Customer Personalization
Problem :
The same recommendation of products for every visitor.
Need to cross-sell products during checkout.
Low average browsing time.
Data Science Challenges :
Converting transactional data into one row per customer dataset. 70% of the work.
Extreme messy data, including multiple formats in one variable type, extreme values, missing values, strings in numeric variables etc.
Approach/Solution :
Built a recommender system based on visitor's history, interests of users like him, and people's pattern in the same geography.
Suggest products that the user will pair up with current products at the time of checkout. Also called cross-selling. This is based on the history of users like him/her. For example, if a shopper buys jeans, we can be suggesting him belt and shoes to match them.
Automated personalized email every time user abandons cart to incentivize the user to come back.
Results :
Browsing time has gone up from an average of 2 minutes to 3.5 minutes.
The percentage of abandoned carts went down by 12%.
Higher customer satisfaction in surveys by 2 points (on a scale of 10).
130 more products sold every month due to precise suggestions at checkout (cross-selling).
