Why do operators need to do data analysis?

Data analysis sounds like something that programmers do, but it is quietly approaching every operator.

Not to mention giants like Alibaba and Tencent, which have long used data analysis as an objective basis for decision-making. Even in start-up products, user, operation, sales and other data have been proven to have huge analytical value.

How to iterate product functions? How to optimize the conversion process? How to make precise delivery based on user portraits ? How to efficiently promote activation based on user behavior ? ...These things that seemed to be decided on the spur of the moment still have great shortcomings and limitations in the face of objective data.

In the past, the days passed slowly, but now new things are born, flourish and die every day. You may find that the experience that worked in the past is no longer as reliable in the rapidly iterating business environment, and its life cycle has become significantly shorter.

The era of data-driven and refined operations has arrived earlier than we expected.

For example, in terms of traffic operation, it is far from enough today to only focus on vanity indicators such as PV and UV. More sophisticated indicators such as CPC, DAU, average visit duration, visit depth, bounce rate, average traffic conversion, and large-scale data analysis based on these indicators are more analytically meaningful for judging user behavior.

For example, in content operation , personal experience and feelings may require long-term training, and how long this experience can be used is unknown, but the situation is becoming less and less optimistic. Based on user browsing preferences and usage habits, we can almost get very accurate and instructive conclusions.

For example, Taobao recommends products based on browsing history, NetEase Cloud Music recommends music based on similar users, and Toutiao recommends content based on user portraits. These are all examples of data analysis supporting content operations.

In fact, in slightly more formal product operations , the trend towards refined operations is becoming more and more obvious. The following are some of the most common processes in operation:

User Profile

User portrait seems simple, but in fact it is easy to guess based on your imagination, but it is difficult to define it accurately. So in the past, our definition of users was usually like this: "white-collar workers aged 25-30", "young professionals with reading habits", "programmers with 1-3 years of experience"... but this actually doesn't make much sense.

What products users are using, how they perceive the products, how often they use them, how their consumption levels are distributed, etc. However, in fact, more profound conclusions can be drawn by analyzing the user data of the products themselves.

Alternatively, we can also obtain external data for analysis through public data sets or crawlers, which is a very reliable industry research. Through data analysis methods, we can quickly understand new areas and new markets, so as to know where the opportunities are and effectively reduce the cost of trial and error.

Advertising & Channel Selection

You need to analyze the degree of compatibility between users of specific channels and target users . When faced with multiple similar channels, a rough estimate will often yield low returns. If you want to accurately position yourself through raw data such as user portraits and market research, and how to optimize through advertising data, these are areas that require precise calculation and analysis.

The first step in commercialization is to understand the complex transaction structure of the advertising market and the characteristics of one's own products and users, and to choose reasonable and efficient marketing plans and technical architectures.

Conversion analysis

There are many areas in the product that require conversion analysis: registration conversion, purchase conversion, activation conversion, etc. Generally, we use funnels to measure the user's conversion process. We can roughly draw some conclusions from the conversion funnel, such as at which stages users are hindered, whether the copy is not attractive enough, or the functional experience is too poor.

But if we do more in-depth data analysis, we can get more things, such as the differences in user conversion funnels from different channels, which can provide a reference for channel selection and advertising optimization. For example, which tags have better user conversion? Are there different barriers to churned users? What are they? How to prioritize various influencing factors during the conversion process?

More refined analysis and more hypothesis testing can help us come up with more detailed solutions.

Accurate recommendations

User behavior analysis has gradually become an integral part of various excellent products that cannot be ignored. The sudden rise of Toutiao and the good reputation of NetEase Cloud Music are all related to it.

We need to analyze user needs, such as the content that users are interested in, the proportion of content reading and dissemination, etc. How to label and divide users, and how to accurately recommend products and content based on users' historical habits are gradually becoming the key to promoting user activity and improving user stickiness.

Product iteration

Whether it is a product iteration plan or a strategy to promote activation, how to define the product iteration direction. Based on the analysis of user behavior data, heat map positioning of user browsing and clicking, and traffic monitoring of different pages and functions, new features to improve user experience are gradually and targetedly introduced .

Data is penetrating into every detail of our work. As the people who come into contact with users and products most frequently, operations is the first link to come into contact with data. At the same time, operations are usually the makers and implementers of strategies, but data is often controlled by the technology department or sits quietly without any effect.

For operations, there must be a lot of room for optimization between data generation and decision-making, and this room can largely come from data. Discovering knowledge from data and optimizing decisions will greatly improve operational efficiency.

On the other hand, with the support of data, you can better convince your boss, even if you have a fight with technology or products, it is not at all in vain. But the prerequisite is that you can make good use of the data, analyze the conclusions to support decision-making, and describe them in a visual way that ordinary people can understand.

Nowadays, any product of a certain scale has a very large amount of data and many fields. You may be confused about where to start.

But it would be much better if you have some experience. For example, if you want to study the physical factors that affect a runner's speed, we might study the athlete's height, leg length, weight, or even heart rate, blood pressure, and arm length, but we are unlikely to study the length of the athlete's armpit hair. This is based on the knowledge we already have. For example, if you want to analyze several indicators that affect product quality and the priority of factors that affect conversion, then if you conduct preliminary analysis, you can draw some preliminary conclusions.

So when you analyze more problems, you will have some sensitivity to data and develop the habit of analyzing and speaking with data. At this point, you can even make preliminary judgments and predictions based on some data and your own experience (of course, this cannot replace the accurate predictions of a complete sample). At this point, you basically have data thinking.

I believe you have heard the term data-driven a lot. The low signal-to-noise ratio and waste of resources caused by rough operation strategies can be effectively alleviated through refined operations. When indicators such as UV and PV are no longer able to make accurate decisions, more sophisticated data analysis will be the driving force for future operations.

More importantly, the most basic, repetitive, and low-value operational work will gradually be replaced by automated operations. For example, the recommendation system based on data analysis has effectively replaced the screening and push of some content. One of the core competitive advantages of future operators will be to achieve efficient and automated operations through data analysis.

So can data analysis skills be acquired quickly? Of course you can.

From an operational perspective, the general data analysis process is: problem definition, data acquisition, data cleaning, data modeling and analysis, data visualization and conclusion.

● For operators, problem definition is an internalized skill. Because of their familiarity with the business and their understanding of products and users, they should be considered to be at the top of the Internet circle. Even now, you have indicators that you want to improve and processes that you want to optimize, and these are all good sources of questions.

● In the data acquisition part, the operator’s data generally comes from the database of the enterprise’s products. At this time, you need to understand SQL operations, or at least be able to extract data from the database proficiently. In addition, mastering crawlers and being able to obtain industry data from external websites will open up ideas for global analysis.

● Even the company’s own data is mostly incomplete, inconsistent and dirty data, which cannot be directly analyzed or the analysis results are unsatisfactory. For example, duplicate data, accurate data, invalid data, etc., only by properly processing these data that affect the analysis can more accurate analysis results be obtained.

● Use basic data analysis methods or data mining algorithms to obtain the conclusions you want, such as the regression models (linear regression, logistic regression) commonly used in analysis. The most direct results of the analysis are the description and display of statistics, while others require in-depth exploration of internal relationships in order to make accurate predictions about future situations.

In general, there are three parts of skills that need to be mastered:

● SQL (database). Enterprise data is generally stored in a database, so how do you retrieve data from the database? How to establish the relationship between two or three tables? How to get the specific data you want? These data selection problems are your first consideration, and these problems are all solved through SQL, so SQL is the most basic skill for data analysis.

● Basics of statistics. The premise of data analysis is to have awareness of data. How to collect data? What is the overall distribution of the data? If there is a time dimension, how does it change over time? What guiding significance do the median, mode, and significance of data have? How to perform preliminary analysis using hypothesis testing?

● Python basics. This is a must and also a plus. Compared with tools , language is more flexible and practical, which can make it easier to realize your ideas. Python has many open source libraries (such as numpy, pandas, scikit-learn, seaborn), which makes scientific computing and data visualization a piece of cake.

The fastest learning path is based on the problem-solving process, so that you know what each part of the knowledge is used for and where it can be used, and each part can solve some practical analysis problems.