15 essential data analysis methods for operations

When it comes to data analysis , people often think of densely packed tables of numbers, advanced data modeling techniques, or gorgeous data reports.

In fact, "analysis" itself is a skill that everyone possesses, such as deciding whether to buy or sell based on stock trends; choosing a driving route based on daily time and past experience; and making a final choice after comparing prices from multiple companies when buying air tickets or booking hotels.

These small decisions are actually made based on the data points in our minds, which is the process of simple analysis. Business decision makers need to master a set of systematic, scientific, and business-compliant data analysis knowledge.

1. Strategic thinking in data analysis

Whether you are in product, marketing, operation or management, you must reflect on: What is the essential value of data? What can you and your team learn from this data?

1. Goals of Data Analysis

For enterprises, data analysis can help them optimize processes, reduce costs, and increase sales. We often define this type of data analysis as business data analysis. The goal of business data analysis is to use big data to make fast, high-quality, and efficient decisions for all professionals and provide scalable solutions. The essence of business data analysis is to create business value and drive business growth.

2. The role of data analysis

The corporate growth model we often talk about is often centered around a certain business platform. Among these, data and data analysis are indispensable links.

Products or services are provided to target user groups through enterprises or platforms, and the interactions and transactions generated by users during the use of products or services can be collected as data.

Based on these data insights, we can reverse engineer customer needs through analytical means, create more value-added products and services that meet these needs, and re-invest in user usage, thus forming a complete business closed loop. Such complete business logic can truly drive business growth.

3. Evolution of Data Analysis

We often position the different stages of data analysis based on business return ratio, so we divide it into four stages:

Stage 1: Observe what is happening with the data?

First, the basic data display can tell us what happened. For example: Last week, the company launched an advertisement on a new search engine A. It wants to compare the performance of the new channel A with the existing channel B within a week. How much traffic did A and B bring respectively? What was the conversion effect? For example, how many users like the newly launched product and how many people register in the new registration flow. All of these require data to demonstrate the results, and are based on the "what happened" provided by the data itself.

Stage 2: Understand why it happened?

If we see why channel A brings more traffic than channel B, we need to combine business to further determine the reasons for this phenomenon. At this time, we can further conduct in-depth segmentation through data information. Perhaps a certain keyword brings traffic, or perhaps the channel has acquired more mobile users. This kind of in-depth data analysis and judgment has become the second advanced level of business analysis, and can also provide more business value.

Phase 3: Predict what will happen in the future?

Once we understand the amount of traffic brought by channels A and B, we can predict what will happen in the future based on past knowledge. When launching channels C and D, we guess that channel C is better than channel D. When launching new registration flows and new optimizations, we can know which node is more likely to have problems. We can also use data mining to automatically predict and determine the differences between channels C and D. This is the third level of data analysis, predicting future results.

Phase 4: Business Decision

The most meaningful part of all work is business decision-making, using data to determine what should be done. The purpose of business data analysis is business results. When the output of data analysis can be directly converted into decisions, or decisions can be made directly using data, then the value of data analysis can be directly reflected.

4. EOI framework for data analysis

The EOI framework is the basic way many companies, including LinkedIn and Google, define the goals of analytical projects. It is also a basic and necessary means for chief growth officers to think about business data analysis projects.

Among them, we will first divide the company's business projects into three categories: core tasks, strategic tasks, and risk tasks.

Take Google as an example. Google’s core mission is search, SEM, and advertising. This is a proven business model and has continued to make a lot of profits from it.

Google's strategic mission (around 2010) was the Android platform. In order to avoid being occupied by Apple or other manufacturers, it had to spend time and energy on it, but the business model might not necessarily take shape. Risky tasks are very important for innovation, such as Google Glass, self-driving cars, etc.

The goals of data analysis projects for these three types of tasks are also different. For core tasks, data analysis is an enabler (E), helping companies to make better profits and improve profit efficiency; for strategic tasks, it is an optimization (O), how to assist strategic tasks in finding directions and profit points; for risk tasks, it is a joint venture (I), working hard to verify the importance of innovative projects. The Chief Growth Officer needs to have a clear understanding of the company's business and development trends, reasonably allocate data analysis resources, and set data analysis goals.

2. Three major ideas for data analysis

Faced with massive amounts of data, many people don’t know how to prepare, how to conduct the work, or how to draw conclusions. Here are three classic ideas for data analysis. I hope they can help you in the actual application of data analysis.

1. Basic steps of data analysis

Above we mentioned the importance of the connection between data analysis and business results. All business data analysis should start with business scenarios and end with business decisions. What should be done first and what should be done later in data analysis. Based on this, we propose five basic steps in the business data analysis process.

1. You need to first explore the business meaning and understand the background and premise of data analysis as well as the business scenario results you want to associate with it.
2. An analysis plan needs to be made, including how to split the scenario and how to make inferences.
3. Separate the required data from the analysis plan and actually implement the analysis itself.
4. From the data results, determine and extract business insights.
5. Insights are gained based on data results and ultimately business decisions are made.

For example:

The marketing department of a domestic Internet financial management website has been continuously placing advertisements on Baidu and hao123 to attract web traffic. Recently, internal colleagues suggested trying to place ads on Shenma Mobile search channels to gain traffic; in addition, we also need to evaluate whether to join the Kingsoft Network Alliance for in-depth advertising.

How to make in-depth decisions in this multi-channel delivery scenario? Let’s break down this problem according to the five basic steps of the business data analysis process above.

Step 1: Uncover business implications.

First, you need to understand what the marketing department wants to optimize and use this as the North Star indicator to measure it.

For channel effectiveness evaluation, what is important is business conversion: for P2P websites, whether to initiate "investment and financial management" is far more important than the "number of visiting users".

Therefore, whether it is Shenma Mobile Search or Kingsoft Channel, the focus is on how to measure the conversion effect through data means; it can also further optimize the operation strategies of different channels based on the conversion effect.

The second step is to develop an analysis plan.

Taking "investment and financial management" as the core conversion point, allocate a certain budget for traffic testing, and observe and compare the number of registrations and the final conversion effect. Write down the number of times these people repeatedly purchase financial products and continue to pay attention to them to further judge the quality of the channel.

The third step is to split the query data.

Since the analysis plan requires comparing channel traffic, we need to track traffic, landing page dwell time, landing page bounce rate, website visit depth, orders and other types of data from each channel for in-depth analysis and implementation.

The fourth step is to extract business insights.

Based on the data results, we compared the effects of Shenma Mobile Search and Kingsoft Network Alliance after launching the campaign. Based on the two core KPIs of traffic and conversion, we observed the results and inferred the business implications.

If the mobile search results for Shenma are not good, you can think about whether the product is suitable for the mobile customer base; or carefully observe the performance of the landing page to see if there is content that can be optimized, etc., to find business insights.

Step 5: Produce business decisions.

Guide channel decision making based on data insights. For example, stop advertising through the Shenma channel and continue to follow up with Kingsoft Network Alliance for evaluation; or optimize the mobile landing page, change the user operation strategy, etc.

The above are the basic steps to deconstruct business data analysis and complete inferences. In the following content, we will have this analytical idea.

2. Internal and external factor decomposition method

In the process of data analysis, there will be many factors that affect our North Star indicators, so how do we find these factors? Here I recommend the internal and external factors decomposition method to everyone. The internal and external factor decomposition method is to break the problem into four parts, including internal factors, external factors, controllable and uncontrollable factors, and then solve each problem step by step.

For example:

A social recruitment website, divided into job seeker side and enterprise side. Its profit model is generally to charge businesses, and one of the charging methods is to purchase advertising space for job positions. Business personnel have found that the number of "job postings" has been slowly declining over the past 6 months. How can we analyze this kind of problem of a decline in a certain data indicator?

According to the internal and external factor decomposition method, we can analyze the possible influencing factors from four angles in turn.

1. Internal controllable factors: recent product launch updates, changes in market channels, product stickiness, retention issues for new and old users, and conversion of core goals.
2. External controllable factors: recent behavior of market competitors, changes in user usage habits, and changes in recruitment needs over time.
3. Internal uncontrollable factors: product strategy (mobile/PC), overall company strategy, and company customer group positioning (for example, only recruiting in the medical industry).
4. External uncontrollable factors: Internet recruitment industry trends, overall economic situation, and seasonal changes.

With the internal and external factor decomposition method, we can analyze data indicators more comprehensively, avoid possible missing influencing factors and prescribe the right remedy.

3. DOSS idea

The DOSS approach is to break down a specific problem into its overall impact and find a scalable solution from a single solution. Chief Growth Officers need to scale effective growth solutions quickly, and DOSS is an effective way to do so.

For example:

An online education platform provides free course videos and sells paid memberships, providing paid members with more advanced course content. If I want to push a set of paid computer technology courses to a group of users who continue to watch free C++ courses, how should data analysis be supported?

We break down the four steps of DOSS thinking as follows:

1. Specific problem: Predict whether a group of customers is likely to purchase a course.
2. Overall impact: First, we conduct data analysis and data mining predictions based on the usage of free courses by this group of people, and then extend it to, for example, the overall impact, focusing on other types of courses in addition to computer courses.
3. Single answer: Build a model for this group of users and monitor the impact of the model on final conversions.
4. Scaling solution: Later, we will launch a scaled solution to model behaviors that match certain behavioral trajectories and characteristics, and productize the course recommendation model

8 methods of data analysis

The above introduces three classic analysis ideas, which can help you build a clear data analysis idea framework. So what should we do about specific business scenario problems? We take an e-commerce website as an example, and use the data analysis product GrowingIO to quickly collect data from the website, and clearly and visually display it, and then share these 8 common data analysis methods with you.

1. Numbers and trends

Looking at numbers and trends is the most basic way to display data information. In data analysis, we can quickly understand market trends, number of orders, performance status, etc. through intuitive numbers or trend charts, so as to intuitively absorb data information and help make decisions more accurate and real-time.

For e-commerce websites, traffic is a very important metric. In the above figure, we aggregate indicators such as the number of visitors (UV) and page views (PV) of the website into a unified data dashboard and update it in real time. Such a data dashboard shows the core figures and trends at a glance, which is easy for the chief growth officer to understand.

2. Dimensional 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. When choosing dimensions, you need to think carefully about their impact on the analysis results.

For example, when you detect abnormal website traffic, you can find the problem by splitting it into dimensions such as region, access source, device, browser, etc. In Figure 7, the number of users visiting the website on that day was significantly higher than that of last week. What is the reason? When we split the traffic by access source (Figure 9), we can easily find that the traffic from direct access sources has increased significantly, which further focuses on the problem.

3. User Segmentation

Classifying users who have certain behaviors or background information is what we often call user segmentation. We can also create a profile of a group of users by extracting their specific information. For example, users who visit shopping websites and whose shipping addresses are in Beijing can be classified as the "Beijing" user group. For the "Beijing" user group, we can further observe the frequency, category, and time of their product purchases, so that we can create a portrait of this user group.

In data analysis, we often conduct targeted user operations and product optimization for users with specific behaviors and backgrounds, and the effect will be more obvious. In the above figure, we use GrowingIO's user grouping function to select users who failed to pay in a promotion, and then push corresponding coupons. Such precise marketing promotion can significantly increase users' willingness to pay and sales amount.

4. Conversion Funnel

Most business monetization processes can be summarized as a funnel. Funnel analysis is one of our most common data analysis methods, whether it is a registration conversion funnel or an e-commerce order funnel. Funnel analysis can be used to restore the user conversion path from the beginning to the end and analyze the efficiency of each conversion node.

Among them, we often focus on three key points:

1. What is the overall conversion efficiency from start to finish?
2. What is the conversion rate for each step?
3. Which step has the most loss and why? What are the characteristics of lost users?

The registration process in the above figure is divided into 3 steps, and the overall conversion rate is 45.5%; that is, there are 1,000 users who come to the registration page, and 455 of them successfully complete the registration. However, it is not difficult to find that the conversion rate of the second step is 56.8%, which is significantly lower than the conversion rates of the first step (89.3%) and the third step (89.7%). It can be inferred that there is a problem with the second-step registration process. It is obvious that the second step has the largest room for improvement, and the return on investment is definitely not low; if we want to increase the registration conversion rate, we should prioritize solving the second step.

5. Behavioral Trajectory

Paying attention to behavioral trajectories is to truly understand user behavior. Data indicators themselves are often just an abstraction of the real situation. For example, if you only look at indicators such as the number of visiting users (UV) and page views (PV) in website analysis, you will definitely not be able to fully understand how users use your product.

Restoring the user's behavior trajectory through big data can help the growth team focus on the user's actual experience, discover specific problems, and design products and deliver content based on user habits.

The above picture shows the detailed behavior trajectory of a user on an e-commerce website, from the official website to the landing page, then to the product details page, and finally back to the official website homepage. The website purchase conversion rate is low, and previous business data cannot tell you the specific reason; by analyzing the above user behavior tracks, you can discover some product and operation problems (such as whether the products do not match, etc.), thereby providing a basis for decision-making.

6. Retention Analysis

In an era when the demographic dividend is gradually fading, the cost of retaining an old user is much lower than acquiring a new user. Every product and every service should focus on user retention and ensure that every customer is realised. We can understand retention through data analysis, and we can also find ways to improve retention by analyzing the relationship between user behavior or behavior groups and return visits.

At LinkedIn, the growth team discovered through data that if a new user adds more than five contacts (the red line in the above figure), his/her retention on LinkedIn will be much higher than those who do not add contacts (the green and purple lines in the above figure). In this way, adding contacts becomes one of the core means for LinkedIn to retain new users. In addition to paying attention to the overall user retention, the marketing team can pay attention to the retention of users acquired through various channels, or the return rate of registered users attracted by various types of content. The product team can pay attention to the impact of each new feature on user return visits, etc. These are all common retention analysis scenarios.

7. A/B Testing

A/B testing is used to compare the impact of different product designs/algorithms on results. A/B testing is often used during the product launch process to test the effectiveness of different product or feature designs. Marketing and operations can use A/B testing to evaluate the effectiveness of different channels, content, and advertising creativity.

For example, we designed two different forms of product interaction and evaluated which form of interaction was better by comparing the visit time and page views of the experimental group (Group A) and the control group (Group B).

There are two essential factors for conducting A/B testing: first, sufficient time for testing; second, a high volume and density of data. Because when the product traffic is not large enough, it is difficult to get statistical results by doing A/B testing. A large company like LinkedIn can conduct thousands of A/B tests simultaneously every day. Therefore, A/B testing is often more accurate and can produce statistical results faster when used when the company's data is large.

8. Mathematical modeling

When a business goal is related to multiple behaviors, portraits and other information, we usually use mathematical modeling and data mining to build a model and predict the business results.

As a SaaS company, when we need to predict and determine customer churn, we can build a churn model based on user behavior data, company information, user portraits and other data. By using statistical methods to perform some combinations and weight calculations, we can find out which behaviors will increase the possibility of user churn.

We often say that if you can’t measure, you can’t grow. Data analysis plays a vital role in enhancing the business value of an enterprise. Of course, it is far from enough to just master the theory, practice is the only way to gain true knowledge. You may want to try using data analysis methods in your daily work and analysis-related projects. I believe you can achieve twice the result with half the effort and create more business value.

Author: Chen Ming

Source: Qinggua Media