Data analysis is ubiquitous in operational work. Whether it is activity review, special reports, project optimization, or job interviews, data analysis has a place.Regarding data analysis, I found that many operations have some confusions:
Don’t know where to get the data;
I don’t know what tools to use;
Unclear methodology and framework for analysis;
Most data analysis is just a formality;
…
In fact, data analysis is not as difficult as you think! I have come into contact with many data practitioners and summarized this article, hoping that it will be helpful to operations students who are interested in learning data analysis.1. Concept: Data and Data AnalysisIn fact, we have always been in contact with data and data analysis, but it is difficult to give a clear definition of the two. I once conducted a survey and asked some operations colleagues which of the following five options fall within the scope of the concept of "data".Most people know to select "4. Report", but it is difficult for anyone to think that all 5 options above are correct. In fact, this reflects a very common phenomenon: many people have a preconceived notion that data are various tables and numbers, such as Excel reports and various databases. In fact, this is a wrong or biased understanding, which will make our understanding of data very narrow.1. What is data?Data is a symbolic record that describes things and is the raw material that constitutes information or knowledge. This philosophical definition greatly enriches the scope of data and meets the current needs of the development of "big data". Think about it, isn’t the “image recognition” and “audio recognition” that many search engines are doing now part of data analysis?As an operations practitioner in an Internet company, the data we come into contact with may not be that complex, but there are also many categories.From the source of the data, it can be divided into external data and internal data of the enterprise. External data mainly includes social demographic, macroeconomic, news and public opinion, and market research data; internal data includes user behavior data, server log data, CRM and transaction data. The ways of acquiring different data, the analysis methods and the analysis purposes are all different. Different industries and enterprises also have their own preferences in actual analysis.So what is the difference between the common “information” and “data”?Data is the carrier and expression form of information; information is the connotation of data, and information is loaded on data. Taking books and knowledge as examples, books belong to the category of data concepts, and knowledge belongs to the category of information concepts; books are a carrier and form of expression of knowledge, and knowledge is the connotation and sublimation of books.2. What is data analysis?Data analysis refers to extracting useful information from data and guiding practice.There are two points to note here: first, what we need to extract is useful information, not just self-satisfaction; second, this information needs to be used to guide practice, not just for formalities.2. Ideas: Methodology and methodsWhen many newcomers get started with data analysis, they either get confused or have no idea where to start. These are all manifestations of a lack of analytical thinking, which requires guidance from both macro-methodology and micro-methodology.So what is the difference between methodology and method?The methodology is an analytical framework proposed from a macro perspective, from the perspective of management and business, which guides the direction of our subsequent specific analysis. Method is a micro concept, which refers to the method we use in the specific analysis process.1. MethodologyThere are many methodologies for data analysis. Here I will introduce some common frameworks.
PEST analysis method: Analyzes internal and external environments from four aspects: politics, economy, society, and technology. It is suitable for macro analysis.
SWOT analysis method: Analyze the internal and external environment from four aspects: Strength, Weakness, Opportunity, and Threat. It is suitable for macro analysis.
5W2H analysis method: analyze problems from 7 common dimensions: Why, When, Where, What, Who, How, and How much.
4P Theory: A classic marketing theory that holds that product, price, channel and promotion are important factors affecting the market.
A AR RR: Pirate's Law of Growth Hacker, an important framework for lean entrepreneurship , which aims to achieve growth from five aspects: acquisition, activation, retention , revenue and referral.
There are many methodologies for data analysis, which cannot be listed here one by one; there is no best methodology, only the most suitable one. Below I will introduce the AARRR methodology in detail. This methodology is very suitable for issues such as lean operations and business growth.For Internet products, users have obvious life cycle characteristics. I will use an O2O industry APP as an example to illustrate.First, acquire new users through various online and offline channels and have them download and install the APP. After installing the APP, users are activated through operational means; for example, first order free, vouchers, red envelopes, etc. Through a series of operations, some users are retained and generate revenue for the company. During this process, if users think the product is good, they may recommend it to people around them; or encourage sharing to their friends circle, etc. through incentives such as red envelopes.It should be noted that these five steps are not necessarily in the order above; operations can be flexibly applied according to business needs.The five links of AARRR can all be measured and analyzed through data indicators to achieve the goal of lean operations; the improvement of each link can effectively grow the business. Our following analysis is also based on this methodology.2. MethodsAccording to the actual needs of operational work, and based on Chen Ming's article "How to Become an Excellent Data Analyst" from GrowingIO, I have compiled 7 analysis methods. With the help of common website/APP data analysis products, we can complete these 7 types of analysis very quickly.1. Trend analysisTrend analysis is the simplest, most basic and most common method of data monitoring and data analysis. Usually we create a line graph or bar graph of data indicators in data analysis products, and then continue to observe, focusing on outliers.In this process, we must select the first key indicator (OMTM, One Metric That Metter) and not be confused by vanity metrics.Taking social apps as an example, if we use download volume as the first key indicator, we may go astray; because users downloading the app does not mean that they use your product. In this case, it is recommended to use DAU (Daily Active Users) as the first key indicator, and only users who initiate and perform a certain operation can be counted; such indicators are of practical significance, and operations personnel should focus on such indicators.2. Multidimensional decompositionMulti-dimensional decomposition means splitting indicators from multiple dimensions based on business needs; the dimensions here include but are not limited to browsers , access sources, operating systems, advertising content, etc.Why is multi-dimensional disassembly necessary? Sometimes you can't see any problems with a very general or final indicator, but after breaking it down, many detailed problems will emerge.For example, the bounce rate of a certain website is 0.47, the average visit depth is 4.39, and the average visit time is 0.55 minutes. If you want to increase user engagement, obviously such data will make you feel at a loss as to where to start; but once you break down these indicators, you will find many ideas.Shown below is the user engagement metric data for a product on different operating systems.If you look closely, you’ll find that user engagement on mobile platforms (Android, Windows Phone, IOS) is extremely poor, as evidenced by extremely high bounce rates, low visit depth, and low average visit duration. In this case, you will find the problem: is it because our product is not optimized on the mobile terminal, resulting in poor user experience? In this era of mobile Internet , this is a very important issue.3. User SegmentationThere are two main ways to segment users: dimension and behavior combination. The first type is grouping based on user dimensions. For example, based on the regional dimension, there are users from Beijing, Shanghai, Guangzhou, Hangzhou and other places; based on the user login platform, there are PC, tablet and mobile phone users. The second method is to group users based on their behavior combinations, such as the difference between users who sign in to the community three times a week and users who sign in less than three times a week. I will introduce this in detail in the retention analysis later.4. User scrutinyAs mentioned earlier, user behavior data is also a type of data. Observing the user's behavior path within your product is a very intuitive analysis method. Based on user grouping, generally 3-5 users are selected for detailed investigation, which can cover most of the behavioral patterns of the grouped users.Let’s take the registration process of a product as an example:The user went through the following operation process: [Visit the official website] - [Click to register] - [Enter number] - [Get verification code]. It was supposed to be a very smooth process, but it was found that a user clicked [Get Verification Code] three times in a row and then gave up submitting. This is strange, why would users click the verification code multiple times?At this time, I recommend that you try your product in person and go through the registration process. You will find that after clicking [Get Verification Code], you often don’t receive the verification code for a long time; then you will continue to click [Get Verification Code], so the above situation occurs.The vast majority of products have some anti-human designs or bugs to a greater or lesser extent. Through careful inspection by users, problems in the products can be easily discovered and resolved in a timely manner.5. Funnel AnalysisThe funnel is a tool used to measure conversion efficiency. It gets its name because the model from beginning to end resembles a funnel. There are two key points to note in funnel analysis: First, 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; second, 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.A certain company's registration process uses email, and its registration conversion rate has always been very low, at only 27%. Through funnel analysis, it was found that the main loss was at the [Submit verification code] stage.After investigation, we found that it is very easy for the registered email address to not receive emails during email verification. The reasons include the email agent being blocked, the email containing sensitive words being sent to the spam mailbox, the email taking too long to be delivered, etc. Since so many uncontrollable factors affect the registration conversion rate, let's try a different verification method. After switching to SMS verification, the overall conversion rate increased to 43%, which is a very large increase.6. Retention AnalysisRetention, as the name suggests, means that new users stay and continue to use the product. Common indicators for measuring retention include: next-day retention rate , 7-day retention rate, 30-day retention rate, etc. We can analyze retention from two aspects, one is the retention rate of new users, and the other is the retention of product functions.The first case: Taking the community website as an example, the retention rate of users who "check in 3 times a week" is significantly higher than that of users who "check in less than 3 times a week". The sign-in function has invisibly improved the stickiness and retention rate of community users, which is why many groups or communities promote this function.The second case: When you register on Weibo for the first time, Weibo will recommend you to follow 10 influencers; when you register on LinkedIn for the first time, LinkedIn will recommend you 5 colleagues; when you apply for a credit card, the issuer will say that you can enter a draw for a [drone] grand prize if you make 4 or more credit card transactions; many social products stipulate that if you sign in 5 times a week, users can get double points or virtual currency.Among them, "follow 10 big Vs", "follow 5 colleagues", "4 purchases", and "check in 5 times" are the Magic Numbers I want to talk about. These numbers are discovered through long-term data analysis or machine learning. Practice has shown that users who meet these characteristics have the highest retention rate; operations personnel need to constantly push and motivate users to meet this standard, thereby improving retention rate.7. A/B Testing vs. A/A TestingA/B testing is to achieve a goal by adopting two sets of plans, one group of users adopts plan A, and the other group of users adopts plan B. Through experiments, we observe the data effects of the two sets of solutions and judge the pros and cons of the two sets of solutions. When it comes to A/B testing, Google spares no effort in its attempts. For the display of search results, Google will develop a variety of different plans (including copy titles, font size, color, etc.) to continuously optimize the click-through rate of ads in search results.One thing to note here is that it is best to have A/A testing or similar preparation before 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. In fact, this is essentially the same as the controlled variable method, experimental group and control group, and double-blind trial in school.3. Process: Macro, Meso and Micro1. Macro1. Simple analytical philosophy in ancient ChinaIn fact, data analysis has existed since ancient times. Many famous people in ancient China were actually engaged in data analysis. Their titles may not be data analysts, but more like "prime ministers", "military advisors", or "counselors", such as Zhang Liang, Guan Zhong, Xiao He, Sun Bin, Guiguzi, and Zhuge Liang.They created great value for their organizations through "historical statistics" - "experience summary" - "predicting the future", which is an important part of the ancient Chinese simple analytical philosophy.2. MVP Concept of Lean StartupThe lean startup that is popular in Silicon Valley advocates the concept of MVP (minimum viable product), which aims to continuously optimize products and increase users by taking small steps and running fast.In our operations, we must be bold in experimenting and transforming ideas into products and operations methods . Then analyze the data to measure the effectiveness of the product or operation. If it is good, maintain it and promote it vigorously; if it is not good, summarize the problems and make timely improvements. Gradually optimizing in the continuous cycle of "build" - "measure" - "learn", this process is very suitable for operations work.2. MadhyamikaThe book "Who Says Newbies Can't Do Data Analysis" introduces a more specific analysis process: 1. Clarify the purpose and ideas of the analysis → 2. Data collection → 3. Data processing → 4. Data analysis → 5. Data presentation → 6. Report writing.This process only explains the previous and subsequent processes from the perspective of "data" and does not combine it with business reality; and it is misleading to position the foothold of data analysis at "report writing" because the ultimate goal of data analysis is to guide practice, not to write a report.However, this process still has reference value, especially the "clarification of analysis purpose and ideas" has certain guiding significance for novices.3. MicroThe following is a very detailed analysis process. With the help of certain analysis tools, we can conduct a detailed analysis of your website/APP according to this idea.This is the process introduced by GrowingIO business analyst Tan Runyang in "What are the methods for improving user retention, product, market and operation?" I think it is applicable to most operational data analysis.The premise is to use data analysis tools to carry out data collection and monitoring, and focus on business analysis.The core of this process is the "MVP" concept, "discovering problems" - "designing experiments" - "analyzing results", and continuously optimizing products and operations through data.IV. Application: System and Analysis(I) Case 1: Building a data analysis systemXiao Zhang just graduated this year and works in new media at a company, responsible for the daily operations of WeChat . Xiao Zhang was not clear about the core purpose of WeChat operations and tried many methods. He created, translated, and rewrote many articles and posted them on WeChat, but the number of readers fluctuated and was generally average.The manager asked Xiao Zhang to find a way to improve the operation of WeChat and increase the number of followers and readers of WeChat; but Zhang San had no idea where to start.This is a true portrayal of many operations. Trivial work can easily make people forget to think, and this may very well happen around you and me.We diagnosed this case from the perspective of data analysis and summarized the following problems faced by Xiao Zhang:
Not sure which core indicators you need to focus on;
Unclear about the characteristics of target users (user attributes, user profiles, etc.);
Lack of systematic analysis of one’s past work (data collection, monitoring and analysis).
From the perspective of business growth, I tailored a data analysis system for Xiao Zhang to support the development of his content work.The first point is content positioning.Operations need to clearly know their goals or KPIs, and then select a core key indicator (OMTM) for monitoring. If it is a startup company, it may need to attract new users in the early stages, so the core indicator is the number of registered users or the number of new visiting users. If it is an information media that focuses on influence and coverage, then the core indicator should be the number of WeChat readings or web page PV.The second point is user portrait.Regardless of which operations position you are in, you need to clearly know who your (target) users are? What are the characteristics of these people, and what are their concerns and pain points? If your user is a product manager , you can try to crawl relevant questions on the product manager's website and then do text analysis: this is a quantitative analysis.At the same time, more in-depth information on user characteristics can be obtained through surveys and questionnaires: this is an analysis at a qualitative level.Third, continuous monitoring.With the help of data analysis tools, the core key indicators (OMTM) are continuously monitored. For abnormal indicators, we need to analyze and improve them in a timely manner.Fourth, data analysis.Collect and analyze data from past content to find out which content, titles, formats, and channels are more effective, and then continuously optimize in this direction.Case 2: Analyzing core business indicatorsEmail marketing is a marketing and operation method that many companies still use. An Internet financial company sends activation emails to new users (users who have email addresses but are not registered) through EDM. The registration conversion rate has always been maintained between 20% and 30%. On August 18, the registration conversion rate plummeted and has remained at around 10% since then.This is a very serious recession and the causes need to be investigated immediately. The EDM channel registration conversion rate involves too many factors, which need to be checked one by one. The data analyst helped the operation to list the possible reasons:Technical reasonsProblems with ETL (data extraction, transformation, and loading) caused the backend data to not be presented in the BI report in a timely manner;Macro reasonsSeasonal factors (holidays, etc.), other email shocks (other departments also send emails to users, diluting users' attention);Micro-causeEmail title, copy, layout design, CTA design, and registration process design.A simple business indicator may be affected by a variety of factors, so we need to make detailed measurements of the factors involved in order to continuously optimize it. It was finally discovered that the product manager added a "credit card binding" step during the registration process, which resulted in a significant drop in the registration conversion rate.5. Learning: Business, Tools and Resources1. Business levelData analysis is not as unattainable as imagined. As long as you master the corresponding concepts, ideas, and processes, operations can do data analysis well. One point that needs to be emphasized here is that the purpose of data analysis is to guide business practice; data analysis that is divorced from practice or data analysis for the sake of analysis is just hooliganism.Unlike professional data analysts and data scientists, the prerequisite for operations personnel to do good data analysis is a sophisticated understanding of the business. From a business perspective, data is not just numbers, it is the voice of users. Operations personnel should identify problems from the data, continuously optimize, improve user experience, and create more value for users.2. Tool levelSharpening the knife does not delay the chopping of wood, and good data analysis tools are essential. I have summarized the following tools, which operations can adopt based on their actual needs.Excel is the most common and basic data analysis tool. The charts, functions, and pivot tables in Excel can meet everyone's basic needs. Access is part of the Microsoft Office suite and is a small relational database. When the amount of Excel data is large and there are frequent connections, queries, and updates between tables, Access is a very good choice.Python is a high-level programming language that has developed rapidly in recent years. It can be used for data analysis, programming or crawling; R language is a data analysis tool widely used in statistics. Currently, Python is widely used to write crawler programs to obtain information on the Internet, which is very helpful to operators.Google Analytics, Baidu Statistics, and Umeng are common website traffic analysis tools, while Mixpanel, Heap, and GrowingIO are user behavior data analysis tools, which have richer functions and more detailed analysis than the former.3. Resource levelGetting started with operational data analysis does not require learning complex mathematical theories, but is more about combining business operations with data analysis. I recommend two websites and two books here, I hope they can be helpful.
Mobile application product promotion service: APP promotion service Qinggua Media advertising
The author of this article @官世强 compiled and published by ( Qinggua Media ). Please indicate the author information and source when reprinting!
