- Data science
- Deep learning
- Machine learning
- Business intelligence
Pitch
We pitched whether they face any issues in terms of growth, sales, marketing etc. We presented some services from us, including AI/ML solutions such as Recommender system, Sales forecast, Demand prediction, Churn Analysis, etc. They got back to us as they realized they face high customer attrition.
Our success rate to retain customers with our solution is above 85%
Proposal
We proposed the following services initially:
- A thorough analysis of their past orders and transactional data and make data story.
- Churn and customer retention modelling predict customers with a high chance of going away in a defined timed window.
- Pricing and discount strategy for customers with a high probability of going away.
- Personalized email and SMS marketing.
Personlized email palyed vital a role to customer retention success.
Data Collection
At this stage, we requested their data team coordinate with us to provide information about what kind of data they have about their customers, customersโ behaviour and their transactions. We got to know they collect a fair amount of data, and they are the following:
Feature name | Type | Example |
Gender | Categorical | Male/Female/Unknown |
Age | Numeric | Calculated from birthdate |
Is senior citizen | Categorical | Derived from age |
Location Lattitude/Longitude | Numeric | Derived from address or location |
Delivery distance | Numeric | Calculated from location data |
Delivery time | Numeric | Obtained from transaction data |
Has Subscription | Categorical | Yes/No |
Term | Categorical | Monthly/Annually |
Total subscription time | Numeric | Total days that customer has the subscription |
Gaps in subscription | Numeric | Total days that customer was out of subscription |
Payment method | Categorical | Card/Internet banking/POD |
Avg purchase value daily | Numeric | |
Avg purchase value weekly | Numeric | |
Avg purchase value monthly | Numeric | |
Total purchase value | Numeric | |
Avg time spent on the web app daily | Numeric | |
Avg time spent on the web app weekly | Numeric | |
Avg time spent on the web app monthly | Numeric | |
Avg time spent on the mobile app | Numeric | |
Avg daily time spent on each category | Numeric | One hot encoded for all categories |
Avg daily time spent on each product | Numeric | One hot encoded for all products |
Numer of cart abandoned | Numeric | |
Is Coupon used | Categorical | Yes/No |
Avg Discount given per order | Numeric | |
Churned | Categorical | Yes/No. Labelled. |
Explanatory data analysis and Feature Engineering
After collecting these data, we cleaned up the data by removing blank data or imputing missing data. Then we made a data warehouse effort to create a data frame. Thenย we performed univariate, bi-variate, and multi-variate analyses. We are putting some brief points here.
- There is a 7.5% churned rate.
- Customer who is younger than 34 tends to churn more.
- Gender distribution does not contribute anything.
- Annual subscription tends to stay more.
- More significant gaps in subscription tend to churn more.
- Discount played a dominant player to retain a customer and many more.
The churn rate was 7.5%.
Modelling
We first split the data in train and test split with 80% and 20% ratios. We trained several models from classical machine learning (statistical models) and deep learning such as Logistic regression, Support vector machine, Random forest, ANN, and CNN. We built a classifier to predict customers with a high probability of going away.
Our model test accuracy was 95%.
Result
We measure the outcome of this model by checking how many customers marked to be churned in the next month got retained. After some special treatment early, like providing significant discounts and other perks, more than 95% of those marked customers were retained. Overall the churn rate reduced to just 1.8%.
Overall the churn rate was reduced to 1.8% from 7.5%.