This paper is an integrated data analysis framework summarized by the author based on his years of experience in data analysis, and briefly introduces some points that data analysis can achieve. Enjoy it ~
Big data, a speculative concept, has now been replaced by artificial intelligence. Let's not discuss artificial intelligence. As far as big data is concerned, we are all emphasizing his technology, such as network hot words: hadoop+spark, data mining. When we use big data, we often use it to make up its influence. For example, accurate advertising, orderly social security management, and intelligent pharmaceutical industry.
Of course, these are all our imaginations and cannot be separated from the influence of data analysis. But have we stopped to think about how big data landed and how to analyze it? How to use data to let enterprises make decisions, such as accurate advertising?
Do we know what big data analysis is?
McKinsey defines big data as:
"A large-scale data collection that greatly exceeds the capabilities of traditional database software tools in acquisition, storage, management and analysis has four characteristics: huge data scale, rapid data flow, diverse data types and low value density."
Based on my understanding of the above definition, the big data analysis I summarized is to get through the obtained data, integrate it, find the law, and get the decision information immediately.
Data procurement
The data sources I summarized can be divided into three categories:
(1) Party data: user fact data.
For example, the wealth management products purchased by users in a financial institution, the time, which account, name, telephone number, etc. , or operational data, such as an Internet financial app, user operational behavior data.
(2) Second-party data: In fact, this part is called advertising data.
For example, the number of advertisements displayed, the number of clicks on active pages, the source of advertisements, etc. Some companies will use these data as third-party data, because some advertising monitoring companies will use these data to integrate with crowd data to build their own dmp. Companies like DMP generally claim to be third-party companies, third-party data.
(3) Tripartite data: industry data, also known as public data.
For example, association data, or Internet behavior data, such as the behavior data of users of an Internet company on this website, or the installation active list that we can collect after embedding sdk app, and the offline data that we can collect.
Get through: in fact, it is to integrate the first, second and third-party data by collecting key points. For example, we can integrate the data of one party and three parties through the mobile phone number, or use cookie or imei numbers to integrate the data of two parties and three parties. However, due to the current regulatory system's control over sensitive data of mobile phone numbers and the technical difficulty of cross-platform interoperability between Internet and mobile data, our actual matching rate is very low. For example, it is actually a good situation that the data matching between one party and three parties reaches 20%, except for the operator data.
Looking for patterns: the goal is to clean up data, from unstructured data to structured data, so as to make statistics, explore data, discover patterns and form the viewpoint of data analysis report. This article will elaborate in the third part.
Immediate decision-making: systematizing or productizing the opinions in the data analysis report. At present, most companies will still rely on manual decision-making
Why do you need big data analysis? Big data analysis seems to follow these steps, but from the data source of the first step, it actually reflects the characteristics of big data, that is, chaos. So how to find the rules from these data, and whether the content and objectives of the analysis correspond, seems to be the reason why we need big data analysis.
Nowadays, the analysis of big data usually uses data reports to reflect the operating conditions of enterprises. At the same time, for hot spot and crowd analysis, the statistics we see are all oriented by the viewpoints extracted from the data analysis report. Then the question is, how to use data analysis to guide data decision?
The report idea of data analysis (this paper cuts in from the perspective of mobile terminal)
Based on my understanding of data analysis, I divide data reports into three categories: market analysis, operation analysis and user behavior analysis.
market analysis
Generally speaking, market analysis is qualitative and quantitative. The congratulatory letter of the recent hit drama My First Half Life and Tang Jing's Career is to go to a consulting company. Generally, they will produce a market analysis report through interviews and questionnaires, telling customers their market share and consumer views.
Here, we take the market analysis of mobile Internet data as an example. Generally speaking, the data source is public data or third-party data. As we said, by embedding the sdk into the developer's application, we can collect the installation and use list. Then the more sdk developers use, the more data sources we can collect, so that we can form a ranking of installed apps and used apps. The coverage rate and activity rate mentioned here also mean this, such as the proportion of the installation and usage of this application in the whole financial category.
Then, the role of these market analysis is a summary of the company's marketing in general. For example, the kpi of a financial company is to get customers, and they have done a series of marketing. Next month, we can find out whether the installation of this application has increased compared with last month. How about the performance of our competing products? Did they also do a series of marketing activities, ranking up and down? We can all observe it through market analysis and competing products analysis, but this part of the view is market data, so we can only speculate whether the rise of competitors' rankings is related to these marketing activities through a large number of search activities of official website or Internet advertisements.
At the same time, according to the market trend chart, potential competitors can be found. For example, we can see that ICBC in the picture below is a potential competitor of all banking groups, because mobile phones belong to high coverage and high activity groups, that is, the active people who install xxapp are also the highest. Need to pay more attention to their marketing strategy.
Business analysis
Methodology proposed by mobile Internet: 33r. When I was consulting before, this methodology can also be applied to network analysis. To sum up, 33r is:
Perception → acquisition → activity → acquisition → income → dissemination → perception.
It should be noted here that operational analysis is only a company's baseline, and product managers, operators and marketers can make reasonable decisions according to their own company's data reference. At the same time, the operation data is only a reference or warning. If you want to be specific, you need to analyze specific details, such as whether the app is revised and how to change it. What channels of cooperation need to be increased?
(1) perception perception
According to the analysis of advertising data, the purpose is to judge the drainage of channel advertising pages to app or website, and at the same time, it can help advertisers to design monitoring tables and measure the advertising effect from a digital perspective.
However, advertising data is generally in the hands of advertising monitoring companies or public tools such as GA. We need to rely on advertising companies to design marketing links, such as activity pages, add monitoring codes, or add codes in media and app application stores to facilitate the monitoring of advertising effects. Usually, these data are difficult to load, and are usually provided by app stores or media. At the same time, the above data and monitoring company data are generally not provided to advertisers, but only statistical values.
Anyway, when we look at perceptual data, the purpose is to measure whether we spend a lot of money on marketing. The display volume and click volume of advertisements are the best measures to measure the performance of a company's advertising marketing department. Without advertising, there will be no customers, so the money spent is not worth it. How many customers can you bring before the next acquisition?
(2) Acquisition to win customers
Getting customers is the first step in advertising expansion. Users click on the advertisement, enter the app store or login page to download the app. After visiting the web page, the data after logging in to the app is data that cannot be provided by advertising companies or app stores, so there are actually two purposes to get customers.
Target 1: Measure whether the data provided in the first step is accurate, that is, whether it is channel cheating.
Goal 2: judge whether the channel is good or not.
Goal 3: Judge whether the marketing activity is effective.
For example, in the figure below, we found that the search traffic of 40% users increased by 6% compared with last month. Do we need to increase cooperation with sem? In the media recommendation channel, we measure the customer conversion rate through the channel, click-user activation, activate registration conversion, can we focus on increasing cooperation with an application store?
The figure below shows the application of Goal 3 to measure whether new users and active users are affected by campaign marketing, advertising and version changes within three months. For example, the version changed on July 28th, and a new user's weapon was added, so the product manager needs to analyze where the version changed, so that users can grow so fast. The marketing activities in August will wake up the sleeping users and reflect the performance of operators. Then, can we learn from the successful experience in August when promoting activities? This successful experience needs further thematic analysis.
(3) Active.
After acquiring customers, we hope to see the performance of our new and active users, so the third step is to be active, which is actually to provide data support for product managers to modify apps or pages.
Activity analysis can refer to the following three steps:
First, define the main page analysis from the number of page views and the number of independent visitors.
For example, the homepage of an app is pv, and the uv is the highest. We will focus on analyzing the homepage.
Second, according to the delineated page, make a click heat map, which is convenient for product managers to provide data support for subsequent page conversion. For example, we can delete the buttons with low clicks in the next version and reorder the buttons with high clicks.
Third, according to the delineated page, make a click heat map, which is convenient for product managers to provide data support for subsequent page transformation. For example, we can delete the buttons with low clicks in the next version and reorder the buttons with high clicks.
(4) Retention analysis &; Income and profit. refer to
In fact, these are not used much in enterprises, so here is a brief explanation.
① reservation
When users accumulate to a certain number, we want to see the stickiness of users, so we come to retain them. Retention is usually used to measure the effect of the activity and see if users will use our app after this activity, but because the financial app attributes will not be accessed every day like game applications, they will not be retained in practical applications. The following example is a demonstration and will not be repeated.
② Income
How much cash did these remaining customers contribute to the company? According to the different income steps, the general company will not put the cash flow data into the statistical platform, but we need to put the running amount data contributed by users for our use to facilitate crowd division. For example, the following is a brief analysis:
Reference communication:
Finally, we want to spread these customers; The core is word-of-mouth marketing, that is, users spontaneously forward links to other users to download apps or participate in activities, then the next link of communication will change marketing, but communication will be subject to many restrictions, such as word-of-mouth communication without reward mechanism, and the forwarding volume is almost zero. At the same time, communication is difficult to measure, especially on the basis of a large number of Internet users, which will cause resource code superposition and system burden, and general enterprises will not design such activities for marketers' reference.
User analysis
If the core of big data analysis is actually user analysis, as we said earlier, the steps of user analysis are as follows:
In other words, within the scope of data collection, get through data, get through customers and users, and make accurate marketing.
First, we can filter the list of conditions, and we can integrate the data by applying conditions, locations and label conditions. The purpose of integration is to describe customers and determine marketing strategies.
For example, we need to screen financial customers (applying conditional screening) and appear in five-star hotels (location conditions), especially mothers and babies (labels).
However, it should be noted that the more conditions, the clearer the outline of users and the fewer people.
Second, according to the screened population, do online/online statistics or multi-dimensional analysis of the model.
For example, according to the screened population, we found that there are more men than women, and Apple has the highest mobile phone attributes and often uses mobile phone tools. Then we can cooperate with this target group by adding mobile phone tools, or cooperate with Apple to get customers or promote activities.
Third, integrate the above data analysis to form a crowd portrait.
Concluding remarks
Based on my years of experience in data analysis, this paper summarizes an integrated data analysis framework, which is actually a brief introduction of several points that can be analyzed in data analysis. Of course, this requires a lot of data cleaning and understanding of the industry. This paper is only a summary from the perspective of data analysis, and the refinement of the content can actually be analyzed in detail, especially the chapter on user portraits.