A brand made a set of matrix numbers in Little Red Book, and their operation team had a full-time data analyst, who asked me for advice because he didn't know how to do the data analysis of Little Red Book. The role of data analysis is very important in product operation and marketing business, which is not only an important basis for leaders of various departments of the group to make decisions, but also an important way for marketing operation leaders to find business breakthroughs. Generally, a team with 10 or more people will be equipped with a data analyst to assist the business. A team with a scale of less than 10 can be held concurrently by operators who understand data analysis. Three realms of data analysis Three realms of data analysis: the first level, obtaining data, sorting out data and reporting data. The second layer, processing data, finding problems and asking questions. The third layer, through data analysis, finds insights and influences decision-making. The data analyst on the first floor goes to get the data first thing every morning. If the company has a database, but no visual retrieval tools, then write SQL statements to obtain data on the server. It may take several hours to get the data. If the company has a database and a visual retrieval tool, then let the database calculate the desired data according to the requirements, and then copy and paste it into the excel table. After getting the data that leaders need, they sort out the data and put it in a table or PPT for leaders' reference. Because there are new data every day, such data analysts have to repeat their work every day. Then when leaders meet the demand for upward returns, they will have more tasks to process data. The value of this kind of data analyst lies in obtaining data and sorting it out, thus saving the time of leaders and colleagues. But at the same time, this value can be easily replaced, resulting in overtime, hand speed, skilled shortcut skills and careful spelling. Crazy overtime, dare not complain. The data analyst on the second floor will process the data, so that leaders can easily see key data and find problems in numerous data, thus assisting leaders in making decisions. For example, leaders prefer to see the trend changes of data such as the sales volume of a product, while first-level data analysts will only give daily sales data tables. Data analysts at the second level can use tools such as PowerBI to generate data trend charts, so that leaders can clearly see how the recent data changes every day and whether they should make strategic adjustments. If you can do this, you can be regarded as an excellent assistant and a right-hand man in decision-making. The third layer is the third layer data analyst, who is good at finding problems in a pile of data, analyzing problems and designing strategies to solve problems. According to a legend, a supermarket found through data analysis that when beer and diapers were put together, their sales increased significantly, because fathers with babies would buy both at the same time. Although this is a legend, it means that some marketing insights can be found through data analysis, so as to put forward corresponding strategies and obtain obvious results. In the future, I will write an article about how I make decisions through data analysis, so as to achieve excellent results. Second, the process of data analysis The basic six steps of data analysis: 1, put forward the purpose of analysis This step is done by the boss in many companies, and the boss gives the task to the data analyst. For example, today, the boss said: Give me a table to see the daily statistics of sales in the last six months, the growth trend and the growth of sub-regions and sub-stores. For example, the day before yesterday, the boss said: Help me look at our products, which pairwise combinations are more appropriate. Last week, for example, the boss said that the recent sales figures were a bit sloppy. Help me find out why. But higher-ranking data analysts may find their own analytical purposes. For example, what strategies can be designed to increase the sales of a certain business? 2. Obtaining data Excel tables are suitable for processing data within 6,543.8+0,000 rows, and the data processing within 6,543.8+0,000 rows is also general. It can also handle thousands of online data, such as graphite documents. To process tens of thousands and hundreds of thousands of rows of data, we need to use some professional data processing tools, such as PowerBI. If you want to process millions to hundreds of millions of rows of data, you need to use some database tools, such as MySQL, and learn the basic database language. For us, it is enough to deal with the data related to the little red book, Excel table or graphite document online table. It may be faster to enter a small amount of data manually. Some data are a little large, or it is convenient to use a crawler, so use a crawler to collect data. For example, the results found by Baidu search are more convenient with reptiles. For example, for some web pages that don't need to enter a verification code to turn many pages, it is more convenient to use a crawler than to see the desired content without logging in an account. If you want to learn reptiles, you can learn related Python programming languages for this purpose. There are also some relatively simple crawler tools, such as octopus, such as webscraper. For the data related to the little red book, it is basically recommended to enter the data manually, and hundreds of thousands of data will be settled soon. 3. After processing the data, the data must be processed first. For example, is your data format correct? Some data formats are not convenient for you to calculate and sort later. Some data formats may be inconsistent, some data may be missing, and some data may be wrong. If these problems may affect the subsequent analysis results, they should be dealt with in advance. 4. Analyze the data. This is the most challenging step. What kind of analysis produces what kind of convincing conclusions. Before we come to a correct conclusion, we don't necessarily know what analytical methods to use, nor do we know what analytical ideas to use. In retrospect, you may think it was so simple, but in this process, it was so difficult. Just like we solve big math problems in middle school. 5. After the data visualization analysis is completed, it is often necessary to use visualization, not only to understand and see clearly, but also to make leaders and colleagues understand and see clearly. Make the analysis result more simple and intuitive. The simplest and most common data visualization methods are histogram, pie chart, line chart and scatter chart. 6. Only by drawing an executable conclusion in data analysis can we draw a conclusion, unify everyone's opinions, and then push everyone to take corresponding actions. If you don't approve a decision correctly and don't approve it, there may be conflicts or even naysayers in implementation. Therefore, data analysis is a way of persuasion, which can convince everyone through data, thus promoting business development. In the data analysis related to Xiaohongshu, the analysis tasks mainly focus on the following aspects: self-owned account analysis, benchmarking analysis and non-self-owned account analysis. Three own account analysis As a data analyst of a small red book team, the focus is naturally on analyzing your own account. You may be responsible for only one account, or you may be responsible for multiple accounts. The amount of data to be maintained is only 100-2000, so the fastest way to obtain data is to manually enter the data into an excel form. Because there are two main ways to view the data of the little red book account, one is to log in to the creation service platform on the computer side, and the other is to view it on the mobile phone side of the creation center. There are fewer data display dimensions on the computer side, but you can view the traffic trend chart of each note in the last 30 days. The data of Little Red Book is not arranged line by line, so it can't be copied and pasted into the table directly. The platform has an anti-crawling strategy, so it is not easy to organize it into a table with crawler software. The main data can only be seen on the mobile phone, which is more troublesome. So, honestly organize these data by hand. Collect the original data with tables. In the first table, enter the data of each note, including at least the following fields: account name, title, reading volume, like, collection, comment, sharing, fans, per capita viewing time, click rate, click rate evaluation, 5s playback completion rate (indicator of video content), playback completion rate evaluation (indicator of video content), content richness, content richness evaluation and home page recommendation. Age distribution (on-demand statistics 1-2 age group or all five age groups), urban distribution (before 1-3 city or before 10 city) and audience interest (before 1-3 interest or before 10 interest). In the second form, enter the basic data of the account, including at least the following fields: daily page views, page views in the last seven days, total viewing time in the last seven days, likes in the last seven days, favorites in the last seven days, comments in the last seven days, homepage visitors in the last seven days, notes shared in the last seven days, page views ranking percentile in the last seven days, interactive ranking percentile in the last seven days, recommended traffic ranking percentile in the last seven days, and. Pay attention to the proportion of page traffic in the last seven days, traffic from other sources in the last seven days, the number of new fans in the last seven days, the number of lost fans in the last seven days, the proportion of female fans, the proportion of age distribution, the proportion of urban distribution and the proportion of audience interest distribution. Calculate the reading amount of some indicators: click on the reading amount of notes through the search results, and the reading amount of note search = the reading amount of notes × the search ratio. Because you can't see the click rate of search results, you should look at your notes to search for reading. The higher the reading volume, the more advantages notes have in search results. If you want to layout the search results, you need to study these notes with high search reading and optimize the creative direction of the team. Recommended reading amount of notes: click on the recommended reading amount of notes through the home page, and the recommended reading amount of notes = recommended reading amount of notes × recommended exposure amount of home page: recommended exposure amount of notes on the home page, and recommended exposure amount of notes = reading amount of notes/click rate. A high recommendation value means that the system can recognize the comment more easily. Relevant factors related to the recommendation amount can be found, thus guiding the optimization direction for the team. Reading amount in the same city: the reading amount of users in the same city, reading amount in the same city = reading amount of notes × proportion in the same city. Some businesses have strong attributes in the same city, and the traffic outside the same city is of little significance. It depends on the reading volume in the same city. Interaction rate: interaction rate = (like+collection+comment)/reading volume. Interaction rate is generally regarded as an index to evaluate whether notes are worth recommending, and notes with high interaction rate are more likely to be recommended. Notes with high interaction rate can be used to summarize the experience of improving interaction rate, thus improving the performance of notes in the future. Powder rate: powder rate = powder number/reading volume. The higher the powder rate, the easier it is to attract readers' attention. The matters needing attention of high flour yield can be considered, and the French fries can also be considered to increase flour. How to get higher traffic through data analysis often leads to traffic anxiety, why the traffic has dropped significantly recently, and why the traffic has been unable to go up. At this time, data analysts need to tell everyone why this is the case and what changes should be made. See if the traffic distribution has changed and where the main traffic changes are concentrated. According to some indicators calculated before, generating a line chart can usually clearly see the problem, which is often caused by the obvious decline in the recent note recommendation traffic. Next, let's see what changes have taken place in data such as interest distribution, gender distribution, urban distribution and click-through rate. It may be that the target group corresponding to the notes has changed, or the notes themselves are not attractive enough, the click rate is not high, or the notes are not well written, resulting in low interaction rate. Once the cause is found, the solution can be found. Through data analysis, we can guide how to get more turnover. First of all, we should sort out the transaction path of this business. For example, the path of a medical and beauty institution is reading notes-homepage traffic-private messages-private domain to WeChat-transaction. Then it is necessary to monitor: 7 days of reading, 7 days of homepage visits, 7 days of private messages, 7 days of WeChat, and 7 days of transaction (amount), and then you can calculate the homepage visits, according to the private message rate of reading, according to the private message rate of homepage visits, private messages plus WeChat rate, and the transaction rate every 7 days. Then we can find the problem. For example, sometimes, although the reading volume has increased, the sales volume has decreased. It is found that the private message rate according to the reading volume has dropped significantly recently, and then it is found that the private message rate according to the homepage visit has dropped less obviously. Then the problem is that there is something wrong with the step of guiding the reading volume to the home page, and then the strategy of guiding the home page access needs to be worked out, and it takes iteration from copying the notes to guiding the comment area. IV Benchmarking analysis Benchmarking analysis mainly analyzes a batch of benchmarking accounts and a batch of benchmarking contents. We have a special article on benchmarking. Benchmark accounts are not only direct competing products, but also competitive accounts in some aspects, as well as accounts with high overlap in interest points, target groups and content styles. The main value of benchmarking analysis is to provide reference for content creation and operation action optimization. In the field of e-commerce, competitive data analysis is often done, but in the field of content, there are many accounts that usually grab your traffic, not just from your direct competitors, and lack special data. So we don't do data analysis of benchmark accounts every week, just pay attention to our own content. V. Analysis of Non-owned Accounts If you need to find bloggers for promotion, you need to evaluate the data of these bloggers. It is easy to make big mistakes simply by looking at the number of fans. If a good data analysis model can be established, it will be easier to screen bloggers and give appropriate pricing accordingly. The data with the strongest correlation with the advertising effect is the conversion rate, which needs to be tested many times and gradually accurate. The content published by the same blogger, even if the reading volume is the same, the conversion rate of different copywriters may be several times different. In addition to the conversion rate, the data with high correlation is the reading volume. The higher the reading volume of notes, we think it usually leads to higher sales. However, the conversion rate of different types of content is actually very different, several times or even ten times. Some notes have high click-through rate and low conversion rate, while others have low click-through rate and high conversion rate. There are two solutions. The first is to divide the notes into several types through the continuous accumulation of data analysis experience, and compare the notes of the same type, so that the conversion rate of the notes of the same type will at least not be too far apart. This solution requires continuous data analysis and research, which is very difficult. The second way is to accumulate the average measurement data through many cooperation cases, which can reduce the error to a certain extent and is simple to operate. Because bloggers don't necessarily want you to see the real reading volume, or it is not convenient to investigate the reading volume during the preliminary screening process. So the general popular strategy is statistical preference. However, the favorable rate of some types of notes can reach 10%-20%, and the favorable rate of some types of notes is even less than 0. 1%. And some bloggers' notes mainly come from a few loyal fans or praise each other. Notes with less than 50 likes are easy to be falsified by behaviors such as mutual praise and praise, so there is the possibility of data fraud. But this is not important in the preliminary screening. The way to evaluate explosions according to the standards of thousands of praises is also unreliable. Some notes like 1000 are only read about 1w, and some notes like 10w are only read about 100. Therefore, at the stage of screening accounts before establishing cooperation, at least the following data should be counted: nicknames, fans, total likes, top post titles, top post likes, the latest 10 articles or the average likes in the last two months, the lowest likes in the last two months, the likes of 30% graded works, the style of works and the content form of works. The lowest praise is used to estimate the traffic from fans. When it is hardly recommended by the system, the praise of the work is in a state of depression, and almost all the praise at this time comes from fans. 30% of the work points can be used to predict what kind of optimistic results you can get from your launch. The basic forecasting formula can be referenced as follows, and then adjusted and optimized according to the actual data. Expected reading = expected praise /3% expected sales = expected reading × expected conversion rate (1%) expected output = expected sales × selling price, so that we can preliminarily estimate the expected output brought by this blogger, and then decide how much advertising fees we can pay at most. Note that it is suggested to calculate the expected output repeatedly according to the actual experience, and the initial data we give are only used as reference in the case of lack of data. In the early stage, it is suggested to make a conservative estimate, that is, the expected output will be reduced by 5- 10 times.