The ecommerce company Flipkart is a household name in India today which has made deeper inroads over the last few years across the country. The company, which claims to have come up with new payment model called Cash on Delivery (CoD), is now building artificial intelligence (AI) based products to fix challenges for its customers and sellers.
2015 was when the company set up a lab in the Silicon Valley with the aim to employ latest technologies for its datasets and gain insights on shopping behaviour of customers. F7 Labs over has been working to help Indian ecommerce giant improve its business processes and enhance shopping experience. The director of the lab, Mihir Naware, believes that “online commerce will soon become as easy and familiar as that of brick and mortar experience.’’
The 10-member team of data scientists and data engineers of the lab is betting on data science and machine learning technologies to rethink ground up the challenge areas as their priorities.
The journey so far
The team say they have amassed a lot of fascinating insights on their customer behaviour --from the nature of their ‘search queries to how they evaluate products before clicking the buy button. ‘’These insights have helped our product teams to optimize the experience for customers and sellers,’’ says Jatin Chhugani, Principal Architect, F7 Labs.
Their AI-based model called, DeepInsight for instance, can extract 50 different features of an image while automating image processing for the entire catalogue of Flipkart. The insights are then conveyed back to their sellers to work on them.
‘’One problem we encountered is the difference in language usage of our customers. Compared to customers in the US or UK, the vocabulary used by our audience to express their sentiment is quite different,’’ says Mihir. Given the current size of its audience, the company can’t rely on manual moderation of reviews. They have developed a new NLP-powered algorithm which can generalize the search terms and phrases, including abbreviations. “We are now able to handle 100s of thousands of reviews with 98%+ accuracy,’’ says Mihir.
The image optimization technique they call LightSpeed is another attempt by the lab to reduce the amount of data sent over the wire, which they say, is impacting 100M Indian customers. “For a typical browsing session, a user would save 500Kb - 1Mb data because of this improvement. In other words, customers can save 35% - 40% of image data,’’ claims Mihir.
For the F7 Labs, there are a few things on the anvil. The team is working on a one-step process of information discovery on its platform. The goal of the initiative, called Mira, is to enable customers to express their desire in the most natural way possible. Users of Flipkart may soon get to see this in action, as ‘’this is closer to becoming a reality,’’ says Jatin.
The company is working to come up with chat-bots, as Jatin hints at, to automate its customer interactions. The Lab plans to roll this feature over the next few months.