Shoppers’ annual Spring Festival Gala “Double Eleven” shopping carnival, how many orders have you placed?
Although the editor didn’t buy too much this year, it can still be regarded as a contribution to the “Double Eleven”!
Double Eleven is an annual promotion season for e-commerce retail. All e-commerce companies and brands are gearing up and firing on all cylinders. If brands want to take the lead in the e-commerce war, consumer insights are essential. The more thoroughly the brand understands consumers' purchasing behavior, the more effectively it can help companies carry out more targeted marketing activities.
This article uses data analysis to analyze several key factors of e-commerce, explain which factors may affect sales, and help e-commerce retail find insights to increase sales.
Today, we will use the literature Sakar, CO, Polat, S.O., Katircioglu, M. et al. A public sample data from Neural Comput & Applic (2018), combined with JMP software, is used to conduct some interesting exploratory data analysis on the purchasing behavior of online shopping users.
Step.1 Data acquisition
original source
Original data can be obtained through URL Download the original data in csv format and open it through JMP; you can also directly use the web page reading function of JMP to directly obtain the web page data.
(JMP operation: File -> Open from internet -> Web page)
Information background introduction
This data set contains the purchasing behavior of 12,330 online shopping users on the website within one year, as well as the corresponding 17 user behavior records and final transaction results. There are 17 behavioral records, including 10 numerical features and 7 categorical features.
- Basic information
Management webpages, management stay time, information webpages, information stay time, product webpages, product stay time, represent the total number of users opening and stay time on different types of webpages.
- Bounce Rate
Bounce rate indicates what percentage of users who enter a website page from a specific path without doing anything and simply leave the website. It can be used as a measure of the entire website or as a measure of the page.
- exit rate
The exit rate indicates that for a specific page, the percentage of people leaving the website from this page accounts for all visits to this page. It is generally used as a measure of the page;
- page value
Page value represents the average number of pages a user visited before completing a transaction;
- special day
Special days represent the interval between website access time and specific special days;
- other
In addition, it also includes the operating system, browser, region, traffic type, visitor type used by the user, whether it is a weekend and the month of the year.
Step.2 Visual exploratory analysis
The level of bounce rate is related to the success of online marketing.
Customers only view a single page and then exit, making it difficult for brands to convince consumers to purchase products since they only viewed one page. Let’s take a look at what bounce rate looks like below.
(JMP Operation: Analysis -> Distribution)
As can be seen from the picture, 90% of customers have a bounce rate of less than 6%, and the average bounce rate of all users is only 2%. Isn’t it perfect? Please don't be too anxious to be happy yet. It could be inaccurate or there could be a technical issue with how the website's analytics tracking code is integrated into the website.
Because based on past experience, the "normal" bounce rate is between 40% and 60%. Below 40% is very rare, and above 70% is worrying and requires urgent action.
Bounce rates that are currently outside of the expected range and seem "too good to be true" are likely caused by duplicate analytics code somewhere on the site.
Multiple exploration of user numerical characteristics
When doing shopping behavior analysis, the user's numerical characteristics may have many dimensions. With the help of JMP's multivariate analysis method, the relationship between each dimension can be quickly discovered, and it is possible to implement dimensionality reduction operations, which will reduce the time and effort for subsequent feature monitoring. Necessary waste of resources.
(JMP Operation: Analysis -> Multivariate Methods -> Multivariate)
Based on current data, there is a positive correlation between the number of times a user opens different types of web pages and the time they stay on them. This is easy to understand. Bounce rate and exit rate also have a significant positive correlation because of similar calculation formulas. In addition, no obvious numerical characteristic correlation was found.
The impact of how users access the Internet on sales
Through the statistical results of the chi-square test in the figure below, we can capture the relationship between whether a transaction is completed and the customer's operating system, browser type, and traffic type.
( JMP Operation: Analysis -> Take X FitY )
As far as operating systems are concerned, the proportion of users completing transactions is different for different operating systems.
It can also be seen from the graph that when the operating system is "2", it is slightly higher, while "1" and "3" are on the lower side. This may mean that the website page is not friendly enough to support these operating systems. If you want to improve this part revenue conversion, corresponding improvements need to be made. Similarly, for browser type and traffic type, we have also seen their statistical impact on the proportion of users completing transactions, indicating that the website also has room for improvement in this regard.
New and old customers and working days / Impact of weekends on sales
With the help of the chi-square test method in Part 2.3, we can also quickly discover the sales patterns of some new and old customers and weekdays/weekends, but here we try another method of data table summary.
(JMP Operation: Analysis->Consumer Research->Classification)
Combining the above graphics and information, we can clearly see: Old customers are the main source of website visits , indicating that the website has done a good job in customer retention; but we also see that the proportion of new customers completing transactions is higher whether on weekdays (13.2% vs 26.1%) or on weekends (16.5% vs 21.9%) For old customers, this shows that the website can make some improvements in the conversion rate of old customers.
For example, when old customers purchase goods, they can enjoy greater discounts by introducing new customers. This not only arouses the enthusiasm of old customers for purchasing, but also adds more new customers to the website.
Centralized display of diverse analysis results
If you want to display various analysis charts in the form of a report, you can use the script function of JMP to update the report link data in real time with one click, saving a lot of repetitive manual operations.
(JMP operation: File->New->Application)
The above are the results of some exploratory data analysis. Next, we will further try to use data collection methods to establish a quantitative statistical model for the main user characteristics and transaction results mentioned above.
Step.3 Data Predictive Modeling
Screen the key features that affect sales through decision trees
Decision tree is a supervised learning method that can summarize decision rules from a series of data with characteristics and labels, and present these rules in a tree diagram structure. The results are easy to interpret and have wide application in various industries and fields. Application.
(JMP Operation: Analysis -> Predictive Modeling -> Segmentation)
From the diagram we can observe the various stages of the decision tree, showing from top to bottom the most important features that influence the trading outcome.
The most important one is the value of the web page. The transaction ratios of web page values lower than 0.067 and higher than 0.067 are 3.85% and 56.4% respectively, with obvious gaps. There are also some more important characteristic parameters listed later, such as bounce rate, month and product-related pages, etc. These information are important factors that drive transaction results and can now be quickly selected and displayed, allowing brands to have A clearer focus for improvement and optimization.
Selecting the best from the best, trying more data collection methods
In addition to decision trees, JMP also provides a variety of data collection methods such as neural networks, random forests, boosted trees and support vector machines, and can easily complete comparisons between model algorithms to achieve the best of the best.
Through JMP Pro 16's new model screening, multiple data collection methods can be compared at one time. Before this, in order to prevent the built model from overfitting, you can first follow the training set, test set, and verification set in the ratio of 6:2:2 Proportionally split the original data and generate verification columns.
(JMP operation: Analysis->Predictive Modeling->Generate Validation Column)
As shown in the figure below, once you input the characteristic parameters and transaction results, the platform will build multiple algorithm models at the same time, and automatically select the current best modeling method as random forest. The performance of the model on the test set is an indicator of the future. The predictive ability R-squared of the new data reached 0.6, and the prediction accuracy reached 90.7%.
Accurate prediction models can help brands understand the possible transaction results of each user as early as possible, especially predict transaction failures, and make early responses and remedies, such as discounts, or popping up a retention interface before customers exit the page.
(JMP operation: Analysis->Predictive Modeling->Model Screening)
How about it? After reading today’s analysis, does it make you more confident in the “e-commerce war”?
For brands, providing personalized product, service and commodity recommendations, and conducting one-to-one marketing based on individual needs is a huge advantage of online marketing over traditional marketing.
Combined with JMP data analysis software, you can conduct in-depth analysis of the consumption behavior of online shopping users, which can help companies design product collection pages and promotional activities that better meet the needs of target customer groups, and make timely responses to potential problems discovered. improvements, thereby bringing greater benefits to the enterprise.
>> Download JMP 16 now and get a 30-day free trial <<
