Designed by rawpixel.com / Freepik
Until the project get finished, this README will contain the problem and the steps to be done (So, this won't be the final ReadMe).
If something goes wrong when opening the notebook, please try this link: https://nbviewer.jupyter.org/github/KattsonBastos/health_insurance_cross_sell/blob/main/notebooks/health_insurance_cross_sell.ipynb
In summary:
- Whom: customers interested in the new product;
- how they got the customer's interest on purchasing a new product: a survey of about 380,000 customers;
- How many new customers will participate in the offering campaign: 127,000.
- How will they offer the product to customers: phone calls (capacity for 20,000 phone calls during the campaign);
THE TASK
Our Data Science team was hired to build a model that predicts whether or not the customer would be interested in auto insurance, so the Sales Team hopes to be able to prioritize the people with the greatest interest in the new product and thus optimize the campaign by making only contacts with customers most likely to purchase the new insurance.
Questions we have to answer:
- Main insights on the most relevant attributes of customers interested in purchasing auto insurance.
- What percentage of customers interested in purchasing vehicle insurance will the sales team be able to reach by making 20,000 calls?
- And if the sales team's capacity increases to 40,000 calls, what percentage of customers interested in purchasing vehicle insurance will the sales team be able to contact?
- How many calls does the sales team need to make to contact 80% of customers interested in purchasing vehicle insurance?
Understanding the demand
- The root cause of the demand: to offer the vehicle insurance so that the sales team makes more calls to customers who are interested in purchase hte product.
- Stakeholders: the company's Sales team
- Product Delivering Method:
- Granularity: unique customers data
- Business Model: Cross Sell - classification prolem
- Main Methods: Classification Machine Learning Models
- Solution presentation: Dashboard with the classifications and insights; and provide the model so the Sales team could make requisitions and get the classification of new clients.
- Sprint 01 (05/01)
- Business Demand Understanding
- Initial Hypothesis Creation
- Solution Planning
- Data Collection
- Sprint 02 (12/01)
- Descriptive Analysis
- Business Research and Cross-Sell Understanding
- Hypothesis Creation
- Feature Engineering
- Sprint 03 (19/01)
- Exploratory Data Analysis
- Sprint 04 (26/01)
- Data Preparation
- Feature Selection
- Sprint 05 (02/02)
- Machine Learning Model
- Sprint 06 ( 09/02)
- Business Metrics Creation
- Translating and Interpreting the Metrics
- Sprint 07 (16/02)
- Model Deployment
- Sprint 08 (23/02)
- Accessing Data in Production
- Sprint 09 (02/03)
- Presentation to the Business Team
- Sprint 10 (09/03)
- article creation
- What we learn
The insurance sector is all about taking and managing risks. Thus, offering a new product (vehicle insurance) means taking more risk. offer a new product (vehicle insurance) means taking more risk)
Insurance is a contract, represented by a policy, in which an individual or entity receives financial protection or reimbursement against losses from an insurance company reference. The basic concept of insurance is that one party, the insurer, will guarantee payment for an uncertain future event, and the insured or the policyholder, pays a premium to the insurer in exchange for that protection on that uncertain future occurrence.
The premium is basically the price of the insurance, tipically expressed as a monthly or annually cost. It is based on the customer's risk profile.
According to this article, the premium can vary depending on many factors that are believed to affect the expected cost of future claims. Some of those factors are: gender, age, driving history, marital status, profession, vehicle classification, neighbourhood, behavior-base and even the credit rating.
Insurance companies base their business models around assuming and diversifying risk. They generate revenue in two ways: Charging premiums in exchange for insurance coverage, then reinvesting those premiums into other interest-generating assets. The insurer's real product is the customer claims. Thus, the company must process when the customers files them and then check their accuracy and submit the payments.
The main reason why insurance works is because the likelihood of something unfortunate happen to the insured is low (reference). However, it's not the same between all insurance types.
health insurance
It is an insurance that covers the whole or a part of the risk of a person incurring medical expenses, spreading the risk over numerous persons.
Differently than others, health insurance is used more often.
vehicle insurance
It provides financial protection against physical damage or bodily injury resulting from traffic collisions and against liability that could also arise from incidents in a vehicle. It covers many types of vehicles, like cars, trucks, and motorcycles. It can also cover against non-traffic events, like theft, natural disasters, and weather.
Cross-Sell is to sell related or complementary products to existing customers, maintaining a healthy relationship with them. There are some reasons why cross-sell products is important. This post list some of them:
- Revenue generation
- Customer loyalty
- Improved Customer satisfaction
- Increase customer retention
- Increased Customer base
That post also points that selling to existing customers costs less than selling to new ones. Besides that, it's important to note that cross-selling still is selling, which means the company needs tp help the customer solve his/her proglems. Thinking on insurance matters, to offer two insurance types is good to the customer because he/she could prefer a single place to get them, simplifying the task of searching for different products.
Machine Learning is usefull in many ways to the insurance sector. Among that, it helps to find patterns in the data and then to determine the probability of cross-selling a product to the existing customer and also to predict the right time to pitch the products to existing customers based on inter-purchase time of similar customers
Some hypothesis about the business problem was created. They will guide the Exploratory Data Analysis, offering insights and also the relevance of each feature to the model. That relevance will be compared with the results of some feature selector algorithms and then we can better decide what feature to consider in modeling.
| Hypothesis | Description |
|---|---|
| H1 | Among the customers interested in purchasing the insurance, there are more around the middle age. |
| H2 | The proportion of customers interested in purchasing the insurance is higher among those who have driving license. |
| H3 | At least 40% of customers who have had their vehicles damaged are interested in purchasing the insurance. |
| H4 | Among women, the proportion of customers interested in purchasing the insurance is at least 10% higher than the proportion among men. |
| H5 | Among customers over the age of 60, those who have a vehicle older than 2 years are more interested in purchase the insurance. |
| H6 | The proportion of customers interested in purchase the insurance is at least 10% higher among those who are customers for more than 6 months. |
| H7 | Customers who spent above the average in premiums are more interested in purchasing the insurance. |
| H8 | Customers who spent above the average in less time as a customer are more interested in purchasing the vehicle insurance. |
| H9 | There are more customers interested in purchasing the insurance among those who have a vehicle newer than 1 year. |
| H10 | There are more customers interested in purchasing the insurance among those who have never had a vehicle insurance. |
| H11 | Customers from regions that have a higher mean in premium spending are more interested in purchasing the insurance |
User Level Data
| Variable | Type | Description |
|---|---|---|
| Id: | Discrete | customer's unique ID. |
| Gender: | Discrete | customer's gender. |
| Age: | Discrete | customer's age. |
| Policy sales channel: | Discrete | anonymous code for the customer contact channel. |
| Region Code: | Discrete | customer's region code. |
Vehicle Level Data
| Variable | Type | Description |
|---|---|---|
| Id: | Discrete | customer's unique ID. |
| Driving License: | Binary | 0, the customer is not allowed to drive; and 1, the customer is allowed to drive. |
| Vehicle Age: | Discrete | vehicle age |
| Vehicle Damage: | Binary | 0, the customer has never had his vehicle damaged in the past; and 1, the customer has had their vehicle damaged in the past. |
Insurance Level Data
| Variable | Type | Description |
|---|---|---|
| Id: | Discrete | customer's unique ID. |
| Anual Premium: | Continuous | amount the customer paid the company for annual health insurance. |
| Vintage: | Discrete | number of days the customer joined the company through the purchase of health insurance. |
| Previously Insured: | Binary | 0, the customer does not have auto insurance; and 1, the customer has had auto insurance. |
| Response: | Binary | 0, the customer is not interested; and 1, the customer is interested. |
