How to analyze when the indicator falls?

Lead: “When XX index fell, how would you analyze?” XX including but not limited to sales, the number of users, active rate, ROI, and so on. This problem is often encountered when a interview, a very common problem is the daily work. Students can often forgetful in answer, it seems no matter how speak speak insufficiency. Are often entangled in their daily work the business side, fart bigger fluctuations also allows you to analyze, analyze and finished despise: “I knew it!.” Today we systematically author explain, how to deal with this problem.

First, the interpretation of the principle of large fluctuations in the index

You must first understand one thing: simply relying on data analysis, it is impossible to sort out 100% of all causes.

The real reason often need market research data analysis + + + industry research visits frontline work together to lock. Data Analysis on the Cause, a greater role is to locate the problem rather than solve problems. Want to answer the business logic behind the problem, market research much more useful.

Secondly, business fluctuations are normal, no more day is hell, so the routine will often encounter similar problems.

First priority to solve the problem is to determine priorities. In order to have a limited time and energy situation, select the appropriate method. If each issue must be thorough in-depth analysis, we will waste a lot of human effort, delay other work. Data analysis is also required to input-output ratio.

Finally, many times in fact, the real reason is difficult to parse out 100%, but for business, it does not need 100% of the Cause to action. For the business side, within a specified time, the method can adjust the indicator is very limited, as long as the amount of information sufficient to support operations on the line.

Sales index fluctuations, for example, there are a lot of factors affecting sales targets, see below:

Where: Customer attitude is very difficult to quantify, need research support; external competition, these require the industry boom industry research. The product did not sell well, in the end how many percent blame the user, what percentage of sales blame, blame what percentage of marketing, is difficult to peel off completely clear.

Last ABtest may have to test it. If every time fluctuations are so frustrating to find the cause, the company will be closed.

At the same time, ways to improve indicators is very limited:

If time is a day, just enough to be in store merchandise Duitou; if time is a week, then surgery can do sales training; if the time is in January, before considering driving sales activities; if time is a quarter is enough to adjust the product line layout; if time is a year to plan a new product.

So after determining the priorities of the problem, help businesses find coping strategies in a limited time, to improve the situation is the most important. Indicators can improve back is the ultimate goal, the ultimate goal is not to explore the truth of the universe. This is the enterprise for analysis and universities to do research fundamental difference.

Analysis Index fell when, in: three principles clear case + + focused enough on the line, the actual work will be able to cope with various scenes. Analysis of indicators not drill down the better, not to find the index, the better, not everything had on ABtest.

In the case of meet the business needs, try to rely on daily, weekly, monthly statements and other fixed resolve the fighting. Try to leave the truly meaningful thematic analysis of the topic, rather than in front of it every day drill to drill.

Second, determine the index fell priorities

1. The first step: Make sure data is not abnormal

Actually indicators because the data source problems, abnormalities caused by very, very much, to make specific reference may be buried point, ETL, the number of students in a variety of positions Tucao. Therefore, the first priority has encountered a problem make sure data is not wrong, not false police report.

2. Step two: Make sure the volatility index

This is to confirm the severity of the problem. Common indicators, such as sales / new number of users / active rates, etc., which have a range of volatility is based on historical experience can be classified as light in weight. In the real case, the general severe fluctuations are obvious signs.

For example, affected by the policy you want to stop some operations, the company’s initiative to shut down the business, Spring / eleven holidays and other factors. So for major changes in advance to set a good expected value, so look at the data, they will not Yijingyizha. In the case of severe than expected before doing the focus of follow-up.

3. The third step: to confirm the duration of index volatility

This is a question of priorities confirmation. Index fell / rose, there are usually three forms:

A one-time change: fluctuations occur only at some point in time. Behind the one-time changes are generally short-term / unexpected events, such as the system is down not lead to transactions, such as heavy rain suddenly one day, one day such a big promotion, and so on; periodic change: occurs periodically, such as weekly weekdays and weekends. General business development are cyclical, such as the retail industry, is circulating a weekly basis. Weekday and weekend is to have significant fluctuations; persistent changes: XX time from the start, always appear up / down trend. Behind the persistent changes often have deep-seated reasons, such as user demand transfer, trade and prosperity / withered, channel morphology changes and so on. These are difficult low-grade single enterprise, can only go along with the power, it will show a continuous change.

These three forms of different severity of the problem itself means.

If it is the decline in the indicator, the sustainability fell ¡Ý the falling fell ¡Ý Periodic fell. If it is a periodic decline, it is generally not a big strange. If it is a one-time fell, it is often so fast, and it is necessary to pay attention to the continuity of the incident. Sustained decline, especially if it is not improved, all the roads fall, the longer the problem, the longer the problem. Simply seeing the trend is not a strict explanation, and it is necessary to combine with the volatility. For example, when the period is falling, if the value of this cycle is significantly more powerful than the last cycle, it is possible to pay attention to other issues may be hidden behind periodic changes.

It is also a one-time fell. If the same incident, it will become more and more powerful, indicating that there are other problems hidden. Some seemingly a one-time events, the impact may continue to ferment, and finally evolved into sustainable decline. In the continuous decline, it may be a lot of decline in each period, but the growth line will find that the accumulation decline is particularly large, and the problem is higher.

Many business units will make mistakes here. For example, the sales volume decline in February, it is considered to be the influence of the Spring Festival, and there is no serious calculation. In fact, the recovery speed after this year’s Spring Festival is slower than the year.

At first glance, it is anxious to celebrate, and there is no time to calculate the cost-saving period and the weakness of the promotion. It is very possible to have a structural change. . These are all taste, unkreated to see data.

Classmates who do data analysis often make the opposite mistakes, have not judged that it is proneous, and I heard a lot of data to analyze the XX indicator fell. It is a waste of time, and there is no analysis assumption, and the result is empty as air.

Third, different lightweight is essential, different analysis methods

Heavy + urgent

Hurry and call the relevant departments and report it to the leadership.

If you really have a major problem in a short time, slowly write code analysis itself is the behavior of the delay. It is like a fire to fight 119, and it is still a serious analysis, and the fire is not big. At this time, if you have confirmed the data, there is no problem, you can’t report a fake police. Then grab the phone and quickly feedback to the relevant department.

2. Heavy + slow

For long-term major problems, we must find the crux in depth. At this time, you have to slow down, do a multi-angle multi-dimensional analysis, and analytical conclusions should be returned to market surveys, first-line visit, industry research to mutual verification, This is the way to solve the problem from the root. Otherwise, the analysis is completed, and the business does not feel easy to use, and finally it is white.

3. Light + urgent

Not serious but sudden problems, lock the problem point first. Because the problem itself is not serious, if it is checked, it is likely to be submerged in the calculation of the average. To the average, expose the truly problem of serious problems, so subsequent analysis can be deepened.

4. Light + slow

At this time, I don’t have to take a bunch of data. The problem itself is not serious. I haven’t contracted it, I can observe it again.

It should be noted that we have not talked about any professional analysis or technology so far, as long as it is staring at a sales curve or a number of users, do some simple year-on-year, the ring is enough. This is the legendary data sensitivity. Seeing that the indicator curve begins to fall, the sensitivity is high, and the trend will be interpreted first. This will continue to be shared in this.

With a priority judge, the next step can be reduced, and the analysis assumption can be established. We said before, it is difficult to make the reason for the reason, but when we narrow the problem range, we will easily find the source of the problem. Establish assumptions to help go to fake provens to verify and further approximate the reasons.

Fourth, narrow the suspected range, establish an analysis assumption

1. Angle 1: Event

Often major changes are accompanied by major events. It is therefore reflected in the data, often appears to be decided by the indicator response after the incident. This can see the impact of the event even if it does not require rigorous analysis, as long as it identifies the order of the incident on the indicator chart.

Event can be divided into internal events and external events. Strict distinction between the effects of two events is a very huge project analysis, and is likely to fundamentally not distinguish. However, the internal external events affect the method of indicators and effects are different.

External macro events we used to PEST analysis, P influence is generally fatal blow directly to the index hit collapse, the results do stop. EST affect more gradual, slow, foundation, structural, and therefore reflected in the index, it is more than Yindie.

Internal events can often change quickly in the short term indicators, so when reading index changes can be made according to this simple principle: relentless look at the policy, to find short-term changes in the internal and external factors to find a long-term transaction.

Within a certain period of time may be many simultaneous events, paying particular attention to three categories:

Starting event: The indicator has just started to fall, what happened; often the starting incident is the direct cause of the problem; the inflection point event: Whether there is any event in the process of indicators, make the problem more serious, or start Turn warm. The abduction point of care means that this is a means of improving the indicators; the endpoint event: When the XX event is over, the indicator will return to normal. Or after starting the XX event, the indicator falls. Two forms of endpoint events represent two ways to improve indicators: such as issues themselves, or actively attack the problem. We can label the events on the indicator changes trend, which can clearly see the relationship between events and indicators, thereby better shrinking the suspected range, refining analysis assumptions. Find those events that look like a core problem for in-depth research.

Interestingly, the business unit will do it, and many business units will be directly conclusions. For example, the weather is falling in the rain, and the price increases will rise. They want to say that this is this reason.

This is a very empiricistic approach that may ignore many structural questions. For classmates who do data analysis, there is still no more elegance for non-100,000 energetic problems. Since everyone is so recognized, we also save energy. It can be selected for some heavy + slow problems, in-depth analysis, and find structural changes. This seems to be valuable, and we understand business.

2. Angle 2: Region

Regional / channels that are distinguished from problems can also be reduced the assumption.

For example, it seems that the overall sales performance has fallen by 30%. Whether all stores, all areas are 30%, is it still rising, there is no more miserable, still rumor, the miserable / fall? (Store location, store manager, shop time, sales product line, stock customer group …) By classification, you can help us find problems more easily.

There is also a benefit to solve the problem to solve the problem. For example, in the case of overall sales performance, some provinces / store stores can do not decline, it is very likely that they have a unique means to respond.

For example, in the case of overall flow reduction, some channels still supply high-quality traffic, which is likely to mean new opportunities. Therefore, it is also possible to make a relatively good manifestation by comparing data in different regions, so future finding solutions have also been based, without having to take a head to find a way.

It should be noted that many classmates who do analysis are directly conclusions in this step, and the big title writes “Analysis of Sales Problems in Various Regions”, then gives an answer: because the ABC region is poor, so drag the market.

This analysis is very easy to attract: “I know” the vomiting. Because this only found the location where the problem occurred, there was no reason for the problem. It truly impact retail is the external inside. It is a human airfield. It does not find the root source of the real problem. It can’t say “the problem is XXX”.

3. Angle 3: group

Similar to the region, you can distinguish between different guest groups, see if there are differences in different guests. The product line can also be similar to the observation. Generally, traditional enterprises have no perfect CRM, lack of customer ID, so it is more incoming from the angle of the product line + region. Internet companies are more concerned about user groups, from the user’s perspective.

Here, there is a problem, for example, when I see the area, guest group, product line, there may be a difference between groups, which dimension will be cut from? In principle, it should be in accordance with the business characteristics of the industry. For example, traditional enterprises should first look at the area to see the product. Internet companies are habitually first watching users. If there is no business understanding, it will look at which group difference is large, from a significant problem.

As of this, there are a lot of words, and it is very easy to actually operate. Because all the above analysis can make a daily report based on an indicator daily and a sub-channel / divided user group. Difficulties are not to run a magical indicator, but to carefully interpret the trend of indicator curves, such as collecting and indicators related to data.

V. In-depth analysis within the range of conditions

After locking the problem, we will have a lot of problems. This indicator fell because:

New products don’t give sale, don’t sell channels, the lack of competition is too fierce …

Then, according to time, the conditions of resources are analyzed in-depth analysis. If time is tight, you can call the first line to confirm the problem directly, and then copy those good benchmarks directly. If time is worse, Internet companies can do Abtest and try to eliminate some interference factors. Traditional companies can do pilot, lose some template shop to see the effect.

In short, there is a clear assumption that the speed of verification is very fast, and the speed of finding excellent benchmarks is also very fast. All the work in front is paving this way. The basic work is solid, and the more easily do it.

Sixth, why do you want to say so much?

The length of this article broke the progress of Mr. Chen. Classmates who are familiar with Mr. Chen know that Teacher Chen has always been too lazy to write long. Why is this seemingly simple problem?

Because: Interpretation indicators are the personal business of data analysts, and these years are excessive superstition techniques, defective, and addictive alpha’s big dogs. . Whether it is job search or a lot of things. For example, many students write daily on February, write: due to the Spring Festival factors, sales are sluggish. Then February weekly: due to the Spring Festival factor, sales is sluggish. The monthly reporting of February is also affected by the Spring Festival factor and sales is sluggish. This is the legendary three flowers to parents. The same sentence is three times, there is no interpretation.

For example, seeing a day of data fell, anxiously scratching, the result is a few days ago, see if it is a periodic change, it doesn’t know; Continue to stare at the indicators in front of you; The absolute value of the area, and then, it is considered to fall by 5% and 50% of the same type of question. Write on the report: “All channels are falling” will be full; for example, I only know that the data is urgent, even the business is doing something I don’t know; for example, it is expected that there is a manual intelligence algorithm and there are several factors.

But this can’t blame the students, because I found out with many students in-depth chat: their supervisor has never taught these …

No wonder, many leaders of “Data Analysis” team leaders, in fact, do Hadoop, do Bi bish, write SQL origin, it is only the ability to have no analysis capabilities. So this article is particularly embarrassed, I hope everyone can practice more, Daily, Weekly, Millennium Daily is the skeleton of data analysis, the bones are soft, and people will directly waste.

As usual, such a long jealousy is welfare, Teacher Chen has summarized a thinking map on the upper side of the Laramis, the left side of the picture is confirmed, and the right side is judged. Everyone is in the interview, it is best to write the thinking map while saying that the interviewer is a spike, which is more reasons than those who will, and then have been discovered: nothing, the poor is not enough.