Five pitfalls of data analysis projects, don't step on them!

Data analysis is now a skill that most companies need. Let us follow the author to avoid the five pitfalls of data analysis projects.

In the previous article [Five questions to see if you can make a good data analysis project], everyone has a certain understanding of data analysis projects. Today, we will take a specific scenario to analyze how to make the project outstanding. The five questions correspond to five major pitfalls. Think carefully and don't step into the pit.

Scene restoration:

The B2B business development team of an Internet company mainly contacts potential customers through telephone sales. The outbound call list management is chaotic, with only two fields: customer company name and contact number. The sales success rate is extremely low, and the team management is chaotic. It only records the transaction amount, but does not record the reasons for the failure to close the deal, nor does it have follow-up records. The performance is poor, the team turnover is serious, and the leaders are very anxious.

Question 1 (multiple choice)

You are a data analyst for this company. At this time, you will:

A. Carefully analyze the reasons for the low success rate in the monthly report and write 20 pages of suggestions for improvement

B. Monthly reports only list numbers, waiting for them to come to you to discuss cooperation

(The question is simple, think about it for a second)

After the education in the previous article, everyone chose B. Yes, the problem in this scenario is caused by chaotic business management, and it would be a pity if data can help.

If the business itself is not aware of the pain, and the data analyst is just an outsider blablabla, no one will pay attention. So don't do this thankless job. Even if you do it, and people listen to your suggestions and improve efficiency, the credit will be all the business's, and your analysis has nothing to do with it. How can you prove that they listened? People will say: "I thought of it a long time ago" or "I know it even if you don't tell me."

So the best strategy is to wait for them to come to you, set up a project, such as setting up a project team called "Sales Performance Improvement Project", and send an email to inform the bosses to officially start the work. Ok, let's go!

Question 2 (True or False)

Now the team leader has found you and decided to start the project. You set the project goal as "increasing sales success rate". Is this right or wrong?

A. Yes

B. Wrong

(The question is simple, think about it for a second)

This is a common problem for many students who do analysis: they treat the ultimate goal as the immediate goal. Or they don’t know what the business goal is at all. They just say based on their feelings: “We are an e-commerce company, so we need to increase GMV”, “We are doing growth, so we need to calculate DAU”…

Please note: In the case of chaotic management, missing data, and team desertion, it is completely unrealistic to expect to boost sales by writing 200 lines of code and producing a PPT. So you need to have a serious talk with the business leader to find out if there are any secondary goals to achieve in addition to improving the sales success rate.

for example:

• It takes time to prove and improve results. Buy time
• Demonstrate the inability to make improvements at present and lower the KPI
• Explore ways to stabilize the team and stabilize the morale of the troops
• Find some successful benchmarks and summarize the experience

In fact, when business leaders encounter poor performance, the first thing they think of is to get resources, the second thing they think of is to adjust KPI, and the third thing they think of is to find cases. Others really don’t want to listen to you saying: "The activity rate is low, we need to improve it!" Therefore, it is very important to sort out goals and determine primary and secondary goals. So the answer to this question is B.

Question 3 (multiple choice)

Now confirm the first-level goal: improve sales success rate; second-level goal: find successful sales experience. Someone immediately jumped out and said: "You have never done sales, how can you analyze!!!" Question: What should I do?

A. Analyze the best sales method through data

B. Admit that you don’t understand

(The question is simple, think about it for a second)

This is a common mistake made by many students who do analysis*2: they expect data to directly calculate the best solution. Students with this idea are often beaten down by people who ask: "How many orders have you completed?" or "If you can do it, go for it."

When it comes to the question of “how to do it”, the role of data analysis is not to calculate the best solution. Because every successful case must have unique advantages that cannot be replicated, such as sales. Some people are born with good words and good appearances, and you can't avoid these.

The role of data analysis is to analyze specific cases and distinguish between replicable parts and non-replicable parts. The replicable parts are accumulated as experience, and the non-replicable features are extracted to find more similar features in the future.

For example, if we find that local pretty girls have a high sales turnover, then we can let each city recruit pretty girls locally; if there is a special time and a special action needs to be done, then let others copy this operation. Data analysis is not good at selling goods, but at summarizing experience and finding features. For this question, choose B.

Question 4 (Picture description)

The basic salary is 1,300 per month, and the commission for each order is 250. Which of the following two tiers is better?

A. Figure 1

B. Figure 2

(The question is a little complicated, think about it for a minute)

This question is not about how to stratify, but about a basic idea: find classification standards based on business needs. For example, in this example, there is a big problem: serious team loss. There may be statistical differences between 10, 8, and 4 orders per person, but there is no difference in business. Whether a salesperson has 10 or 8 orders, he will not earn enough for a month's living expenses, and he will still run away. But 25 orders can make him earn 25*250+1300=7550, which is considerable for the phone guy.

This is a core difference between data analysis and data mining. We build data models in order to simulate the real situation with a high probability, so we can process some data and virtually fill in a batch of data, anyway, for the overall effect.

Data analysis has a unique effect: it can guide the business department to create a situation that does not exist now. For example, the business believes that the team is stable only when the backbone personnel who can earn 7,500 yuan a month account for at least 20%. Then the existing salary system, operating system, and recruitment process can be changed, which breaks the status quo. Therefore, when doing data analysis, we often pay more attention to the guiding significance to the business. When looking for standards, we need to find standards that meet business needs. For this question, choose B.

Question 5 (Picture description)

Still using the above figure, if we use the layering of Figure B, can we identify the first layer as the business benchmark and conduct in-depth research:

A. Can

B. No

(The question is a little complicated, think about it for a minute)

A: No. Because we don’t know whether these people’s performance is good continuously or occasionally. As shown in the following figure, there are four different trends for the winners selected within a month:

Note: Generally, for the convenience of data collection, we will not collect all the data at once. Therefore, the project is often promoted from individual cases to universal cases, from a single month to a whole year, and the results are output in steps. On the one hand, it can improve efficiency and prevent the project from being delayed for a long time without output; on the other hand, short-term emergencies are easier to interpret. If you want to know whether you have really found a pattern, you have to promote it from the short term to the long term.

For example, in this example, we can first select quasi-benchmarks from the performance of a month, and then look at their stability. This will help us interpret richer business meanings and establish the next step of analysis hypothesis. With the analysis hypothesis, we can continue to go deeper and do a deeper analysis.

This question is one of the questions that Teacher Chen used for internal training. The original question did not have so many hints, just six fields:

1. Salesperson ID

2. Customer's Chinese name

3. Customer contact number

4. Is the transaction completed?

5. Transaction time

6. Transaction amount

Many students were confused after reading this article: "What the hell is this analysis? There is nothing." But it really reflects the current situation of many companies. They are called "Internet companies", but their actual management is even more backward than that of traditional companies.

From the perspective of the problem-solving steps, as long as the goal is set reasonably and one works step by step, many useful conclusions can still be generated. Even if it is found in the end that sales are very random, it is still a great support for the business. At least in the future, you can recruit as many people as possible and use the human sea tactics. If you can summarize a set of benchmarking words, it will of course be a more ideal result.

Moreover, not all data is unavailable. For example, if we really select a benchmark, his speech, the time he contacted the customer, and the number of follow-up calls can be recorded and supplemented. Based on these analysis results, we can further promote system upgrades. With a better system, the business can improve efficiency and the data can provide more analysis materials, and everyone will benefit.

Now that we have mentioned data collection, the question arises: where do we start?

Question 6 (Sequencing)

After the first phase of analysis, the business recognized the practice of copying benchmarks and wanted to further improve the data. Then the following data all need system support, and the priority order is:

A. Use crawlers to crawl customer details

B. Enter the salesperson's resume into the system

C. Record the salesperson's operations on CRM

D. Complete the customer information form for the salesperson to fill in

(The question is a little complicated, think about it for a minute)

How many programmers put A first? Please raise your hands, haha. Please note that although options ABCD all require a system, the difficulty of obtaining the data itself, the degree of business support required, and the usefulness are different:

• Resume: The most formatted, and not handled by salespeople, so it is least likely to be contaminated.
• CRM data: directly record operations without going through sales.
• Information sheet: Needs to go through the salesperson and has supporting management measures.
• Crawler: It seems possible, but it is difficult to ensure the stability of data quality (specifically in this case, B2B business data may be very scattered, unlike e-commerce sales and comments which can be crawled in a centralized manner).

Therefore, from easy to difficult, the order is B≥C≥D≥A. This example is just to remind everyone: Don't be obsessed with technology just because we are in technology. Many technical tools require supporting systems to ensure that the data is not polluted. At this time, we must work closely with the business and consider the availability and convenience of the technology. Some guys are too obsessed with data, which will make the business process extremely complicated and the data table have too many fields. As a result, the sales will deal with it casually, and in the end they are the ones who are pitted.

summary

In the previous article, we listed the five keys to making an excellent data analysis project. In this article, we summarize the five pitfalls of making an excellent data analysis project:

1. No project, no consensus, only data, no implementation
2. Directly taking the highest goal as the project goal will not be achieved
3. Applying the model mechanically results in being stuck due to insufficient data
4. It's too complicated to do it at one time, and it takes a long time to get results.
5. There is no gradual iteration, and the results are only a PPT

The key to avoiding these pitfalls is to refuse to work in isolation, combine business needs, and evolve from low to high. In this process, a lot of demand insight, communication and collaboration are needed, so that the business test analysis results can be finally separated from the false and the true can be promoted to evolve the business.

This is why things like "Titanic", "Boston House Prices", "US Credit Cards", and "Maoyan Movie Reviews" are not considered projects. These so-called Internet celebrity projects are just running a data table.

Moreover, many self-learners do not run this data table by themselves, and the codes are copied from the Internet. In addition to the ability to type (and read English words), there is no communication, demand analysis, plan formulation, result testing, and iterative upgrade process. Although these Internet celebrity projects are all named "artificial intelligence" and "21 days to change careers and earn a million dollars a year", they are just for self-entertainment.

As the saying goes: A brave general must be born in the ranks, and a prime minister must be born in the prefectures and counties. A good data analyst does not start by manipulating models, but can read the problems of the company from the data details; can design feasible methods based on even the simplest data foundation to help the business upgrade from low-end to high-end. This is the real role of a good data analyst.

However, some students will say: Teacher, this scenario is that the business has a pain point and comes to us to solve it. But there is another scenario: the business itself does not know what it wants? Then ask us "You need to interpret something that we don't know and is very important." At this time, it is vague and chaotic. What should I do?

Author: Down-to-earth Teacher Chen; WeChat public account: Down-to-earth Teacher Chen