Vishal Pandey

Data Scientist with experience in customer segmentation, data analysis, and machine learning. Skilled in Python programming, data cleaning, and data visualization. Able to provide insights and recommendations to stakeholders based on data analysis.

Churn Prediction:

• Extracted data from 15 interfaces saved in multiple tables in different schemas
• Prepared dataset of all customers who have churned as well as renewed in last 1 year
• Analyzed patterns of almost 15k customers' information saved in more than 100 columns extracted from multiple interfaces
• Built more than 10 models after preparing the dataset (data mining, cleansing, pre-processing) and compared results on a score board also tuned a few models using hyper-parameter tuning
• Finally selected LGBM with accuracy 83%
• Each passing month, the model's accuracy increased to 90% with the help of retraining using latest patterns in the latest month's data
• Using our recommendations and churn predictions done every month, the organization has achieved decreasing the churn:renew ratio from 60:40 to 50:50 in 3rd quarter

Lead Prediction:

• As an e-commerce company, we have more than a crore leads saved in our database but the sales team has no idea which are potential leads and to be approached first
• Based on customer’s on-boarding journey, this project is divided into two parts:

a) Registered user to interested customer (Agreed to know about product & pricing)

b) Interested customer to paid customer

• Prepared a detailed report on customer’s on-boarding journey based on 6 months sales team interactions with non-paid customers
• Performed exploratory data analysis on more than 7 Lac registered users with whom sales team have interacted in previous three months and set the target as a meeting done if a customer becomes interested in the current month
• Explored all interfaces related to a newly registered customer, be it company profile, demography, product categories in which he deals, products posted on our website, recency & frequency on our website, etc., and found patterns of conversion
• Performed sentiment analysis based on conversation with sales team
• With the help of Oops concepts like classes, functions, constructor, global and local variables, created seamless flow of code which can be used by multiple projects and inter-usable within the project
• As the target had imbalanced classes 93:7, so we evaluated our models using Recall and Precision
• After comparing all models, we finalized the Catboost classifier model which has 30% recall and 30% precision after adjusting the threshold
• This model has improved the company's sales efficiency by 20%

Upsell Prediction:

• Selling an additional product or selling the validity of an already sold product to an existing paid customer is upsell
• To analyze patterns of upsell in existing customers, we’ve collected data from customers who existed in different months in the previous 6 months
• Average Upsell percent in the training dataset is 2.44%
• Recommendations based on the top ten features of our final Catboost classifier model promise to increase the upsell percent to almost double

Customer segmentation:

• Used RFM (Recency, Frequency, Monetary value) model and TIM (Traffic, Inquiry, Meeting) to understand different kinds of customers and their behaviors
• Segregated customers into 4 groups and provided personalized recommendations based on their behaviors
• Identified potential up-sell/upgrade opportunities for the company's business
• Conducted state-wise top business categories analysis by dividing categories into 10 buckets in each state based on active buyers & sellers, sales meeting percent and conversion percent
• Sorted all non-paid customers based on priority to help sales teams get the best leads first
• Analyzed sales employee performance by running a K-mean clustering model on features extracted from different interfaces, including employees' sales data, no of months of experience, etc
• Found five clusters with employees having different characteristics.