Harnessing Analytics and Data Science for Ecommerce: Insights from Ravi Vijayaraghavan of Flipkart

In the world of ecommerce, the use of data and analytics to enhance customer experience is becoming increasingly crucial. This was the central theme of a recent talk given by Dr. Ravi Vijayaraghavan, Senior Vice President and Chief Data Analytics Officer at Flipkart, at MachineCon India 2023.

The Power of Scale and Machine Intelligence

Vijayaraghavan began his talk by emphasizing the importance of scale in ecommerce. With around 500 million registered users, 150 million products, and billions of monthly visits, Flipkart operates at a massive scale. This necessitates the use of machine intelligence, as manual operations simply cannot keep up.

The Three Pillars of Flipkart’s Data Strategy

Vijayaraghavan outlined three key areas where Flipkart leverages data:

1. Measurement Science: This involves using statistics and machine learning to create a hierarchical structure for targets, connecting strategy to execution through data.
2. Data-Assisted Human Decisions: This includes classic business analytics, such as deciding what categories to invest in, how to drive growth, and operational and planning initiatives.
3. Data-Assisted Machine Decisions: This involves using AI for tasks like discovery, search, recommendations, personalization, listing quality, selection, and trust and safety.

The Importance of Assortment Design

One of the key areas where Flipkart uses data and analytics is in assortment design, which is crucial in ecommerce. Assortment design involves determining the width, depth, and quality of the assortment needed. Width refers to the number of product lines, depth refers to the variety within each product line, and quality refers to bringing in the right designs.

Leveraging Machine Learning for Assortment Design

To determine the appropriate assortment width, Flipkart uses natural language processing to categorize and cluster search queries. Poor performing clusters, which have a low click-through rate, indicate areas where more selection is needed. Flipkart then uses a seller affinity model to identify sellers who could potentially fill that selection.

To determine assortment depth, Flipkart uses machine learning to predict demand. Once demand is predicted, the impact of selection depth on demand can be assessed. This helps Flipkart determine how many styles of a given SKU are needed to meet consumer demand.

Using Generative AI for Assortment Quality

The most innovative aspect of Flipkart’s approach to assortment design is the use of generative AI for determining assortment quality. Flipkart uses a Stable Diffusion model to design new fashion items, particularly in the unbranded women’s fashion area.

First, Flipkart identifies selection gaps in its platform, i.e., attribute combinations that don’t currently exist. These gaps are then fed into the Stable Diffusion model, which generates new designs. These designs are then evaluated using a high potential score, which predicts their potential to sell at a certain price point. Designs with a high potential score are sent to sellers to be produced and added to the platform.

Conclusion

In conclusion, Vijayaraghavan’s talk provided valuable insights into how Flipkart is leveraging data, analytics, and AI to enhance customer experience and drive business growth. By focusing on assortment design, Flipkart is ensuring it offers a wide, varied, and high-quality selection of products to its customers. As we move into the future, this approach will be crucial in harnessing the full potential of ecommerce.