Data Operations: How to build an efficient data analysis system?

With the advent of the data-driven and refined operations era, how to iterate product functions? How to optimize product conversion process? How to make precise delivery based on user portraits?

In the data-driven era, the era of making decisions based on feelings and experience is over. As an operations manager, you need to master certain data analysis capabilities to find, analyze and solve problems from the data.

So, what is data analysis? How does operations use data analysis to guide product optimization? What are the methods of data analysis? Next, I will talk to you about some things about data analysis.

1. Why do we need to do data analysis?

With the advent of the data-driven and refined operations era, how to iterate product functions? How to optimize product conversion process? How to make precise delivery based on user portraits?

…

Faced with this series of problems, you will find that the methods that worked in the past are no longer so reliable, and analysis based on objective data can more accurately assist operations in making decisions.

For example: traffic operation, only focusing on vanity indicators such as PV and UV, seems far from enough now. 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.

2. What is data analysis?

1. concept

Data analysis, as the name suggests, is data + analysis, which means that data must come first and analysis must come second. Using appropriate statistical methods to analyze a large amount of collected first-hand and second-hand data in order to maximize the value of the data is a process of detailed study and summary of the data in order to extract useful information and form conclusions.

1. Purpose

We use data analysis to always solve certain business problems and drive business growth. Depending on the type of problem we want to solve, we can divide the purpose of data analysis into three categories: current situation analysis, cause analysis, and future prediction.

(1) Current situation analysis

The meaning of current situation analysis can be roughly viewed from two points of view: what has happened and what is happening now. Through analysis, we can tell you the overall operation status of the enterprise, let you understand the development and changes of various business operations of the enterprise, and have a deeper understanding of the operating status of the enterprise.

Current situation analysis is generally completed through daily reports, such as daily, weekly, monthly reports, etc.

(2) Cause analysis

After the first phase of current situation analysis, we can understand that there are certain hidden dangers in the enterprise, so we should analyze these hidden dangers. For example: the registration conversion rate of a certain product must be stable at 15%, and one day it suddenly drops to below 5%. At this time, it is necessary to analyze the data of that day, find out the reasons for the drop in registration conversion rate, and provide solutions. This is the cause analysis.

Cause analysis is generally completed through special analysis, and cause analysis is selected for a certain current situation based on actual operating conditions.

(3) Predicting the future

After analyzing the current situation and the causes, the next step is to predict the future. Operators use the data they have and data analysis methods to predict future development trends.

For example: the seven-day repurchase rate of a certain e-commerce company is 30% on average. Now there are 1,000 first-time purchasing users. The behavior of these users is monitored to see whether their repurchase rate reaches or exceeds 30% in seven days. The growth rate of repurchase is judged based on the data results. This is data analysis and prediction of future applications.

Predictive analysis is generally completed through thematic analysis, which is carried out when formulating quarterly, annual, etc. plans. It is not conducted as frequently as current situation analysis and cause analysis.

3. How to do data analysis

Many people who are new to data analysis don’t know how to do data analysis. Either you try to do everything at once, or you don't know where to start. These are all manifestations of a lack of analytical ideas. Today, I will give you an inventory of the workflow of data analysis and the commonly used data analysis methodologies and methods.

1. Data analysis process

Data analysis mainly includes six relatively independent yet interrelated stages, namely: clarifying the purpose and ideas of analysis - data collection - data processing - data analysis - data presentation - report writing.

The picture comes from the Internet

(1) Clarify the purpose and approach of the analysis

There must be a clear purpose before doing anything, and the same is true for data analysis. Before conducting data analysis, you must first clarify the purpose of data analysis, know why you want to do data analysis, and what results you want to achieve. For example: the purchase conversion rate of the original product landing page is relatively low, and a new landing page is needed to increase the purchase conversion rate after traffic enters.

(2) Data collection

Data collection is a process of purposefully collecting and integrating relevant data according to determined data analysis and framework content. It is the basis of data analysis.

One way to collect data is to add "point-of-view" codes to the code of your own products, and another way is to use third-party data statistics tools (such as Baidu Statistics). They can all monitor a series of user behaviors in the product and save the data for subsequent analysis.

(3) Data processing

Data processing refers to the processing and organization of collected data in order to carry out data analysis. It is an essential stage before data analysis. This process takes up the largest amount of time in the entire data analysis process, and also depends to a certain extent on the construction of the data warehouse and the assurance of data quality.

The main tasks of data processing include data cleaning, data conversion, data extraction, data merging, data calculation and other processing methods, which are used to process various raw data into the styles required for data analysis.

(4) Data analysis

Data analysis refers to the process of using appropriate analytical methods and tools to analyze processed data, extract valuable information, and form effective conclusions.

At this stage, being able to control data and conduct data analysis requires the use of tools and methods. We can complete general data analysis through Excel, while advanced data analysis requires the use of professional analysis software, such as Power-BI, SPSS, R and other data analysis tools.

(5) Data presentation

Generally speaking, the results of data analysis are presented in the form of graphs and tables. As the saying goes: words are not as good as tables, and tables are not as good as graphs. With the help of data presentation methods, the information, opinions and suggestions you want to present can be expressed more effectively and intuitively.

Commonly used data charts include pie charts, column charts, bar charts, line charts, scatter plots, radar charts, etc. Of course, these charts can be further organized and processed to become the graphics we need, such as: pyramid charts, matrix charts, funnel charts, Pareto charts, etc.

(6) Report writing

The final stage is to write a data analysis report, which is a summary and presentation of the entire data analysis process. Through the report, the causes, processes, results and suggestions of data analysis are fully presented for reference by decision makers. A good data analysis report needs to meet the following three requirements: a good analysis framework, clear conclusions, and feasible suggestions or solutions.

1. Data Analysis Methodology

There are many methodologies for data analysis, which will not be listed in this article. The editor introduces the more common theories to everyone so that you can use them as guidance when establishing a data analysis framework in the future.

(1) PEST analysis method

The PEST analysis method analyzes the internal and external environment from four aspects : politics, economy, society, and technology , and is suitable for the analysis of the macro environment. The PEST analysis method can better grasp the current situation and changing trends of the macro environment from all aspects, which is conducive to enterprises to take advantage of opportunities for survival and development and to detect and avoid possible threats from the environment as early as possible.

The picture comes from the Internet

The PEST analysis method includes four factors: politics, economy, environment and society, which are also called "PEST hazards". PEST requires senior management to have relevant capabilities and qualities. PEST is a basic tool for enterprise and environmental analysis. When combined with factors of the external overall environment, it can summarize the opportunities and threats in the SWOT analysis.

(2) SWOT analysis

SWOT analysis (also known as TOWS analysis, Dows matrix) is a situation analysis method. S (strengths) is advantage, W (weaknesses) is disadvantage, O (opportunities) is opportunity, T (threats) is threat or risk.

The picture comes from the Internet

SWOT analysis is a scientific analysis method used to determine a company's own competitive advantages, disadvantages, opportunities and threats, thereby organically combining the company's strategy with its internal resources and external environment.

Using this method, we can conduct a comprehensive, systematic and accurate study of the situation in which the research object is located. By analyzing the various major internal strengths, weaknesses, external opportunities and threats closely related to the research object, a conclusion can be drawn, which usually has a certain degree of decision-making. Based on the conclusions, corresponding development strategies, plans and countermeasures can be formulated.

(3) 5W2H analysis method

As shown in the figure below, the 5W2H analysis method analyzes problems from seven common dimensions: Why, What, Who, When, Where, How, and How much .

The picture comes from the Internet

This analysis method, also known as the Seven-Dimensional Analysis Method, is a very simple, convenient and practical tool. It is widely used in corporate marketing and management activities. It is very helpful for decision-making and executive measures, and also helps to make up for omissions in consideration of issues.

To put it simply, the 5W2H method is a way to discover and solve problems.

(4) 4P Marketing Theory

The 4P marketing theory originated in the United States in the 1960s, namely product, price, place, and promotion. In the marketing field, this market-oriented marketing mix theory is the most commonly used by enterprises.

The picture comes from the Internet

It can be said that all marketing actions of an enterprise are carried out around the 4P theory, namely: product, price, channel, and promotion. By combining and coordinating the four, the company's market share can be increased and the ultimate goal of making a profit can be achieved.

For the mobile phone industry, the 4P theory should not be unfamiliar. Take OPPO as an example. Every aspect of its products, prices, channels and promotions is worth learning.

Product: For consumers, a good product is one that solves their pain points. OPPO's product strategy is to continuously meet consumers' higher demands and address their pain points. The slogans "Charge for five minutes, talk for two hours" and "This moment, clearer" reflect this point well. Price: OPPO's overall pricing strategy is to unify and strictly control prices nationwide. This strategy will not result in different prices in different channels. To some extent, it also limits online channels. If the prices are the same online and offline, consumers are more willing to go to physical stores to experience them before purchasing. Of course, this method is beneficial for the company to manage prices. On the other hand, it also makes consumers feel at ease. Although there is no sense of discount, they are not at a disadvantage either. Instead, they will have more trust in the brand. Channel (place): OPPO's channel tends to be flat, "OPPO-provincial agent-agent-user", in which OPPO cooperates with channel partners in a bundled manner. Some partners hold shares in the company, which will make channel partners more attentive and hardworking in sales. It also builds a high degree of trust with channel partners and enables them to survive steadily during fluctuations. Promotion: OPPO’s marketing promotion strategy is: vigorous publicity and large-scale appearances, so that consumers do not have to search for information effortlessly, but can accept it easily, and this acceptance is subjective and willing to accept. A typical example is to invite a large number of popular idols to endorse the brand, such as Yang Mi, Li Yifeng, TFboys, Yang Yang, Di Ali Gerba, etc.; sponsor many popular variety shows, such as "Running Man" and "Go Fighting"; and widely place advertisements in airports, subways, and high-speed rail stations with large passenger flow in various places. This direct and sharp approach allows consumers to quickly receive the message that the brand wants to convey.

(5) AARRR model

The AARRR model is a data model that all operations personnel must understand. The data analysis framework in the famous "Growth Hacker" is also based on this model.

AARRR starts from the entire user life cycle, including acquisition, activation, retention, revenue and referral.

The picture comes from the Internet

Each link corresponds to the five important processes in the life cycle, namely, from acquiring users, to increasing activity, improving retention rate, and obtaining revenue, until finally forming viral spread.

1. Common data analysis methods

The above introduces 5 classic analysis methodologies, which can help us build a clear data analysis framework. So what should we do about specific business scenario problems?

According to the actual needs of operational work, the following editor introduces several commonly used methods in data analysis, hoping to be helpful to everyone in the actual application of data analysis.

(1) Trend analysis

Trend analysis is the simplest, most basic and most common method of data monitoring and data analysis. Suitable for long-term tracking of core product indicators, such as click-through rate, GMV, number of active users, etc.

Generally, a data trend chart is created. Through intuitive numbers or trend charts, you can quickly understand the market, user or product characteristics, etc.; you can also divide indicators according to different dimensions and locate optimization points, which helps to make decisions accurately and in real time.

Taking e-commerce websites as an example, if we take traffic as the first key indicator. We aggregate website visitor volume (UV) and page views (PV) and other indicators into a unified data dashboard and update it in real time. With such a data dashboard, the core figures and trends are clear at a glance.

(2) Multidimensional decomposition

When a single number or trend is too macro, we need to break down the data through different dimensions to gain more detailed data insights. The dimensions here include but are not limited to browsers, access sources, operating systems, advertising content, etc. When selecting dimensions, you need to carefully consider their impact on the analysis results.

For example, when monitoring the website's bounce rate is 0.47, the average visit depth is 4.39, and the average visit time is 0.55 minutes. Then you can break down these indicators into multiple dimensions, such as region, access source, device, browser, etc. After the breakdown, you will find many ideas.

(3) User segmentation

Targeting users with certain behaviors or background information through specific optimization and analysis is what we often call user segmentation.

For example: when considering the registration conversion rate, it is necessary to distinguish whether the user login platform is PC, tablet or mobile phone, as well as the user groups in Beijing, Shanghai, Guangzhou, Shenzhen and other places. This allows for targeted optimization of channel strategy and operational strategy.

(4) Funnel analysis

Funnel analysis is one of our most common data analysis methods. It is widely used in daily data operations and data analysis work such as traffic monitoring of website user behavior analysis and APP user behavior analysis, product target conversion, etc. For example, using a funnel chart to analyze the conversion rate of certain key paths in a website can not only show the final conversion rate from users entering the website to making a purchase, but also show the conversion rate of each node in the entire key path.

There are two key points to note when doing funnel analysis:

Not only should we look at the overall conversion rate, but also pay attention to the conversion rate of each step in the conversion process; funnel analysis also needs to be broken down into multiple dimensions. After the breakdown, you may find that the conversion rates in different dimensions also vary greatly.

(5) Retention analysis

In an era when the demographic dividend is gradually fading, the cost of retaining old users is much lower than acquiring new users, so retention in analysis is one of the most important indicators.

Retention analysis is an analytical model used to analyze user engagement/activity. It examines how many users who perform initial behavior will perform subsequent behavior. This is an important method for measuring the value of a product to users. Every product and every service should focus on user retention and ensure that every customer is realised.

Common indicators for measuring retention include: next-day retention rate, 7-day retention rate, 30-day retention rate, etc.

（6）A/B Testing

One of the main ideas of growth hacking is not to make something big and comprehensive, but to constantly make small and precise things that can be quickly verified. Quick verification, how to verify? The main method is AB testing.

A/B testing is to achieve a goal by adopting two sets of plans. Through experiments, the data effects of the two sets of plans are observed to judge the pros and cons of the two sets of plans.

For example, Google will develop a variety of different plans for displaying search results (including text titles, font size, color, etc.) to continuously optimize the click-through rate of ads in search results.

The picture comes from the Internet

One thing to note when conducting A/B testing is that it is best to have A/A testing or similar preparation before conducting A/B testing.

What is A/A testing?

A/A testing is to evaluate whether the two experimental groups are at the same level, so that A/B testing is meaningful.

4. Common Fallacies in Data Analysis

During the data analysis process, even very experienced data analysts must be wary of data fallacies. Understanding these types of errors can avoid disasters during analysis.

1. Data Bias

It is absolutely essential that one should remain neutral and not fall in love with your assumptions.

— David Douglass, American physicist

Being influenced by personal biases and motivations when analyzing data, i.e. selecting only data that supports your claims while discarding those that do not. "Data bias" will destroy the objectivity of the data.

The way to avoid this fallacy is to collect as much relevant data as possible and ask for other people's opinions when analyzing data.

1. Sampling bias

Drawing conclusions from data that are not representative. For example, a news and information app that is hardly used by people in the Internet circle, why does this app still have such a large number of views?

So when analyzing data, an important step is to ask yourself what data you are missing. Sometimes it may be impossible to grasp the overall picture of the data because they only reflect a part of it.

1. False Causation

When analyzing data, it is easy to judge that two events that occur at the same time (correlated) are causally related.

The way to avoid this fallacy is to collect more data and look at possible third causes, sometimes finding that their correlations may be related to a third independent factor rather than to each other.

1. Simpson's paradox

When adding together the data of two groups with large differences, the one that has the advantage in the group comparison will be the losing side in the overall evaluation.

To avoid the misunderstanding brought to us by "Simpson's Paradox", we need to consider the weights of individual groups and use a certain coefficient to eliminate the impact caused by differences in the base number of grouped data.

5. Final Thoughts

Knowledge gained from books is always shallow. The above content only provides a basic framework and ideas. If you want to truly master the skill of data analysis, you need to apply it to actual work and practice makes perfect.

Source: Activity Box Operation Society (huodongheziyys)