How to design daily reports and how to play the large-scale daily reports of top players

“Today, we return to the original plan - how to design the daily report of the Internet finance project, and by the way, how the large-scale daily report of the top players is played. This article has a total of 2,100 words and is expected to take 8 minutes to read.”

Data is the foundation of operations. The closer to the core of the business, the more important data capabilities are. The operation of Internet finance is no exception. The daily use of data in operations mainly includes three aspects: data monitoring, data analysis, and data conclusion output strategy. Making a good daily data report is the most basic requirement for data monitoring work.

To do a good job of data monitoring, we need to know the purpose of data monitoring, and then we can prescribe the right medicine. The purpose of data monitoring is simple, which is to quickly understand the current data situation and to find superficial problems based on the data. Among them, quickly understanding the data is the main purpose, and finding superficial problems is the secondary purpose. The design of the daily report should aim to meet these two needs.

01—Quickly understand the current status of data

So, how can we design daily reports to help us quickly understand the data? Disassembly is the best way. Disassemble the data into result data and process data, core data and segmented data, attribute data, behavior data, and business data, and display them separately, so that the data can be clearly understood at a glance.

Taking the new user part of the loan assistance business as an example, the core goal of the new user loan assistance is to increase the total amount of loans. Then the result data is UV, loan amount, number of borrowers, and average number of loan items, and the process data is total conversion. Then these five indicators are the most important core indicators.

If we further break it down to the first-level business node, new customers have to go through five steps: active, application, credit approval, loan application, and loan approval. Then the result data should be added with the scale indicators of these five steps, namely, UV, new customer application, credit approval, average number of credit approvals, total credit amount, loan application, loan approval, average number of loan applications, and total loan amount.

The process data is to break down the total conversion into the conversion rates of the five steps, including new account registration rate, credit application rate, credit completion rate, credit approval rate, loan application rate, loan completion rate, loan approval rate, and total conversion rate. With these five core indicators and 12 first-level node indicators, we can quickly understand the overall situation of the business from the business result data and process data, from the core goals and the first-level structure.

Next, let's break down the segmented data. Segmented data can first be divided into behavioral data and business data. The behavior of users on the active, registered, credit, and borrowing pages is behavioral data; the data generated in the credit and borrowing systems is business data; and the user's channel source, credit funding party, etc. are attribute data.

According to this cross-split of behavioral data, active UV can be broken down into data of different drainage entrances. Registration can be further split into step data according to registration page visits, clicks to register, registration success, registration application rate, registration success rate, and total registration conversion rate. Login can split data according to the active sections of logged-in users who have not submitted applications. Submissions can be split according to users' browsing on the submission page, the number of users entering the next step, the application rate of each step, the approval rate, and the completion rate of the overall credit process. Borrowing is similar to submissions, and can be split according to the pages of different investors and borrowers.

Business data can be broken down by the number of people entering the business system, pass rate, average number of cases, total amount, and investor.

This table is just an example and is not completely consistent with the content of the article. It needs to be adjusted according to actual business.

02—Quickly discover data problems

The shortcut to quickly understand data is to decompose the data, and the way to quickly discover problems is to compare them.

There are two directions for comparison. One is vertical comparison, that is, comparison in time dimension, such as day-on-day comparison, week-on-week comparison, week-on-week comparison, month-on-month comparison, year-on-year comparison. The other is horizontal comparison, that is, comparison in space dimension, such as comparison within the industry, comparison of similar business lines, comparison between new and old customers, comparison between different channels, and comparison between different investors.

How to add comparison items in the daily report? Because it is daily data, the key data must have daily month-on-month data, 7-day data fluctuations, and month-on-month data. Among them, daily month-on-month and month-on-month data only need one column each, so the key data and detailed data columns are indispensable. The 7-day fluctuation data has a high update cost, so only the key data can be placed.

The right side of the daily month-on-month, month-on-month, and fluctuation trend is the original data for easy comparison

In addition, when we set lending targets on a quarterly, monthly, or weekly basis, in order to continuously adjust strategies based on target completion, we can add a current indicator completion progress item to the key data in order to control the overall indicator completion status.

In this way, a daily report for the Internet finance industry has been designed. Different businesses can be adjusted according to this logic and business nodes. Next, we will talk about how to obtain and summarize these data.

03—Data acquisition and integration

In terms of data acquisition, data is mainly divided according to different data collection methods. For example, behavioral data is generally collected by embedding points, business data is generally recorded in the business system base table, and attribute data is collected by adding fields in the embedding points or system base tables. Except for channel data, a separate collection method is rarely required. There are two different situations for aggregating data from different places. One is that the behavioral and business data are separated, and the corresponding data needs to be consumed directly in the corresponding system. For example, many behavioral data collection tools have accompanying analysis tools. For example, Sensors, the best in the industry, has a set of Sensors analysis tools that can be used directly for corresponding behavioral analysis. For business base tables, data should be obtained on the corresponding data acquisition platform, such as hive, spark, etc., and sql. In another case, behavioral data and business data are connected. Some are directly connected through data acquisition tools, and behavioral data, business data, and attribute data are all collected in one system for consumption; some are imported into their own data warehouses, and user association is performed by themselves, and the front-end and back-end data are connected. The fragmented data needs to be obtained from two platforms separately and then aggregated on another platform, such as using a python script to capture behavioral data, querying the bottom table data through SQL, and finally manually aggregating it into an Excel table. The connected data can be exported to a third party such as Excel consumption, or made into a report and automatically updated. In comparison, the connected data is naturally more efficient to use.

04—Does the Big Model Daily have a future?

After talking about the design of daily reports, data acquisition and aggregation, let's talk about something more advanced - large model automated analysis. Some leading companies in the industry have developed a set of AI tools for data monitoring and analysis. Based on the designed reports and configured indicator relationships, it can evaluate the changes in each part of the data in real time from both the quantity and proportion aspects, and find the most influential change points.

The advantage of this system is that it can analyze data from all dimensions in the database, automatically produce and send data briefings, and send real-time alerts based on data fluctuations. Its disadvantage is that it cannot fully understand adjustments in business channels, strategies, activities, products, etc., and does not have the experience of senior operators. It can only see the change points of the data, but cannot find the business reasons for the data changes and produce strategies. At present, it is only a briefing generation tool, but with the rapid development of artificial intelligence, it is not impossible for it to replace operators in monitoring, analysis and decision-making in the future.

The above is a summary of the operation methods and evolution stages of the top players in the credit industry. Next week, we will continue to update the basic practice chapters. There will be basic practice chapters such as how to establish a business growth model, how to design a business estimation model and a business strategy library, and how to design a strategy execution table. There will also be method application chapters such as how to apply the growth model to decompose indicators to quickly find growth points, how to apply the UJM model to land new customers, and how to apply the life cycle model to manage old users. There will also be scenario practice chapters such as how to conduct AB experiments, how to attribute traffic sources, how to design regular activities, and how to design intelligent decision-making models.