Flipkart's data plan

Utkarsh Bhriguvanshi is a sort of a freewheeling one-man army at Flipkart. Formally, he is a principal architect and an ‘IC’ or an individual contributor who loves to code, offers his expertise where needed and is generally among the top tinkerers at the Bengaluru-headquartered company, India’s biggest ecommerce business.

“I came here more than six years ago; people must think I’m a dinosaur,” he jokes. The self-deprecatory humour hides a razor-sharp mind that is among those constantly tapped by the company’s seniormost people, including Chief Executive Officer (CEO) Kalyan Krishnamurthy, 45, and Chief Technology Officer (CTO) Ravi Garikipati, 52.

Bhriguvanshi, 37, says Flipkart is in its third phase of technological advancement: From solving problems like ensuring a trustworthy sale of a book or a mobile accessory in the early years, to figuring out how to support scale—meaning, a very large number of transactions—“by 2014, when we had become a monster [company]”, to now doing things intelligently.

“That means personalised experience. That is the holy grail,” said Garikipati in a recent interview to Forbes India. “The app experience that you see should be different from what I see.”

For instance, at its five-day Big 10 sale that concluded on May 18—part of celebrating the company’s 10th anniversary—Flipkart was able to target specific products at women in metros, while men were shown something else on their app home page. “Gender affinity” is what they call it.

In another pilot during the sale, it experimented with providing real-time inventory information to consumers, in the midst of a buying frenzy. Leading up to and during the Big 10 sale, Flipkart also ran a host of social media promotions and contests: Algorithms handled 20 lakh entrants in a bid-to-win contest. In another promotion, natural language processing software moderated 2 lakh reviews—such as flagging reviews that might be deliberate attempts to add negative comments about a product—on one of the sale days.

Flipkart was able to do all this because of its work over the past few years at collecting data from the innumerable transactions on its website and app, and building the big-data analytics infrastructure needed to process that information. Today, it claims to have one of the most extensive data across urban and middle-class India, and is now gearing up to go after the rest of the country with a host of other technologies.

In April, Flipkart scored $1.4 billion in funding from Microsoft Corp, eBay Inc, and Tencent Holding Ltd, and agreed to purchase eBay’s Indian operations in the bargain. All of this has brought the company some serious strategic heft—not just money, but partnerships with global companies with deep tech knowhow: From Microsoft’s depth in artificial intelligence (AI), to eBay’s experience in online commerce, to Tencent’s ability to exploit mobile commerce platforms that handle hundreds of millions of users.

Number crunchers: (From left) Utkarsh B, principal architect, Ravi Garikipati, Ram Papatla, Anand Lakshminarayanan and Mayur Datar
Image: Nishant Ratnakar for Forbes India

At stake is nothing short of Flipkart’s position as the top ecommerce business in India, where Amazon.com Inc is widely seen as a close second, but also one that is making gains. For instance, Amazon surpassed Flipkart in the number of app downloads in the first three months of 2016, according to multiple media reports in India, which cited data from App Annie, a market research company that tracks apps and their publishers worldwide. Flipkart still led the field in India on monthly active users, the reports said.

Further, while Flipkart started 10 years ago, Amazon entered India only four years ago and is widely seen to have overtaken local competitors such as Snapdeal, run by Delhi’s Jasper Infotech Private Limited, to emerge as a strong ecommerce business in India. Amazon has entered the groceries segment ahead of Flipkart, and is said to be in early talks to acquire local online groceries startup BigBasket, the largest such business in the country. Amazon’s Prime Video service also helps it engage customers beyond ecommerce.

Founder Jeff Bezos, over the past three years, has committed to spending at least $5 billion in expanding Amazon in India, where the US giant has had a development centre for nearly 15 years, and has opened a global data centre facility last year in Mumbai for its Amazon Web Services.

And although big data analytics is par for the course among ecommerce companies, Flipkart is adding more muscle to its game. On an average day, Flipkart’s 100 million plus users generate as much as 10-15 TB of data, says Anand Lakshminarayanan, 41, a vice president and head of the company’s Big Data Services group. During a sale, that volume can increase to 50-60 TB. That helps the company figure out users’ “affinity” for price and brands, among other habits.

With all the data it has collected over the years, the company can now say that when it shows a woman’s handbag to a user, the probability is very high that the user is a woman in a particular metro, with a propensity to buy such products.

Collecting and mining data is an investment Flipkart has been making from fairly early on, recalls Bhriguvanshi, putting in place several “big data platforms” in 2011-12. All of those have evolved and coalesced into the Flipkart Big Data Platform—a growing, changing combination of hardware and software—of today, that Lakshminarayanan’s group runs, and which is affectionately called Bigfoot within the company.

Bigfoot processes all the data it is fed, runs algorithms that are constantly being built and tweaked, and brings up information like a customer’s affinity for a particular brand, times and days of purchases, and browsing behaviour.

Flipkart is using this knowledge to achieve a “long-term engagement with the user and his or her family,” says Lakshminarayanan. It is harnessing the power of machine learning (ML) and deep neural networks, and getting closer than ever to you and me and our families, friends, neighbours, colleagues and the outwardly extending network of all our touchpoints. Each of us will eventually be an open book.

The latest funding has also brought Flipkart serious strategic heft “Can we predict users’ shopping behaviour based on their ‘life stage’?” “The key is to fail fast, learn from the experience, and iterate quickly.”

The flip side (pun intended) will be that we will no longer get search results or recommendations or even home pages or notifications that are apparently irrelevant to us. And AI will be at the heart of how we’re not just served, but anticipated.

Flipkart already knows a lot about its users—based on a combination of their login ID, IP address and device, and even their search queries—and is in a position to recommend not just products, but specific features, based on increasingly accurate user profiles. “Ultimately what we’re trying to get to is, by looking at your past behaviour, we’re looking at building a predictive model. Predict a probability, saying ‘for this query, this product, this user, how likely is a given outcome’,” says Mayur Datar, chief data scientist at Flipkart. Datar, 40, spent a dozen years at Google including two leading its research group in India, before moving to Flipkart two years ago, where he works with 20 to 25 data scientists.

The scale of what Flipkart was trying to do was what attracted Datar. “You have an opportunity to change ecommerce in India,” he says. Data science requires data, and for Datar, Flipkart’s petabytes of data is a bottomless, shape-shifting cookie jar throwing up never-before flavours.

As his models of customer behaviour become more accurate, Flipkart can begin to exploit them to do a multitude of things—from maximising the profit on the next sale of a high-value item for a customer, to making multiple offers at very attractive rates that complement an existing sale. This kind of predictive analysis is increasingly common in industries such as oil and gas exploration and commodity market movements, and “we are on the cusp of it”, says Datar.
“At some point can we start predicting the users’ shopping behaviour based on the ‘life stage’ they are in? That’s the pivotal point,” says Ram Papatla, 43, vice president for product management, and an important recipient of all the information that Lakshminarayanan’s group brings up.

Life at Flipkart for Papatla is a never-ending pursuit of that next bit of end-user insight. He came to Flipkart from Amazon, and has also worked at Microsoft for 15 years. The answer to his pursuit of user-insight lies in the field of deep learning.

This is a fairly new technology, says Datar. It is a class of algorithms that can be applied to multiple areas. Training machines to automatically figure out the meaning of text—“red means colour, Samsung means brand, Galaxy is a make”, and so on.

One of the more well-known applications of deep learning, particularly relevant to ecommerce because of its cataloguing needs, is image-recognition; to teach software to search for similarities in images and group them, so that they can be recommended to users. This also benefits suppliers; if one supplier has put up a red T-shirt with a graphic logo, the software will comprehend these features for other sellers too.

Understanding the user is also going to increasingly include the use of video and, in the foreseeable future, voice. “We are going to even out the field there. That includes Indic language awareness,” Papatla says. So expect talking to your Flipkart app sooner, rather than later.

“There’s been a lot of academic work in speech technology in India, but not so much in actually building systems,” says Kalika Bali, a researcher at Microsoft Research, working on ML, natural language systems and applications and technology for emerging markets. That could change now as the interest in speech-enabled applications is really growing, she says.

At Flipkart’s F7 Labs, its research arm in Silicon Valley that comprises just about 20 people—data scientists, product engineers and analysts—it’s happening. They are now going beyond image recognition into speech recognition. Jatin Chhugani, head of F7 Labs, and Mihir Naware, director for products, two of the seniormost people at F7, expect voice recognition to become fairly mainstream within the next two years as a means of interacting with computers. They are, however, careful to not give away any Flipkart-specific timelines.

Flipkart is also fostering many academia-industry partnerships in its fields of interest, from IIT Bombay and Stanford University to Carnegie Mellon University, says Muthusamy Chelliah, 52, who joined Flipkart two years ago as director of academic engagement. Today, senior scientists as well as research students from some of the top universities in the world are identifying research problems around the data that Flipkart has amassed.

“World over, companies are in the early stages of applying AI and ML in ecommerce. Many players are investing in those technologies to improve business performance and customer experience,” says Sandy Shen, a research director at market research and business tech consultancy Gartner. For Flipkart, just as for its competitors and others, “this will be a continuous journey as players learn about AI and ML technologies, and how to best use the data they own to generate the best results for various use cases,” Shen says. “There will be many failures, but the key is to fail fast, learn from the experience, and iterate quickly.”

And Flipkart’s new strategic investors can help. “As a strategic investor, Tencent can share what it has already learned from its social media and entertainment businesses. Likewise, Flipkart can also share what it learns from its Indian business,” she adds.

Flipkart has come a long way from when founder Sachin Bansal was building the website himself and fellow-founder Binny Bansal focussed on the catalogue of products. The company has assembled a crack team—people like Garikipati, Papatla and Lakshminarayanan have all come from the best-known tech businesses, including Google, Microsoft and Amazon. And Flipkart’s team of roughly 1,100 engineers has the depth it needs at the next levels as well.

Can they successfully build the technologies that will help them eliminate the basic friction points—the need to first open at least the mobile site and enter a search query, and choose a product—before a user even makes a purchase? Or, in groceries, for instance, “how can we do something that is even easier than a customer calling a store and asking for a delivery?” muses Papatla.

How will they get the non-English-speaking India to identify with Flipkart? Will that also be through speech recognition? The data they hold is showing the way, and they have begun to walk and jog down the path. The time to sprint will soon be upon them.