good or bad judgment, you can think further about the reasons. But before thinking about the reasons, it is better to look at the internal structure of the indicator and find the bulk of the influence of the indicator. This makes it easier to see where the problem is.
For example, looking at the sales situation, sales pay attention to people, goods, and fields, first from the three dimensions of users, goods, and channels, and look at the internal structure separately to see which type has a high proportion and which type is performing well/poorly at present. In this way, it is easy to form ideas by distinguishing the key points.
For example, if you look at the cost situation, you can distinguish between variable costs and fixed costs, and variable costs can be distinguished from commodity costs and marketing costs. Fixed costs distinguish between front-end and back-office costs. This makes it easier to see which piece is the source of the fluctuation.
With this step, it is much easier to find the cause in the future, and you can get straight to the point.
Step 5: Column assumptions
Some lazy students draw conclusions directly from the previous step. For example, the reason why product A has not been selling well recently is because product A has not sold well. The cost is high because the promotion costs too much......
But this reason is often too superficial. First, it is possible that product A does not sell well because of other factors that hide deeper (there are deeper factors); second, it is possible that product A does not buy well because certain types of users are losing (other factors affect it);third, even if A does not buy well because A is not good, it may not be corrected in the short term, and we still have to think of other ways (the feasibility of problem analysis)
Therefore, if you go further, you must make clear assumptions and figure out the logic behind the problem. Many students will be dumbfounded at this point, thinking that there are thousands of reasons, how should I list them reasonably?
Here are two simple ways to do this:
1. Start with recent events.
2. Start with possible actions to be taken by the business.
Start with what has happened recently to quickly find hypotheses that explain the source of the problem. We can start by collecting the recent positive/negative events and then look at them one by one:
Theoretically: which indicators this event has an impact on
Actually: the degree to which the event occurred corresponds to the change in the data
Investigate one by one to find out the source of the problem.
Starting with possible actions for the business can quickly identify assumptions about how the business will respond. For example, in the face of declining performance, the business will be three-pronged in the short term:
1. On the promotion, send a bunch of coupons
2. Engage in training and grasp a few typical demonstrations
3. Change the copywriting and change the promotion link
So, we can make assumptions:
1. According to the past input-output ratio, promotion can boost performance
2. The personnel are uneven, and there are benchmarks to refer to
3. The promotion is uneven, and there are benchmarks to refer to
After that, you can check them one by one.
Step 6: Verify authenticity
With a hypothesis it can be verified. Note that a lot of daily data fluctuations are not available to do ABtest for us to verify them one by one. Therefore, the verification here is more about finding evidence. Find enough obvious, data evidence to substantiate a point.
For example, if I receive a recent product price adjustment information, then theoretically, if it is a best-selling product, the price adjustment will increase income if the supply exceeds demand, while the price adjustment of ordinary goods will only hurt sales. So the idea of verification is:
1. What is the past sales and inventory data of the adjusted product (judgment type)
2. On which day did the price adjustment start to be adjusted, and what changes did the sales have after the adjustment?
3. How big is the impact of the price adjustment commodity, and there are other problems if you eliminate this commodity
In this way, the data can be used comprehensively to make judgments.
For example, let's assume that doing promotions can boost performance. Then you can take out the previous promotion effect data for reference
1. How much was invested at that time, and how many days did it
2. How much it improved at that time
3. At present, according to this amount, can the pit be filled
In this way, you can also make a judgment: if you go on a promotion now, can you save the situation, and what measures are needed.
Step 7: Draw conclusions
At this point, we have done enough homework, and when we turn in our homework, we can make a very detailed report:
1. The status quo is good/very poor, and the performance is ...... (Conclusion of step 123)
2. The status quo is good because of ...... (Conclusion of Step 4)
3. The deeper reason is ...... (Conclusion of Step 5)
4. This good expectation is sustainable/unsustainable because ...... (Step 6 Conclusion)
5. Therefore, it is recommended to ...... (Continue to observe/take measures/brainstorm further proposals)
In the annex, the detailed data process is attached, which is both comprehensive and in-depth.
The sequence of the 7 steps
Note that these 7 steps don't have to wait until someone asks a question to start doing. Because:
Steps 1, 2, and 3 are completely basic data interpretation, and you can do it on a daily basis
The fourth step is to collect recent business actions and major industry events, which can be done on a regular basis
The fourth step is to do a review of the past actions of the business, which are recorded in history
In fact, there are only two things to do in step 5: using historical data to measure and verify the impact.
Therefore, we often say that if data analysts want to strengthen their data insight capabilities, they need to accumulate more analysis experience, collect business actions for specific business problems, and review them more often, so that they can understand more and more deeply. Every time a specific question arises, there is an abundance of ammunition depots available.