Alibaba eCommerce Analysis

Avatar
Posted on Sep 3, 2019

Team Alibaba 

Lan Mond | Yuqin Xu | Maomao Yi | Xiaofeng Zeng (Fred)

Project & Dataset background:  

Alibaba Group Holding Limited is a Chinese multinational conglomerate holding company specializing in e-commerce, retail, Internet, and technology.  Tmall.com is one of Alibaba's online trading platform which provides a B2C trading platform for officially registered business and their customers.

Huitong (shorten as HT) is one of the 267,631 registered stores on tmall.com. Started from 2010 , HT sells backpack, toys etc. and had its peak performance through the year but recently, the business is not doing well.

This project aimed to conduct analysis into HT's historical sales records and locate the problems and possible solutions to improve its performance. 

Dataset

  • 2010 ~ 2019
  • about 143,000 order records
  • 30 collected features (60 features started from 2019)
  • selling products: backpacks and cases for school, electrical puzzles,  Toys(including creative thinking enhancement)
  • end users:  kindergarten ~ 12th G
  • buying customers:  retail customer (parents ) and whole-sale stores.

Data Cleaning:

  • Missing data imputation
    • CellPhone = AlipayAccount 
    • Null ==> No (For note and messages)
  • Delete rows:
  • Duplicates
  • Exceptions (Shipping only orders)
  • Delete columns:
  • ContactPhone
  • StoreID
  • StoreName
  • DeliverNo 
  • New columns: 
  • Address  ==> Province & City 
  • OrderToPay(s)
  • PayToReceive(h)

(includes but not limited to the above)

Sales Performance

Historical Sales Trend

Y- Monthly sales amount
Accumulated sales: ¥20,704,459 
Best month: May-2015 sales: ¥930,000

Order Count Trend

Y - Order Count/Monthly

Sales -- By Region


Top five provinces:  JiangSu, Shanghai, Shandong, Guangdong
(江苏省 浙江省 上海市 山东省 广东省)
Top 5 provinces on the east coast generated 53% revenue

Sales -- By Product


146,162 products sold which fall into 8439 distinct Items
The top 30 items generated 36.6% of revenue

Order Activity Pattern

Monthly Pattern

Weekly Pattern

Tuesday and Monday top the Weekend

Daily Pattern (2010~2019)


The most orders are generated in 10pm ~ 12am
Follows by lunch break 12pm~ 2pm

Daily Pattern (2019)


93.5% total orders were placed  by cell phone
37% orders fell into  6pm-12am

Order Placement & Payment Process

Time lapse between order placement and payment
Average: 19.5 seconds
Maximum: 27 seconds

Visitor & Customer Behavior

Visitor Behavior

Y - percentage of visitors who made purchase after visiting the HT Store

Customer Behavior


117,678 registered customers
26,252 Customers came back & bought one or more other items in 30 days
77.7%  customers left after first purchase

Customer Review - Word Cloud

Reviews - Successful Orders
Reviews - Cancelled Orders

General Observations:

  1. Best performance 2015
  2. Bad performance 2016
  3. Sales volume clearly declined in 2019
  4. Top 5 selling regions are on the east coast and generated 53% of total sales. (there are 35 provinces + admin territories in China)
  5. Top 30 products out of 8439 (distinct) generated 36% of total sales
  6. less than 2% of the visitors actually converted to a customer
  7. 77.7% customers left after purchased the first item
  8. Average time from order placement to payment is 19.5 seconds

Industrial Comparison

We did research and acquired 2017 ~ 2019 data of both industrial average performance and of the top 10% competitors' performance. Thus we made the following comparison:

Visitor Conversion Rate

Conversion Rate: number of new customers / total store visitors
Conversion Rate Gap
The biggest gap is more than 5 times
Gap = (TOP(AVG) CVR - HT_CVR)/ HT_CVR

Transaction Volume

Transaction volume is decreasing, especially in 2019
Performance clearly bellows industrial Average
Gaps between industrial average & Top 10 competitors

Increase visitor volume 
Increase order quantity / visitor

Conclusions:

  1. HT visitor conversion rate is very much below the industrial average level for instance in 2018: 1.76% vs 3.38%. If HT can improve its conversion rate to industrial level, with an assumed profit rate 10%, HT can bring in extra profit ¥193,386.53
  2. HT transaction volume is clearly below the industrial average performance

HT can invest in Tmall promotion to increase the number of visitors thus increase sales. Here for this project, we focus on current customers: convert inactive customer to active customer and increase per customer sales.

Customer Retention

1. Return Pattern

  1. 117,678 Customers
  2. 22,969 Customers came back in 30 days
  3. 26,252 Customers came back and bought other items

We can see from the graph that most customers purchased the second item in 30 days time period. If we can predict the possibility of the return of a customer, then we can follow up with specific policy to encourage the return. Thus increase per customer transactions.


2. Customer Retention Prediction

  1. Objective :  Predict the probability of consumer return rate in a 30 day period.
  2.  Model    :  Logistic Regression
  3.   Result    :
 Predict 0Predict 1
True 0787312930
True 1113184317
  • 0: Not come back in 30 days
  • 1: Come back in 30 days
  • Accuracy: 85%
  • Recall = TP/(TP+FN) = 28%

We first split the dataset to 80% : 20% and created a logistic regression classifier with the training set. The prediction on a customer's return in 30 days period on the test dataset shows in the table shows an accuracy of 85% but poorly on Recall which is TP/(TP+FN) = 28%. What cause this poor performance ?

Recall that 77.7% of the customers never came back after the first purchase. This indicates that the our dataset is very imbalanced between return & non_return which is 77.7 : 22.3. In this case, we tried three mathematical techniques to improve the efficiency of the model

1. Cut off

Raise the threshold of classifier to 0.6 ~ 0.8 in stead of the default value of 0.5

2. Up-sampling

The theory is to fabricate synthesized minority class data to reduce the imbalance between the majority and minority data. return. In this case, we up-sampling bootstraps only the return customer samples to make the major:minor ratio more balanced.


3. Class_Weight

Classifier Recall Score
Classifier Accuracy Score


3.Market Basket Analysis

  • Market basket analysis
    • Uncover associations between items 
    • Looking for combinations that occur together frequently 
    • Provide information about purchase behavior
  • Use NLP on Itemname
    • Create 8 most popular categories
    • 93% item names are in those categories
  • Use 1 or 0 to represent each consumer’s purchase behavior
  • Apriori algorithm
    • Calculate confidence of buying 2 products together
    • Rank top 10 confidence results
The above chart shows the confidence in customer's purchase of consequents after purchased the antecedents


Also with data engineering, we were able to locate total 482 product combinations customers purchased and found 84 most popular combinations.


With the above tools, we will be able to follow up with specific promotion policy to encourage the potential purchase of the current customers.

Recommendations

  1. UI structures & features improvement
  • Especially item interface pulled by search engine on the platform. 
  • Research on the top competitors’ design performance

        Convert more random visitors to customers

      2. Inactive customer recover 

  • Database
  • Promotion policies accordingly
  • Reach out to customers

         Bring inactive accounts back to life

     3. Set up customer loyalty reward system 

  • Accumulated purchase amount - wholesale customer
  • Accumulated purchase times - retail customer

         Drive increasing activities on the active accounts

     4. Set up a systematic filing & tracking system

  • Customer follow up
  • Customer feedback
  • Birthday gifts  (such as store branded backpack hanging toys. For different ages) 

         Maintain & increase customer satisfaction rate

     5. Research on tmall.com promotion policies 

  • Target windows
  • Target customer groups
  • Target season

         Generate higher visitor traffic

 

 

Further Research Suggestion

1.Further analysis with AB test on UI 

Deep research on top competitors UI:  page structure, graphic, promotion, product group and item links etc. Thus create systematic AB test plans

2.Further analysis on the 8400 accumulated sold items with NLP

Add catch more features such as age group to the item groups; Together with improved logistic regression model to form a more efficient and accurate group product recommendation system

3.Further analysis on customer reviews with NLP

Help determine more effective marketing approaches

4.Further analysis into the regional features 

Between east coast (top 5 provinces, 53% revenue) &  western regions(30 province, 47% revenue) to drive regional expansion into the vast western market

5.Further analysis into city & town level sales density 

Indications of group customer influences such as school clusters. Thus specific regional group marketing plan may be considered: such as the purchased backpack being delivered with store branded fashion toys (hanging on backpack) .  

About Author

Avatar

Lan Mond

Recently certified as Data Scientist and Masters in Electrical Engineering alongside with rich international business experience in helping companies to gather and analyze data to make more informed decisions regionally and globally to achieve their business goals while...
View all posts by Lan Mond >

Leave a Comment

No comments found.

View Posts by Categories


Our Recent Popular Posts


View Posts by Tags

2019 airbnb alumni Alumni Interview Alumni Reviews Alumni Spotlight alumni story Alumnus API Application artist aws beautiful soup Best Bootcamp Best Data Science 2019 Best Data Science Bootcamp Best Data Science Bootcamp 2020 Best Ranked Big Data Book Launch Book-Signing bootcamp Bootcamp Alumni Bootcamp Prep Bundles California Cancer Research capstone Career Career Day citibike clustering Coding Course Demo Course Report D3.js data Data Analyst data science Data Science Academy Data Science Bootcamp Data science jobs Data Science Reviews Data Scientist Data Scientist Jobs data visualization Deep Learning Demo Day Discount dplyr employer networking feature engineering Finance Financial Data Science Flask gbm Get Hired ggplot2 googleVis Hadoop higgs boson Hiring hiring partner events Hiring Partners Industry Experts Job Job Placement Jobs Jon Krohn JP Morgan Chase Kaggle Kickstarter lasso regression Lead Data Scienctist Lead Data Scientist leaflet linear regression Logistic Regression machine learning Maps matplotlib Medical Research meetup Networking neural network Neural networks New Courses nlp NYC NYC Data Science nyc data science academy NYC Open Data NYCDSA NYCDSA Alumni Online Open Data painter pandas Part-time Portfolio Development prediction Prework Programming PwC python python machine learning python scrapy python web scraping python webscraping Python Workshop R R language R Programming R Shiny r studio R Visualization R Workshop R-bloggers random forest Ranking recommendation recommendation system regression Remote remote data science bootcamp Scrapy scrapy visualization seaborn Selenium sentiment analysis Shiny Shiny Dashboard Spark Special Special Summer Sports statistics streaming Student Interview Student Showcase SVM Switchup Tableau team TensorFlow Testimonial tf-idf Top Data Science Bootcamp twitter visualization web scraping Weekend Course What to expect word cloud word2vec XGBoost yelp