\
This user behavior report is based on users’ orders from Alibaba between November 25th, 2017, and December 3rd, 2017 from the Alibaba platform. There are around 1 million customer activities in our dataset.

\
Dataset: <https://github.com/kirstenpan/Personal_Project/blob/96493ec3bed25a6862ffd69cbf4770667cdb08f5/alibaba_user.csv.zip>

\
We only have one table for analysis. The table has one million rows (events) and five columns originally (excluding the column `name` that we created as the primary key). Here is a brief description of our table:

\
· `id` is the primary key column for this table. Since the original dataset doesn’t have a primary key, so we have to generate an auto-increment primary key for the table.

· `user_id` is the id for each user.

· `product_id` is the id for each product.

· `product_category_id` is the id for each product category

· `type_of_behavior` describes users’ actions, which include “pv”, “buy”, “cart”, and “fav”

· `timestamp` describes the timestamp of each event

 ![](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-bzj3o68.png)

# Step 1: Create Tables ALI_USER_STAGE and ALI_USER in MySQL Database: Notes: Why does a staging table necessary in data ETL?

A Data Warehouse Data Staging Area serves the dual functions of storing unprocessed data from Source Systems and providing storage for temporary tables produced during the transformation process. In other words, all of the information needed to update or insert into the target database is stored in the staging table (a temporary table) before inserting into the data warehouse.

\
`CREATE TABLE 'ALI_USER_STAGE" (`

`"user_id' numeric, `

`"product_id" numeric, `

`"product_category_id" numeric, `

`"type_of_behavior" varchar(255) `

`"timestamp" numeric)`

\
`CREATE TABLE 'ALI_USER' ( `

`"id" int NOT NULL AUTO INCREMENT, `

`"user_id" numeric, "product_id" numeric, `

`"product_category_id" numeric, `

`"type_of_behavior" varchar(255), `

`"timestamp" datetime, `

`PRIMARY KEY ("id"))`

# Step 2: After importing data to ALI_USER_STAGE in MySQL Database, insert data into ALI_USER:

`INSERT INTO ALI_USER(user_id, product_id, product_category_id, type_of_behavior, timestamp)`

`SELECT user_id, product_id, product_category_id, type_of_behavior, FROM_UNIXTIME(timestamp, ‘Y-%m-%d %H:%m:%S’) FROM_ALI_USER_STAGE`

\
# Step 3: Metrics Design

 ![](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-wxh3o6r.png)

# Step 4: Data Analysis

## Hypothesis №1: Customers tend to be more active outside of their working hours (in the data source's local timezone)

`SELECT HOUR(timestamp) AS hour, COUNT (user_id) AS hourly_total_customer FROM ALI_USER GROUP BY hour ORDER BY hour`

\
 ![](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-tgg3o2z.png)

\
Based on the Customer Count by Hours metric, the peak of the number of customers visiting the platform is between 8 pm to 10 pm in a day, and the lowest is during midnight which is around 2 am to 5 am. It is obvious that customers tend to visit the platform outside their working hours and 8 pm to 10 pm is the time that customers have more free time to focus on shopping online. For this reason, Alibaba should target its marketing campaign at the right time because the time customers spend on the platform is precious.

\
## Hypothesis №2: Customers tend to spend more time on the platform during the weekend

`SELECT DATE_FORMAT(timestamp, "%m-%d-%Y %W") AS date, COUNT(*) AS visit_count FROM ALI_USER WHERE timestamp BETWEEN '2017-11-24' AND '2017-12-03' GROUP BY date ORDER BY visit_count DESC`

\
 ![](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-jbf3o9n.png)

From the Tableau area above, there is an increasing trend in visit counts starting from December 1st, 2017. This is because the coming Alibaba sitewide discount “12/12” (December 12th, 2017) activate many silent customers. Before December 2nd, 2017, the trend of the whole graph presents flatly, we can conclude that customers are not spending a lot more time on the platform during the weekend regularly. Thus, our hypothesis is not accepted.

\
## Customer Count by Category:

`SELECT product_category_id, COUNT(user_id) AS category_total_customer FROM ALI_USER WHERE type_of_behavior = "buy" GROUP BY product_category_id ORDER BY category_total_customer DESC`

\
 ![](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-xye3ohu.png)

product_category_id = 4801426 has the most purchases while 181182, 194104, 245030, and 344221 have the least purchases.

## Customers’ Buying Power:

`SELECT user_id, CASE WHEN customer_count = 1 OR customer_count = 0 THEN "Bronze" WHEN customer_count = 2 THEN "Silver" WHEN customer_count = 3 THEN "Gold" ELSE "Platinum" END AS customer_buying_power FROM( SELECT user_id, COUNT(user_id) AS customer_count FROM ALI_USER WHERE type_of_behavior = "buy" GROUP BY user_id ORDER BY customer_count DESC) t`

\
 ![](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-6td3oe1.png)

\
Notes:

· Bronze = customers who haven’t made any purchases or purchased once in the platform

· Silver = customers who purchased twice in the platform

· Gold = customers who purchase three times in the platform

· Platinum = customers who purchased 4 times or more in the platform

\
According to the list of classified customers generated from the query and the pie chart above, most of the users had made purchases at least twice between November 25th, 2017 to December 3rd, 2017 from Alibaba. In this case, further recommendations are encouraging the customers who are on the “Bronze” list and maintaining the rest of the customers by keeping up their purchasing desire.

## Repurchase Rate:

`WITH cte AS ( SELECT user_id FROM ALI_USER WHERE type_of_behavior = "buy")`

`SELECT COUNT(t.user_id) / COUNT(cte.user_id) * 100 AS repurchase_rate FROM( SELECT user_id FROM ALI_USER WHERE type_of_behavior = "buy" GROUP BY user_id HAVING COUNT(user_id) >= 2) t RIGHT JOIN cte USING(user_id)`

 ![](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-j2c3o4p.png)

\
The repurchase rate (customers who purchased at least twice/customer who purchased) is 4.2989%.

## Customer Conversion Rate by Category:

`SELECT product_category_id, ctvisit, ctbuy,concat(round(ctbuy100/ctvisit,3),'%') AS conversion_rate, product_number FROM (SELECT product_category_id,count()as ctvisit, SUM(CASE WHEN type_of_behavior='buy' then 1 else 0 end) as ctbuy, COUNT(DISTINCT product_id) as product_number FROM ALI_USER GROUP BY product_category_id) t ORDER BY conversion_rate`

\
 ![](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-zjb3okw.png)

There are a couple of product categories that have high conversion rates which show in the graph above. In order to investigate the reason behind the fact, we drilled down to the detail by adding the number of products in each category in our graph. In the graph, you can see that the higher the customer conversion rate, the lower number of products in the category. This is because when customers have too many options or substitutions, they might need to spend more time comparing products, which leads to a lower customer conversion rate. On the other hand, for the product category that has fewer options or substitutions, customers tend to spend less time choosing the product which leads to a higher customer conversion rate. To sum up, the product categories that have the most purchases are ranked by the combination of customer conversion rates and page visits.

\
Reference: <https://beeketing.com/blog/connection-customer-choice-conversion/>

\
## Step 5: Conclusion and Recommendations


1. Most of the customers are active on the Alibaba planform around 8 pm to 10 pm, we recommend the Alibaba marketing department enhance the marketing campaign from 8 pm to 10 pm by focusing on suggested product push.
2. Since the number of customers who were active during the weekend is very similar to other dates regularly, so we don’t suggest the platform needs to spend extra effort on the marketing campaign for the weekend. However, the marketing department should be aware of the coming visiting peak for Alibaba sitewide discount “12/12”.
3. There are a few product categories that have low purchase rates like 181182, 194104, 245030, and 344221. We suggest the marketing department increase product exposure for these categories to their target customers and be prepared for Alibaba sitewide discount “12/12”.
4. By analyzing the customers’ buying power, the majority of customers are in the “Bronze” and “Silver” levels. In this case, we recommend the marketing department analyze customers’ shopping carts and enhance the recommended product algorithm to improve the purchase rate of customers.
5. Businesses that fall in the low customer conversion rate product categories are very competitive. After digging into the “eCommerce Conversion Funnel”, we discovered that customers prefer to choose among the top brands when they have too many choices. Such a competitive market is very unfriendly for small businesses in the same product category. In this case, we encourage small businesses to break through from their original product category and cover the product categories that have high customer conversion rates by using some of their product.

   \
    ![https://www.convertcart.com/blog/ecommerce-conversion-funnel](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-e9a3o23.png)

# Step 6: Data Visualization

Tableau Dashboard: <https://public.tableau.com/app/profile/jiayi.pan4997/viz/alibaba_user_analysis/Dashboard1?publish=yes>

 ![Tableau Dashboard Preview](https://cdn.hackernoon.com/images/tOTkIHjitVMoN6FWiwd4YTerx6i1-rn93ogn.png)

\


This story contains new, firsthand information uncovered by the writer.

Comprehensive Data Analysis with SQL and Data Visualization: Alibaba User’s Behavior Investigation

Too Long; Didn't Read

@kirstenpan

Credibility

RELATED STORIES