【Data Analysis】: Three major ideas and methods of data analysis

Three ideas and methods of data analysis

In the previous blog post [What is data analysis], we introduced the basic concepts, processes, and methods of data analysis. In this article, let's take a look at the basic ideas of data analysis and common data analysis methods. In Internet analysis, the following three steps are basically followed:

• find out the problem
• analyse problem
• Solve the problem

Next, let's take a look at how to conduct a complete data analysis process.

1. Identify the problem

Step 1: Identify the problem, use descriptive statistics

The precise use of descriptive statistics can explain the basic situation and characteristics of the sample

How is the flow of reading good times?

Step 2: Identify problems, focus on changes, and find outliers

Traffic data, user data, conversion data, and transaction data in daily work have a fixed fluctuation cycle. The data changes in each cycle should be stable. If the data on a certain day does not change within the expected range, it is Data exception. The sensitivity to data anomalies mainly includes: Is data fluctuation abnormal? The scope of the anomaly, the degree of volatility, and whether in-depth analysis is required.

Methods of assessing change

Year-on-year: Comparison of changes over the same period (compared to last week)

Ring-to-Ratio: Comparing Continuous Cycles

Growth rate: an indicator that can effectively assess the cumulative type

2. Analyze the problem

The third step: subdivision indicators, multi-dimensional analysis values

Through multi-dimensional and split indicators analysis, we can find possible reasons for indicator changes.

• The same indicator, different dimensions
• Dismantling indicators, thinking deeply

Commonly used indicator dimensions:

User indicators : region, age, gender, education, device model, operating system (user portrait)

Product Metrics : Product Type

Operational Metrics : Activity Type

Marketing indicators : marketing channels, marketing methods

Regional dimension : overseas, domestic

Product dimension : children's books

Channel dimension : on-site advertising, advertising cooperation

Age dimension : age distribution

Through analysis, we found that the UV of APP users in overseas regions is gradually decreasing.

cross analysis

Focusing only on a single dimension and indicator will cause bias in our subjective judgment:

● Survivor bias

● Dimensional deviation: Although the UV decreases, the conversion rate may increase.

To analyze from multiple dimensions and multiple indicators.

Correlation Analysis:

The conversion rate is introduced as an indicator for cross-analysis. From the line chart, the relationship between conversion rate and UV is not large.

Step 4: Correlation analysis to find out the key factors of the problem

Correlation analysis analyzes two or more correlated variable elements to measure how closely two variables are related. Correlation analysis is difficult to be rigorous at the mathematical level, but has great interpretability at the business level.

Step 5: Anticipate trends and look to the future

In business, we usually judge the development trend of core business indicators through predictive analysis. Although predictive analysis can provide guidance for future trends, forecasting is only an estimate. The quality of data and the stability of business status determine the accuracy of forecasting, so this type of analysis often requires continuous optimization.

3. Solve the problem

Step 6: Actively communicate to promote business landing

We made the above analysis based on the data. Now we need to implement our conclusions and analysis logic with the business side to ensure that our analysis is correct and the conclusion can be accepted by the business side

How to turn business problems into data problems?

• Metrics: Identify key metrics for business problems and split them appropriately
• Dimension: Determine the latitude of the business problem, and build the problem around the latitude

Why does this problem occur?

Why analyze this?

Why is this important?

Common Analysis Methods

1. Funnel Model

The funnel model is a set of process data analysis models used to reflect the key behaviors of users in the process and the conversion and loss of each stage from the starting point to the end point

• process
• key behavior
• Conversion and Churn

value:

• Can help analysts locate weaknesses in the process
• It is conducive to multi-dimensional segmentation, capturing user behavior changes, and discovering abnormalities
• It is helpful to observe and compare the differences between different user groups, and continuously improve the user sports institute

How to Build a Funnel Model

• Sort out main paths and churn nodes

• selected core path

  • Choose a path with a large opening: such as Taobao to buy goods, search position, banner position, top ten diamond positions, operation position

  • Don’t have too many funnels: too many can’t focus on the problem

  • The amount difference between the funnel links should not be too large:

• Observe and compare data

  • Vertical comparison
  • Horizontal contrast

Case: E-commerce website uses funnel model to increase conversion rate

Because the payment is made directly after the checkout until the order is completed, so 20% here may be too low.

search list

• Search CTR (CTR on the first page of search results)

• Details page conversion rate

• Search with no results

• Ways to pay attention to search: search box search, voice search, recommended search

Order page and cashier page

• Effective order conversion rate

• UV transaction conversion rate

• Average UV value

• Need to pay attention to logistics, returns, invoices, shipping costs, payment methods, payment exceptions (wrong passwords, impulse purchases, technical issues…

Details page

• Average time on page

• Add to cart number

• Need to pay attention to the layout design of the details page, comments, customer service...

Looking at the curve of the order page conversion rate over the past six months, we found that the order page conversion rate dropped by more than 25% in June. In June, the Jingdong APP has started the "6.18" coupon grab to win discount campaign. Is the decrease in the conversion rate of the order page related to this campaign?

Horizontal comparison: look at users, find differences, and further verify assumptions

Through comparative analysis of users who participated in the event and users who did not participate in the event, we found that the conversion rate of the order page of the users who participated in the event was nearly 40% lower than that of the users who did not participate in the event, which preliminarily explained the conversion rate of the order page and the "6.18" activity There is a strong correlation. We took a closer look at the event details and found that this event is to issue coupons between 5 and 6 pm every day. The reason for choosing this time is that a large number of users commute during this time, and we have many active users during this time period.

According to the details of the campaign, we further compared the order page conversion rate between 5:00 pm and 6:00 pm on weekdays and weekends, and the results show that the order page conversion rate on weekdays is almost half of that on weekends. Considering that we have a large number of active users during this period, and a large number of active users are on the way to commute, then we can dig deeper in two directions:

• APP stability
• The user's network status has been communicated with the IT department, and the stability of the APP during the event is good, so let's further dig into the user's network status.

We compared the payment success in the dimension of the user's network environment, and found that about 20% of the users were in the 2G, 3G or even offline environment during this period, indicating that the user's network environment does have a great impact on the payment success rate. influences. Based on the conclusions we have obtained earlier, we can judge that those who participate in the activity during the activity period on weekdays may cause unstable network environment due to commuting or other reasons, thereby reducing the possibility of successful payment and reducing the conversion rate of the order page.

Jump out of the funnel and focus on user experience

We have analyzed the network conditions above.

Next, we jump further out of the funnel and focus on user experience

In the payment interface, it is always displayed that the payment is being made, and we do not know whether the payment is successful or unsuccessful (at this time, consider optimizing the interface, and the payment failure will be displayed directly, allowing users to re-pay. Or consider delaying the payment time during special events)

We can see that users with failed payments stay longer.

2. Matrix model

The Boston Matrix (BCG Matrix) is also known as the market growth rate-relative market share matrix, the Boston Consulting Group method, the four-quadrant analysis method, the product series structure management method, etc. The Boston Matrix is ​​a method of planning enterprise product portfolio pioneered by the large US business consulting firm - Boston Consulting Group. The Boston Matrix uses two indicators, sales growth rate and market share, to measure products. Sales growth rate and market share both influence and condition each other. Through the interaction of the above two factors, four different types of products will appear, forming different product development prospects:

• Product groups with "double high" sales growth rate and market share (star products)

• Product groups with "double-low" sales growth rate and market share (skinny dog ​​products)

• Product groups with high sales growth rate and low market share (children's products or problem products)

• Product groups with low sales growth rate and high market share (Golden Bull products)

Star products: high growth and high market share, good development prospects, strong competitiveness, need to increase investment to support its development;

Taurus products: low growth but high market share, mature market leader, should reduce investment, maintain market share and delay recession

Problem products: high growth but low market share, good development prospects but insufficient market development, need to invest cautiously;

Thin dog products: low growth and low market share, low profit margin or even loss, should adopt a retreat strategy

Four principles of the Boston matrix model:

• The principle of the moon : Our products are locked in problem products, star products, and Taurus products.
• Black ball principle: Sometimes, the company is in the initial stage of development and may not have cash cow products, so it needs to develop rapidly
• Diagonal principle: if on a star product, we want to switch to a Taurus product
• Transition principle: problem product –> star product –> golden bull product

Star users: The number of consumption and the unit price of the customer are higher than the average level, which is worthy of our focus, and increase the operation activities in a targeted manner to improve the loyalty of users

Taurus-type users: The unit price per customer is lower than the average level, but the consumption frequency is high, which can bring stable income to the enterprise, and generally does not require special attention

Problem users: The unit price per customer is higher than the average level, but the consumption frequency is low. For such users, push, email, coupons and other activities should be used to promote return visits and increase consumption frequency

Thin dog users: The unit price and consumption times are lower than the average level. You should further understand the user's pain points, and appropriately take personalized recommendation, tying and other activities to increase the user's unit price and consumption times.

3. Personalized Recommendations

Association rule-based recommendation (user-item) The premise of association rule-based recommendation is that the user has purchased a certain item, and then the recommendation is made according to the correlation between the user's purchased item and other items. The establishment of association rules is to use probability statistics to judge the correlation between two or more commodities.

Collaborative filtering recommendation Collaborative filtering recommendation is to use the nearest neighbor algorithm to obtain the similarity between users and users, and items and items to generate recommendation results.

● User and user (user-user): As long as you find out the items that similar users like and predict the target user's rating for the corresponding item, you can find several items with the highest ratings and recommend them to the user.

● Items and items (item-item): As long as the target user's ratings for certain items are found, we can predict similar items with high similarity, and recommend several similar items with the highest ratings to the user.