What kind of data analysis can be considered "advanced"?

As data analysts, we run numbers and do analysis every day, but the results of our analysis are always regarded as not very high-level by others. So, what kind of data analysis can be considered high-level? The author shares several tips for doing high-level analysis. Let's take a look.

We often complain that we are just running numbers at work every day, but the things we produce are always criticized for being "lacking in depth". They are all statistical calculations, so why do they need to be "sophisticated"? Today, through a real case of mine, I will see how data analysis is evaluated as "sophisticated".

Problem scenario: The marketing department leader of a traditional enterprise's e-commerce department found a data analyst and took the current inventory data and weekly sales data report of the goods, asking for help in analysis. The business side said: There is no clear analysis requirement, so do an exploratory analysis.

Q: How to do this analysis?

01 Analysis of negative points

• Yesterday's inventory 120
• Today's stock 220
• The month-on-month growth was 83%. It is recommended to lower the

Comments: I am too lazy to comment, you can just make your own comments.

02 General Analysis

• Product A has a weekly sales volume of 100 and 400 units in stock, which can be used for another 4 weeks
• The replenishment cycle of product A is 3 weeks, which can be further observed
• Product B has a weekly sales volume of 100 and 200 units in stock, which can be used for another 2 weeks
• The replenishment cycle of product B is 2 weeks. It is recommended to start replenishing this week.

Comments: Knowing that the purchase, sales and inventory data must be viewed together, and knowing how to consider the replenishment cycle, this at least passes.

03 A bit of high-level analysis

• Product A has a seasonal pattern and is currently at the end of its cycle. It is estimated that the inventory will be enough for 3 weeks. Restocking is not recommended.
• Product B has reached the second half of its life cycle, with weekly sales decreasing. It should be shipped faster.
• Product C is a non-seasonal standard product, with weekly sales of 100 and inventory of 200. It can be used for another 2 weeks and can be replenished this week.

Comment: The biggest difference between doing data analysis and doing homework is that "there are no standard test questions in actual work, only problems to be solved". Therefore, it does not mean that the business department can only look at a piece of data! In addition, the product life cycle, natural cycle, and whether sales are stable do not need to be told by the business department. Data analysts can discover them through data themselves. Therefore, if you want to find problems, you don't need to wait for the business to teach you, you can take the initiative to attack.

From the perspectives of business common sense and data performance, we can find the rules behind the data. Seasonal products: For example, products that keep you warm in winter and cool you down in summer have good sales in season, and if the winter is not so cold/the summer is not so hot this year, the sales will be discounted, as shown in the data below:

Life cycle products: such as electronic products and new clothing of each season, are sold at the highest price and have the best sales volume after they are launched on the market, but become less valuable as time goes by. When the season changes and new styles are released, they can only be cleared by reducing the price. This is reflected in the data as shown in the following figure:

Stable sales products: such as rice, noodles, oil, eggs, paper towels, shampoo, shower gel, which are needed in daily life. They may fluctuate with the overall sales volume (such as store traffic and website traffic) every week, but generally they will not fluctuate greatly. The data is shown in the following figure:

These can be discovered in daily data and accumulated experience without having to wait for the business side to nag you before taking action (it is very likely that the business side knows these differences themselves, so they can do the calculations themselves and won’t ask again).

04 There are 2 small details to deal with here

Detail 1: If the product is not frequently replenished but a new product, how to predict the future trend? The simplest way is to classify the product according to the product brand and price range, and use the performance of products of the same level and price range in the past as a reference (as shown in the figure below).

Detail 2: If the product is sold periodically and is affected by external factors, such as rain and no business, how to evaluate the trend? The simplest way is to look at the duration of the influencing factors and make adjustments directly in the future duration based on past impacts (as shown below)

Of course, more complex forecasting methods can be used. Interestingly, this kind of forecasting will change the behavior of the business side, so it is not necessary to have a particularly accurate forecast. For example, a binary forecast: whether the goods can be sold out within 2 weeks/cannot be sold out can already prompt the business side to decide whether to clear the goods. For more knowledge about forecasting, click [Don’t just blame the data analyst for inaccurate forecasts!! ]

This brings us to a key question: how much accuracy is acceptable? The second biggest difference between data analysis work in an enterprise and homework is that "real work requires communication with various departments, not just finishing the test and waiting for the correct answer."

Especially in the scenario of this case: the business department itself is confused and has no clear purpose! At this time, there is no need to force the accuracy of the calculation results. Instead, after reaching a preliminary conclusion, a meeting can be held to communicate, remind risks, and understand the inside story. Unless you encounter special products such as imported fruits and seafood in fresh food, and vaccines stored in refrigerators in medical supplies, which have extremely high inventory costs and short shelf life.

Generally, products have a certain turnover space, so the ultimate goal of inventory control is not to clear 100% of the inventory at the right time, but to control the inventory within a safe range. So instead of worrying about it, it is better to see whether the current inventory pressure of the marketing department can be tolerated. So, is there a more advanced analysis?

05 More advanced analysis

Let me ask a simple question: As the operation center, why does the marketing department have to ask for such core data as sales and inventory? The fact that the marketing department came to ask is a big problem in itself. A highly sensitive data analyst should have felt that something was wrong before running the numbers. There must be something wrong when things are abnormal, so communication in advance is very important! There may be many specific situations, but there is a core problem that must be solved first: Does the business side really not understand, or is it pretending not to understand?

Don't laugh! Don't assume that the core data business side really knows everything. It's possible that the business development was going smoothly before, so everyone is used to making decisions based on their own ideas; it's also possible that they have benefited from the industry dividends before, and they are really soaring. In short, if you really don't understand, you need to:

• Establish a data monitoring system
• Diagnose existing problems
• Provide suggestions to deal with the current problem first

Train relevant personnel and establish a long-term supervision mechanism to advance the work of commodity operation from the original state to the state of digital management. If it is fake, it depends on whether it is a lack of manpower or a lack of helpers!

If you are short of manpower and need help cleaning up the data, remember:

1. Previous Products
2. Previous Products
3. Previous Products

Data products are the best solution to solve the shortage of data processing personnel, not hiring a few more Excel boys to fill the gap. The existence of Excel boys/Sql boys is a burden on data work. It wastes costs and cannot reflect performance. If they have strong business capabilities, let them drag and drop on the data mart; if they have weak business capabilities, just fix the dashboard and then teach them how to read.

If you are in need of help, it depends on what you need help with:

• Some products are not performing well, so I want to clear out the inventory in advance
• I want to launch some new products and use the previous ones as benchmarks and support.
• I want to restock a few products, but I'm afraid the boss won't approve it. I want someone to speak up.
• …

These purposes are not something that the data analysts can directly guess without the business side expressing their opinions. So if the business side wants the data analysts to help speak up, they should honestly state their intentions and everyone can discuss it together.

There are indeed business people who like to be sarcastic and don’t say what they think. Instead, they ask data analysts to guess. If they can’t guess correctly, they say, “It seems that the analysis is not advanced enough and does not meet business expectations.” This practice is just a preparation for passing the buck later. It’s better to ignore it.

06 Summary

Talking about analytical depth in an enterprise is completely different from talking about depth in scientific research at school. Only in the field of scientific research can we talk about high-precision and cutting-edge technology. The more complex, forward-looking and advanced the methods used, the better. The purpose of an enterprise is to make money, and making money requires efficiency, coordination and practicality. Therefore, the order of depth is:

1. Combined with business situation ≥ simply calculate the numbers
2. Discover business problems ≥ simply present results
3. Achieve business goals ≥ simply count the numbers

In this process, you need to have a basic understanding of the business, communicate closely with the business, and tailor your analysis to specific goals. This is the most effective way to improve the level of analysis.

Author: Down-to-earth Teacher Chen

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