Rahul Anand

Developed a machine learning model which resulted in a widget that presented pertinent information about a customer's buying and viewing habits. For example, this customer bought this item and also bought these items, or customers who viewed this product also viewed these things. This generated an eight percent lift in engagement and a four percent increase in total purchases.

I developed a mathematical formula that gives weight to the marketing channels for the purchases made by users. It was based on the probability distribution of the landing page and channels. This helped the finance team create a budget for all the paid marketing channel, and helped the team create more efficient campaigns.

I developed a framework on Spark, that helped identify the correct attribution of the order of the marketing channels in real-time. This helped us adjust the budget of inefficient campaigns to save money. Having this system work in real-time, saved millions of dollars a month by reducing the impression share of inefficient campaigns, and replacing them with the impression share of efficient campaigns.

Built a deep neural network model to predict the propensity of purchases for each user towards each product available on the site. This was used in email marketing and push notifications. The model added an eight percent incremental gross profit.

Developed an automated system to predict how much to place on bids for each campaign on Facebook and Criteo, and bid for keywords for SEM. The system also helps select the top keywords to bid for SEM.

Created an end-to-end framework to test two variants of the product, and find the winner based on statistical techniques. It started with basic sample size calculations, Next, we ran a smoke test to identify any fundamental flaws in the system, after which we ran a calibration phase to get the impact and an even more accurate sample size. Finally, we ran the main test to get the overall lift and t-value to calculate the statistical significance. I applied group sequential analysis to end the experiment in less than the prescribed time.

Convinced top management to look at the customers based on long-term impact rather than just RFM segments. This shaped the company to look at a broader segment of users rather than simply focus on short-term gain. I built a machine learning model to predict the one-year downstream value generated by all customers, and segmented them based on the prediction. This model is being used to target customers appropriately via paid marketing channels, spend more on low-value customers to encourage them to move up the ladder, and shape their spending habits in a way that is most profitable for the business.