Data analysis report, how to write the [Suggestions] section?

How to make effective strategic suggestions in data analysis? This article uses a simple example to show how to use data thinking to dig deep into the causes behind the problem and put forward specific and actionable suggestions.

Don’t just report numbers! Think strategically! Make feasible suggestions!

Many students who work with data have been called out by their leaders and colleagues. However, what is strategic thinking? Often when students hear this term, they can’t wait to pull out their magic weapon, such as the McKinsey Method, or search online for articles such as “underlying thinking”, “core logic”, and “analysis framework”.

As a result, I didn't remember anything except the vague terms like "fission", "pain point" and "disruption". Next time I write a report, I will continue to compare year-on-year, month-on-month and three-year comparisons. If they are too low, I will try to make them higher...

What to do?!

Let’s take a simple example. Today, Li Qianying, a young girl from HR, was scolded by her boss and cried because, as an HR, her attendance sheet this week looked like this:

SO, as data analysts, what suggestions do you have after seeing this?

1. Lack of strategic performance

Soon, the four students who were doing data analysis gave their answers.

Student 1's answer:

• There are 22 working days in this month, 11 working days are late, and the lateness rate is 50%.
• The second week was the most late, with 4 days of lateness and a lateness rate of 80%.
• The third week had the least number of late arrivals, with 1 late day and a lateness rate of 20%.

Student 2's answer:

• There are too many times of being late, so it is recommended not to be late.
• It is especially recommended not to be late on Monday.

Student 3's answer:

• The data source is...
• The modeling process is...
• After regression model analysis, it is predicted that there will be 12 days of delay next month.
• Recommended to reduce lateness.

Student 4, has not given an answer yet:

He was looking for "Employee Lateness Analysis Model" on the Internet. He searched for a whole morning but couldn't find it, but he joined five data analysis discussion groups, and each group asked:

• Any data analysis experts out there?
• Are there any data analysts in the HR industry?
• Are there any books on HR analysis, preferably in PDF format?
• Urgent! Can pay! Waiting online!

Question: Which of the above four students can pass?

2. The core problem

Obviously, none of the above four are qualified!

They failed not only because they said empty words and nonsense, but also because they all made the same mistake: they only discussed numbers and were divorced from the process.

As an HR manager, the advice I want to hear is:

• Tip 1: Get out early.
• Suggestion 2: Take a taxi when you need to. Why save the money?
• Suggestion 3: If you make a mistake, just admit your mistake. There’s no point in crying!

As Li Qianying's younger sister, the advice I want to hear is:

• Suggestion 1: Reduce the workload for Comrade Li Qianying.
• Suggestion 2: Since Comrade Li Qianying lives too far away, it is recommended to grant a special exception for a few more days.
• Suggestion 3: Comrade Li Qianying worked too hard last month, so it is recommended that she be exempted from punishment.

See the difference? No matter the business leaders or subordinates, they don’t care about the specific numbers, let alone the model used to get the numbers. What they care about is what can be done. What they do must have a basis, and it’s even better if it can convince people! The so-called suggestion is a specific action that the business department can take. This action is closely related to the business workflow. It must be able to achieve a result that everyone agrees on.

So when deriving suggestions, don't just get hung up on numbers, especially on the "result numbers" of questions like this. If you just get hung up on the results, it will turn into a childish argument like "You say I'm lazy, I say I'm not lazy." You have to find a way to get into the process of the problem before you can find the answer.

3. Ideas for solving the problem

When we relate it to the specific process, we can find that data is of great help in quantifying the process and locking in the problem.

For example, the simplest suggestion is "get out early", which sounds reasonable, but in fact has at least three loopholes:

1. It is not clear what time to go out early, 6 o'clock? 7 o'clock? 8 o'clock? Just saying "go out early" is the same as not saying it at all, it needs to be quantified.
2. It is unclear whether there are any special reasons. It is very likely that the girl will burst into tears: "Is it normal that I worked overtime until midnight the day before and couldn't get up the next day!!! Isn't it normal to ask for formal wear and put on makeup before going out!!! You want me to be busy and blame me, wuwuwu"... Quantification without distinguishing specific scenarios is simply unconvincing.
3. I don't know if the special reason is true or not. Who knows if she is really busy or went out partying until midnight the day before yesterday. What's more entangled is that the data that can directly deduce the answer may be missing. You are not her boyfriend, how do you know whether she went out partying or worked overtime the night before yesterday.

In the absence of direct evidence, we have to proceed step by step:

1. First clean up the available data and establish a basic analysis framework
2. Let’s look at how to explore specific scenarios and eliminate abnormal situations

Only in this way can we be well-founded and convince others with reason.

4. Order of answering questions

The first step is to find out what data is available.

When it comes to commuting, we don’t actually need that much privacy information:

The second step is to establish a basic analysis framework.

In the basic analysis framework, various unexpected situations and special scenarios are not considered, and only the most basic data logic of the business is considered.

For example, when it comes to commuting, as long as you select the starting point (the neighborhood where Li Qianying lives) and the end point (the company), you can open the Amap and see:

• How far away
• How long does it take to take the subway?
• How much does the ride cost and how long does it take?

With this basic information, we can determine whether the distance is really too far, thus eliminating many excuses/suspicions (as shown below):

The third step is to discuss special cases that can be quantified.

Don't gossip about a pretty girl as soon as you see her. It will only lead to arguments and do nothing good. First, collect data on the obvious issues, such as overtime and taxis. This way you can see if there is really an unequal division of labor and people are being wronged. Secondly, it can also stop people from making excuses (if they really didn't work overtime).

Step 4: Derive recommendations.

With the above preparation, the derivation suggestions can be justified and very specific (as shown below):

5. Back to the real work

Of course, the above is just a funny example, but it clearly reflects the problem in reality:

• Business departments often think in an egocentric way, and the suggestions they make are all in their own interest/what they want to express. They are too lazy to consider the facts, let alone make detailed classifications.
• Data departments often get caught up in number games, focusing too much on number calculations and ignoring business processes, and end up just talking about numbers and stopping at numbers.

This is not conducive to drawing correct conclusions and suggestions. The best approach is to start from the process, advance layer by layer, and build a logic tree. However, the concept of algorithm model has been widely circulated in the past two years, which has made business parties and data think that as long as LR, CNN, and XGBOOST are pushed up, the computer can speak: "Li Qianying, I am the omniscient AlphaGo, and you are late this month. It's your own fault"... So more jokes were caused.

Of course, all of this is based on a basic premise: you have to be able to distinguish whether you are looking at result data or process data. A student once asked Teacher Chen: "Teacher, how can I improve my strategic thinking ability? You see, we are doing everything well now, but the conversion rate is not going up. Why?"

Answer: You are now Li Qianying, pouting her pink lips and looking aggrieved: "I go to work very actively every day, but why am I always late?"... If you want to find the answer, it is useless to just worry about the result, you have to go deeper into the process.

So don't just pile up numbers or put a bunch of circles and boxes. The right way is to analyze specific problems specifically.

Author: Down-to-earth Teacher Chen

WeChat public account: Down-to-earth Teacher Chen (ID: gh_abf29df6ada8)