Project Background and Objective

#Project-Background-and-Objective

I was asked to find out insights from data to cast light on increasing revenue for next quarter by 10%. So in this project I took every effort to explore interesting things hidden behind the data and inform decision making to increase revenue.

Data Loading, Interpreting and Preprocessing

#Data-Loading,-Interpreting-and-Preprocessing
Loading output library...
Loading output library...

Missing Values Check

#Missing-Values-Check
Loading output library...
Loading output library...
Loading output library...
Loading output library...

Found max value of "spend_usd_next_14_days" is 22519.57 -> irregular

Ourliers sometimes misguide the analysis and therefore lead to incorrect results. To avoid this I will use a method named winsorization to replace the outliers.

Tackle outliers

#Tackle-outliers
Loading output library...
Loading output library...

Exploratory Data Analysis for Business Insights

#Exploratory-Data-Analysis-for-Business-Insights

Insight 1: Different form of payment make difference in spending?

#Insight-1:-Different-form-of-payment-make-difference-in-spending?
Loading output library...
Loading output library...

From above we can see users who added credit_card have much higher(5x) conversion rate than those with paypal. Then we can say that incentivizing users to add credit cards would make them purchase more.

Insight 1-b: Adding gift cards to acounts make a surge in average spending for dcb users

#Insight-1-b:-Adding-gift-cards-to-acounts-make-a-surge-in-average-spending-for-dcb-users

First, Let's see whether there is a difference in spendings among users with different payment types.

Loading output library...
Loading output library...

We can see from the above chart, users will spend more than 2x on purchase when they add a gift card to the account. Then we can say that incentivizing users to add gift cards would make them purchase more. Then I wouder whether I should target all users or just subsets of the users. In order to answer this question, I conduct further data analysis.

Loading output library...

Among all users who added gift cards to their accounts, 32.6% are only adding a gifts card to their accounts. 23.5% are adding credit cards and gift cards simultaneously, 22% are adding dcb and gift cards simultaneously. From the above findings, a conclusion comes to my mind that customers tend to use the combination of gift cards and credit cards or the combination of gift cards and dcbs to pay their orders. Then I consider that whether I could increase the spendings by incentivizing more credit card users and dcb users to add a gift card to the account.

Loading output library...

From the above two pie charts, I know there is very few proportion of the credit card users and dcb users who have added a gift card to the accout. Then I need to check the impact of adding gift cards would have on the average spendings in credit card and dcb users.

Loading output library...

The graph told us that there was a siginificant increase in average spendings in dcb users when adding gift cards to the accounts.

Recommendation 1

#Recommendation-1

Given this I get my first insight that we can increse revenue for the next quarter by targeting the dcb users and incentivizing them to add gift cards to accounts. In order to incentivize them, we will show a mesaage under the check out button saying "Good News! You can also check out this item through the gift card!" and then offer a link to gift card purchase page.

Insight 2: A large gap between adding a cart and clicking checkout

#Insight-2:-A-large-gap-between-adding-a-cart-and-clicking-checkout
Loading output library...
Loading output library...

Recommendation 2

#Recommendation-2

The funnel shows there is a steep decrease in number when users who added a cart click checkout buttons. If we can take meaaures to lower the number of decrease we will make a increase in revenues.

Recommendation: we should sent direct eamils to cusotmers who have been dormant for 3+ days after adding items to cart to remind them to check out.

Insight 3: Spending for the next two weeks arise critically comparing between purchase weeks equal to 7 and purchase weeks equal to 8

#Insight-3:-Spending-for-the-next-two-weeks-arise-critically-comparing-between-purchase-weeks-equal-to-7-and-purchase-weeks-equal-to-8
Loading output library...

We can see from the above graph, total spending for the next 14 days arise significantly if it is an existing customers.

Then I am thinking which groups of the existing customers we should target? the Whole or some segments

Loading output library...
Loading output library...

Recommendation 3

#Recommendation-3

If buyers make a purchase every week in the past 8 weeks , he would have a steep increase in spending for the next 14 days, compared to those whose number of weeks making a purchase is 7.

Recommendation: I will target the US buyers who have 7 weeks making a purchase out of every 8 weeks and send coupons to incentivize them to make more purchases.

Experiment Design

#Experiment-Design

Marketing Campaign

#Marketing-Campaign

Now I select one of the three insights to carry out an experiment in order to examine its impact on the total revenue increase.

Strategy: Conduct a markeing campaign targeting the US existing customers who made purchases 7 weeks out of every 8 weeks. In the marketing campaign, we send coupons by direct emails to incentivize them to purchase more.
Senario: David was considering to buy a new earphone but he thought the curent one was ok, so buying a new one was not urgent. However when he gets the coupon he thinks it is a good idea to but it now.

A/B Test Design

#A/B-Test-Design

Business Goal: increase total revenue for Q3 by 5%

Metric: Total Spent USD
Population: all exiting costumers who have 7 weeks of purchase out of every 8 weeks.
Sample Size: In order to get 80% of statistical power, 5% of significance level and 5% of sentivity, a sample size of 6282 is needed to be reached.
Experiment Launches: Randomly select two groups (treatment vs. control) with each group size 6282.
Treament Group: cusomters in the treatment group will see coupons sent by direct emails: Every 50 dollars you spent in the next 14 days, you got 10 dollars off your purchase.
Testing Approach: Two Sample t-test
Duration: 1 Week (computed according to daily traffic and sample size required)

Sanity Check

#Sanity-Check

Done with data gathering, I need to do sanity check first on the invariant metrics.

Invariant metrics: total number of records for each group

Loading output library...

The invariant metric for these two groups are comparable, which validates the data collected.

Result Analysis

#Result-Analysis

Experiment Result Visualization

#Experiment-Result-Visualization
Loading output library...

It seems that there is a 5% uplift of total revenue in experiment group. Then the task goes to test if this uplife is statistically significant.

In order to compute a 95% confidence interval, a method named Jackknife is applied here. The reason why I use Jackknife is that it works well when we don't know the distribution of population or we can't make any assumption for the distribution of population.

Creating 95% confidence interval using Jackknife

#Creating-95%-confidence-interval-using-Jackknife
Loading output library...

In order to figure out the reason behind the insignificance, I will selice data further to pictures which segments cause the this result. For example, I could slice data according to 'vertical' or "country", then create the 95% confience interval to see whether the result is significant for each groups. Suppose that they were neither significant however I slice the data.

Recommendation

#Recommendation

There is no statistcal significance in uplift of total revenue of experiment group. This means there is no impact made on the total revenue by the marketing campaign if we run the same campaign multiple times. In other word, this strategy just works less than 95 times out of every 100 times of campaign made.

As a result, we can't reach the goal to increase total revenue by 5% for the next quarter. In order to reach the goal, we have to switch to another strategy and then conduct a A/B test to see the significance.

Plan B: Incentivize cusomters to add credit card, gift cards to their accounts.

Plan C: Send direct emails to remind customers to check out after they have been dormant for 3+ days after adding items to cart