Increasing the sign-up rate of a credit card offer by analyzing customer demographics

Click for recording of final presentation

Click to view presentation slides

Please note: Project based on real bank data, but bank name has been changed

---

Background

Hawkins Bank designed a focused marketing study, where they sent 18,000 current bank clients a credit card offer in the mail. This approach allows the bank to know who does and does not respond to the offer, and to analyze this data with existing demographic data of each customer.

Objective:

6%, or about 1,000 customers signed up for the credit card after receiving the offer by mail. How can we increase this number next time?

I will build a classification model to predict which bank customers will sign up for a credit card offer sent by mail. Then I will use this model along with my analysis to suggest actionable insights to Hawkins Bank to improve the sign-up rate of their next credit card mailer campaign.

---

Quick view of actionable insights:

1. Send the offer 100% by postcard, NOT enveloped letter.

2. Send a higher proportion of customers the Air Miles Rewards program rather than the Points or Cash Back

3. Clients with medium-to-low incomes and/or medium-to-low credit scores are the ones most likely to sign up for a credit card offer sent by mail. Focus the next mailing campaign on them. Note** I personally would not feel comfortable sharing these kind of demographic findings about customers. I acknowledge this more in my presentation.

Main Insights:

Type of mailer

50% of customers were sent the offer by postcard and 50% by mailed envelope. Customers responded better to the credit card offer sent by postcard rather than by mailed envelope. 70% of customers who took the credit card offer had been sent the offer by postcard.

Action: Save money and paper by sending 100% of the offers by postcard next time.

Type of Rewards Program

The bank offers three Rewards programs to go along with the credit card- Cash back, Points, or Air Miles. The majority of customers who accepted the credit card offer were those sent the Air Miles Rewards program.

Action: Instead of offering each rewards program equally, offer a higher proportion of the Travel Rewards.

Demographic Features of Bank Customers

For this part of the analysis, we compare the demographics of customers who signed up for the credit card offer (about 1,000 customers) verses customers who didn't sign up for the credit card offer (about 17,000 customers)

Two features seemed particularly indicative of determining whether or not a customer would sign up for the credit card offer:

Income Level and Credit Rating

Action: Focus the next credit card mailing campaign on customers with medium/low income and low credit score.

---

Data: The data set consists of 14 demographic features on 18,000 current bank customers:

• Customer Number
• Offer Accepted: Did the customer accept (Yes) or reject (No) the offer.
• Reward: The type of reward program offered for the card.
• Mailer Type: Letter or postcard.
• Income Level: Binned into Low, Medium or High.
• Number of Bank Accounts: How many non-credit-card accounts are held by the customer.
• Overdraft Protection: Does the customer have overdraft protection on their checking account(s) (Yes or No).
• Credit Rating: Binned into Low, Medium or High.
• Number of Credit Cards Held: The number of credit cards held at the bank.
• Number of Homes Owned: The number of homes owned by the customer.
• Household Size
• Own Your Home: Does the customer own their home? (Yes or No).
• Average Balance: Average account balance (across all accounts over time). Q1, Q2, Q3 and Q4
• Balance: Average balance for each quarter in the last year

---

Process

1. Exploratory Data Analysis and Visualizations using Python
2. Cleaning data (dealing with NaN values, empty values, duplicate rows, standardizing heading names, checking outliers, etc.)
3. Querying in SQL
4. Pre-processing data for modeling (X/y split, train/test split, encoding categoricals and scaling numericals)
5. Defining and running the model (Logistic Regression) and choosing which metric is most important (precision, recall, or f1)
6. Treating data imbalance (upsampling, downsampling, SMOTE) to improve model scores
7. Re-running model with each data balance technique and comparing their confusion matrix and metrics
8. Drawing conclusions about data from the model
9. Hypothesis testing : Is the (average) average account balance of all Hawkins Bank customers the same as the (average) average account balance of customers who didn't take the credit card offer?
10. Data visualizations in Tableau comparing demographics of customers who did and didn't take the offer
11. Preparation of presentation to explain findings and actions to be taken by the bank to increase the percentage of customers who will take a credit card offer sent in the mail.
12. 7-minute live presentation to 12 colleagues to present project and data-driven insights followed by 10-minute discussion