Introduction: Have you ever found learning data analysis methods painful? In this article, the author summarizes an easy-to-learn and easy-to-use data analysis methodology in simple and understandable language, allowing beginners to quickly grasp the most core and commonly used points of data analysis methods, meeting at least 90% of daily needs. __ Learning is painful for most people, especially when looking at thick professional books and various difficult-to-understand terms without explanation, which intensifies the pain. However, some books or articles can explain complex theories in a very colloquial and conversational way, making it easy for readers to understand without effort. These contents are truly a blessing for readers. After all, we have talked about user thinking in internet thinking for so long; creators of educational and training content should also change and speak from the reader's perspective. This article discusses data analysis methods. Based on the author's contact and understanding of many enterprises, although most companies now pay more attention to data, there are still quite a few enterprises and practitioners who have not yet figured out the tricks of data analysis, not knowing how to analyze their data, and hoping to get help from professionals. · Data analysis methods are not mysterious at all**** The author also felt pain when learning data analysis methods before, reading many books with lots of content, but it was hard to remember and apply. Later, joining Yonghong Technology to provide data analysis systems for many enterprises, through a large number of project practices, gradually became an entry-level expert. A good methodology should be easy to learn and use. Now, this article tries to use the simplest and most understandable language to let beginners understand and master the most core and commonly used points of data analysis methods after reading, meeting at least 90% of daily needs. To achieve this, it is necessary to distill the broad and profound data analysis methods into 3 points that people can remember, not 30 points, and condense them into the length of an article, not the thickness of a book. 1. Data is divided into two types: dimensions and measures. Analysis is the combination of dimensions and measures.**** Below is a simple example of consumer shopping data. Regardless of whether this data table is stored in Excel or a database, only focus on the data itself. The data items (or fields) involved in the table include "Order ID", "User ID", "Region", "Age", "Order Amount", "Order Product", and "Order Time". What are the differences among these data items? Overall, data is divided into two types: one is called dimension, and the other is called measure (or indicator). In the above example, "Order Amount" is a measure, and the rest are dimensions. It can be seen that measures are specific quantitative values used for calculation, while dimensions are various attribute information describing things. When we do data analysis, in the final analysis, we are constantly combining various dimensions and measures, such as the total order amount in the Beijing region, the average order amount of users aged 21 to 30; or performing mathematical formula calculations on dimensions and measures alone, such as the total order amount of all orders, the number of users (distinct count of User ID), etc. From the data type perspective, measures are all numeric, but numeric values are not necessarily measures. For example, Order ID, although numeric, is a dimension, not a measure, while time and text data are dimensions. One point that needs special attention is that dimensions and measures can be converted. For example, to look at the average of "Age", here "Age" is a measure; to look at the order situation of 19-year-old users, here "Age" is a dimension. For a data item, whether it is a dimension or a measure depends on the user's needs, much like the quantum effect, where the state is determined only after the need is determined. In addition, dimensions can derive new dimensions and measures. For example, using the "Region" dimension to derive a larger region dimension, where "Beijing" and "Tianjin" correspond to "North China Region", or using the "Age" dimension to derive an age range dimension, where 20-29 years old = "Young Adults", 30-39 years old = "Middle-aged", 40-49 years old = "Senior Middle-aged". Another example is the average age mentioned above, which uses the "Age" dimension to derive a measure. Measures can also derive new dimensions and measures. For example, using the "Order Amount" measure to derive an amount range dimension, where below 100 yuan corresponds to "Small Orders", above 500 yuan corresponds to "Large Orders", etc. Another example is subtracting the "Revenue" measure and the "Cost" measure to obtain a "Profit" measure. 2. Use comparison for judgment**** Now raise a question: Company A's revenue this year is 80 million yuan. Is it high or low? Looking at this question, everyone should feel unable to judge because there is no reference, i.e., no comparison. Therefore, when you get a piece of data, to judge whether it is good or bad, high or low, you must compare. First, Company A can compare with itself. If the revenue two years ago was 20 million, and last year was 40 million, then this year's 80 million is very good. If last year's revenue was 100 million, and this year is 80 million, it is bad. This is called vertical comparison. Second, Company A can also compare with others. If several competing companies in the same industry have revenues of several hundred million this year, then Company A's 80 million is not ideal. This is called horizontal comparison. Third, Company A can also compare different dimensions and measures. For example, if competitors all do the national market, but Company A only does the Shandong market. If Company A's revenue in the Shandong market is higher than competitors' revenue in the Shandong market, then in the local region, Company A is doing better, but looking at the whole country, Company A's performance is limited. For example, if competitors have been doing it for over a decade, while Company A has only been doing it for four or five years, then Company A is doing well, but if competitors of similar establishment time have already exceeded 100 million, then Company A is not doing well enough. This is called comprehensive comparison. A child scored 95 points in an exam, and the parents were happy because they knew the full score was 100, so there was a reference. The most recent exam scored 80 points, and the parents were angry because the previous 95 points became a new reference. Later, upon asking, they found that the test was difficult, and the child was already first in the class, so they turned anger into joy. Here, other children became the reference (sacrifice). Different references for comparison lead to different conclusions. To avoid one-sided and non-objective conclusions, comprehensive comparison should be used as much as possible. 3. Use segmentation to find causes This year's profit declined, and the boss is very angry, ordering to find the cause and capture the "suspect". How to find the cause? Note that it is to find the cause, not the excuse. Many people often do not know how to find the cause, and in the end, they give excuses. First, look at an example of a cause conclusion: "Because the sales of washing machines in the South China region declined in the fourth quarter, leading to the decline in profit this year." Let's analyze what characteristics this cause has. We will find that this cause is composed of three dimensions (time, region, product) and one measure (sales). Thus, we can know that finding and locating the cause of a problem is essentially answering which measures under which dimensions declined or rose, leading to the problem. This is doing segmentation. We can segment by dimensions. How many dimensions there are, there are that many directions for segmentation. For example, check whether all months of last year declined, or only a few months declined. If the latter, then the data range for searching can be narrowed. After focusing on these months, you can then see which regions declined, further segmenting. The order of starting dimensions does not matter much, and the dimensions involved in the cause of the problem cannot be predicted, so you can start segmentation from any dimension as an entry point. If the problematic indicator has related leading indicators, to further dig into the cause, after segmentation, you also need to look at different measures. For example, the above cause conclusion is "Because the sales of washing machines in the South China region declined in the fourth quarter, leading to the decline in profit this year." The problem is "profit" but the cause is "sales", because profit is derived from other measures. Segmentation is endless. How far should you segment? The answer is: until you reach an operable interval. For example, if you segment to "Profit declined in the fourth quarter, but not in other quarters," there is still no solution to the problem. You must segment to which time period, which region, which product line, until you reach a specific responsible person, to be operable. It should be noted that in real situations, problems often have not just one cause but multiple causes combined. Our company, Yonghong Technology, promotes a one-stop big data analysis platform software. Why does it provide two interaction operations, "Zoom" and "Brush"? It is to meet the two scenarios of "comparison" and "segmentation". For example, as shown in the figure below, the left chart is the revenue and gross profit comparison of various products, and the right chart is the profit trend of various categories. Now the user wants to focus on the three products under the "Flower Tea" category and see their profits. At this time, the user can use the "Zoom" function to circle the three bars representing these three products, click the "Zoom" button, and then the left chart will only show these three products, while the right profit trend will show the total profit trend of these three products. This is doing "segmentation". Some may ask, this effect is very similar to filtering. Why not put some filters next to it? Filters can be used, but in reality, when we find a problem on a chart, it may not be easy to find the corresponding filter conditions, especially for scatter plots. Therefore, selecting directly on the chart is very convenient and efficient. Another example: the figure below is a product profit trend analysis. The user finds that starting from July 2009, profit has declined for four consecutive months (as shown in the red box), and the user wants to know why. At this time, the user can use the "Brush" function to select the points of these four months on the trend chart, click the "Brush" button, and other charts on the same report page will be dimmed, then highlight the proportion of the selected July to October in this chart. So in the figure below, the short green bars highlighted in the left chart are the sales revenue of these products in these four months. Unlike "Zoom", "Brush" is convenient for users to compare local data with overall data. Because in the above example, simply looking at which products have low absolute sales revenue in these four months does not explain anything; some products naturally sell less. You must look at which products performed poorly relatively in these four months. First judge whether the data is good or bad, then analyze the cause. The chain of data analysis is basically complete. · How to view high-sounding things like machine learning and data mining**** When to touch high-sounding things like machine learning and data mining? In one sentence, first master the above data analysis methods to the point of ease, then do those high-sounding things. Do not be superstitious about complex algorithms. Many big shots in internal enterprise data analysis often deeply understand the business and use ordinary calculation methods to complete brilliant and practical analysis processes. When are machine learning and data mining used? Simply put, when there are so many data items that the human eye cannot see them all. If there are only about a dozen data items, and you can take each one out and make a chart to see the clues, then you probably do not need mining algorithms. If there are hundreds of data items, and you want to see which data items have the greatest impact on a certain data item, the human eye cannot see them all, so mining algorithms are more appropriate. Wang Tong, Vice President of Yonghong Technology Personal profile: Master of Engineering from Beihang University, with 8 years of product sales and marketing experience in the business intelligence field. Previously worked at Oracle and IBM, holding important positions in consulting and sales. He has successfully promoted the implementation of multiple large projects and accumulated rich experience in e-commerce, government, finance, internet, and other industries. Wang Tong is currently mainly responsible for product sales and channel expansion, and has provided comprehensive data visualization analysis solutions for hundreds of enterprise users, including star companies in e-commerce such as BabyTree, as well as traditional giants like China Mobile. -END- The best FMCG distributor learning platform in China Focusing on providing professional, practical, and actionable tutorials for enterprises and distributors Committed to helping Chinese FMCG distributors grow rapidly The most professional and practical knowledge base in the FMCG industry Reply with the red number below to get the corresponding content Reply with number 1 to view the complete knowledge base | 001 Excellent article selection | 002 Distributor market operation | 003 Terminal visit management | 004 Sales supervisor skills | 005 Sales improvement techniques | 006 Channel expansion | 007 Managing distributors | 008 Distributor development | 009 Distributor internal operations management | 010 Team management | 011 Efficient distribution techniques | 012 Sales manager's eighteen skills | 013 KA operation methods and strategies | 014 First lesson for new sales | 015 Internet, brands | [Long press QR code to follow] To join QQ/WeChat group, please click: Read original text