Big Billion Grin
How Flipkart, under CEO Kalyan Krishnamurthy,
sailed through its Big Billion Days 2017 sale and
looks stronger on technology, selection and affordability
Late one evening in January 2017, an email from Binny Bansal hit the inbox of Flipsters, as Flipkart employees call themselves. Another e-mail was sent to journalists across the country at the same time. It was front page material. The kind of story that makes beat reporters drop whatever they are doing and take to Twitter to inform the world.
Flipkart had started sending emails to journalists and employees together to prevent leaks. The thinking was simple: if the media got the wind of it, it would create chaos inside the company. If the employees knew it first, it would anyway find its way into the press.
In the e-mail, Binny announced that Flipkart will now be run by Kalyan Krishnamurthy, who was earlier a managing director at Flipkart’s largest investor Tiger Global. Binny himself would become the Group CEO of Flipkart. His job would be to oversee capital allocation and also making Flipkart ready for a public offering, the email said.
It was a crucial moment in the company’s 10-year history: the two founders – Binny Bansal and Sachin Bansal – had ceded control of the day to day operations of Flipkart to a “new chief executive officer”. Not many had thought this day would come. Perhaps not even Sachin and Binny under whose stewardship, Flipkart had grown from a tiny outfit run out of a two-bedroom apartment in Koramangala to the country’s largest e-commerce player with over 100 million registered users, 80 million products and 100,000 registered sellers on its platform.
It was the third top-level change at Flipkart since 2014. The top deck was shuffled in 2014, 2016 and now in 2017. It was only last year in January that Binny had taken over from his other co-founder Sachin as the CEO of the company. At the time Mukesh Bansal was heading the commerce platform and the company's advertising business. Soon, Mukesh quit the company. A series of other exits had rocked the boat. Employee morale was down and fresh capital wasn’t easy to come by. Flipkart had set upon a plan to turn things around and Kalyan Krishnamurthy was to play an important role here.
Cut to present: Flipkart’s first Big Billion Day with Krishnamurthy as the chief executive officer is underway. We meet him for the first time on Friday, around 10 pm. It’s been just over a year since his comeback at Flipkart in July 2016.
Krishnamurthy, 45, doesn’t like talking to the press so this meeting is a rare event. He comes from Tiger Global, a hedge fund founded by billionaire investor Chase Coleman and run by a handful of people including Lee Fixel, a long time India bull and Flipkart’s most fervent backer. Tiger is awfully secretive even by hedge fund standards.
But then, Big Billion Days are big and he’s been giving out a few interviews talking up the company’s focus on customers, how they made products affordable, and why this time the sale is bigger and better than those of the last few years. The five-day festive season sale that ran Wednesday through Sunday, started off midnight Tuesday. The sale is crucial for Flipkart, which is battling Amazon on one side and Alibaba backed Paytm on the other. Even more so, for Krishnamurthy.
“This is probably the first time both these players have a head-on competition that ultimately is a win-win for the consumers,” says Amit Bhawani, the founder of gadget blog PhoneRadar, referring to the rivalry between Flipkart and Amazon India.
Ask any Flipkart employee who has stuck around for long and he or she will wince at the 2014 sale. “It was super intense. Everybody was shouting on the floor. We were firefighting. Heads rolled,” one employee described the events in office on October 6, 2014.
Flipkart had failed to anticipate the huge spike in demand during the sale and its systems didn’t hold up. Sachin Bansal wrote an email to customers in which he said: “We failed to live up to this promise yesterday and would like to apologise once again to every single customer for our failure.” After the sale, airports were choked with undelivered merchandise as there weren’t enough trucks to ship goods on time. Overall, Flipkart shoppers made for unhappy customers.
The stakes are higher now and the margin for error lower. Flipkart has raised over $7 billion to fund its growth — with the latest $2.5 billion round led by SoftBank, the largest round of funding by an Indian venture-funded company. That’s a lot of risk capital riding on one company. Earlier this year, it brought in more funds from the likes of Microsoft and Tencent. At a valuation of over $11 billion, Flipkart is now in the too big to fail territory.
At the time when fresh funds were raised, there was speculation that Krishnamurthy might move on from Flipkart as founders Sachin and Binny took a more active role at the company. However, at least five current and former Flipkart executives told FactorDaily that much of the new funds were riding on the fact that Krishnamurthy is at the helm of the company.
There was a sense of calm about Krishnamurthy when we met him on Friday. “The first BBD was a big experiment. And clearly like any other experiment, some things go right and some go wrong. Since then, everything has matured,” he says. He is hoping that the company will have a crushing 70% market-share during the festive month.
"The first BBD, the second BBD to some extent and maybe even the 2016 BBD, was heavily about offering pure value. But this time we have gone way beyond that and said that we'll offer Indian customers selection that they have not seen before and at very good value," Krishnamurthy told FactorDaily.
It’s Day 3 of the 2017 sale and all systems at Flipkart’s Big Billion Day command centre on the 3rd floor in the 11-storey office, were glowing green. “We spent a disproportionate amount of time and investments thinking about that almost 7-8 months prior to the event itself,” he says.
Krishnamurthy was part of the team that originally conceptualised the Big Billion Day in 2014. After his return to Flipkart as the head of category design in June 2016, the first thing he did, was to set sights on the Big Billion Days sale of October 2016. The festive sale was a sort of test for him. Krishnamurthy went around the company and recruited a team of operators across functions for the sale day. Going by the numbers last year, he (and Binny, who he was running the sale together with) passed the test with flying colours.
Big Billion Days have been historically big for Flipkart. In 2014, Flipkart saw more than a billion hits across its properties. Goods worth over $100 million were sold on that day. The Big Billion Days sale in 2016 was a whole new level altogether. About 2.5 million smartphones were sold. On a single day, October 3, the second day of the 2016 sale, the gross value of products sold was over Rs 1,400 crore — more than two times the value of goods sold in the 2015 sale.
This time around, Flipkart focused much of its energy on tying up exclusive deals making products more affordable for customers and spent heavily on digital marketing as opposed to other forms of advertising. "We started by saying we should look at selection for Indian customers which is not available easily and drawn from customer insights. The other part we really went after was affordability," said Smrithi Ravichandran, Senior Director - Flipkart.
Customers have noticed this push. “This year was huge for Flipkart. Not only did they manage to a large number of good deals, they also went on the offensive when it came to payments,” said Harpreet Singh, who runs DealsforGeeks, a website that spots deals on e-commerce sites.
To be sure, Flipkart's selection has some catching up to do with Amazon India, widely seen to have a better selection of products. Amazon India says it has over 100 million products listed and is adding at the rate of 180,000 a day — nearly a million every week.
Big Billion days are a healthy addition to Flipkart's top line. Flipkart grew its revenues by 50 % in FY16 to Rs 15,403 crores. However, the company is still far from making profits, as it battles rivals. Flipkart lost Rs 5,768.8 crore in FY16, up 86% from the previous year.
Flipkart said on Sunday evening that it has doubled its sales compared to last year with 70% market-share. The growth came at roughly the same amount of money spent on marketing as it did last year, it said. (We couldn't independently verify the marketshare numbers). Reports suggest online retailers put together sold goods worth $1.2 billion to $1.7 billion during the sale, nearly 50% more than last year. Across its properties, Flipkart estimates that it will get four billion to six billion eyeballs this month.
Doubling sales and cornering a majority of the market sounded ambitious. But as it turns out, they were prepared.
The weekend before the Big Billion Days 2017, we meet the Flipkart technology team. On Sunday, we spent several hours late into the night interviewing the people heading the tech and product teams. For almost a month now, most of Flipkart’s tech team has been working the graveyard shift to stress test its systems for scale. “For us, it's like a war without an enemy,” says Ravi Garikipati, Flipkart’s Chief Technology Officer.
Every year, Flipkart’s ambitions grow bigger and bigger and there’s much ground for the tech team to cover. In the last couple of years, it has improved its search experience, made it easier for users to pay, created a content platform that can serve up fresh personalised content for users without manual intervention, saved cost by deploying machine learning algorithms to support customers, made its systems more reliable by building its own private cloud, and created playbooks to follow in the event of a meltdown.
“The strategy is laid out months in advance and to be able to work with various leaders across functions and bringing all of that together into some sort of actionable plan. It's soon becoming a major event in the digital world, almost like Fifa of the digital world,” says Garikipati.
The private cloud
Right after the Big Billion Day in 2014, Flipkart engineers set out on a big project to make sure that it has its own data centres that can handle the load. “Until then we were a smaller set-up,” recalls Vinay Y S, VP-Engineering at Flipkart.
Today Flipkart claims that it has about a million compute cores in its private cloud, making it one of the largest such deployment in the country. Flipkart’s data centres, split between Mumbai and Chennai handle all the traffic these days. “Our footprint is about 10 MW capacity in one location,” says Vinay. The second data centre is powered for 4 MW. Data centres are heavy consumers of electricity thanks to computer processors and air conditioning and often are rated by the electric load they are provisioned for.
Flipkart built a data centre, a cloud platform on top of it and migrated all operations to the new setup in 2015-16. “The 2016 BBD was really flawless. We had done multiple spike events before that,” says Vinay.
Sherlock and the search for efficiency
Improving Flipkart’s search has been big on the company’s agenda from the start. Mayur Datar, the chief data scientist at Flipkart, spends much of his time thinking about this. In the last year or so, he’s cut down the amount of time taken to search Flipkart’s catalogue to nearly a third. It takes about 500-700ms for Flipkart to surface results now as compared to 2 seconds in 2015 for 99% of its queries.
To deal with scale and improve efficiency, his team has changed the architecture of its search engine in the last two years. "I need to look more and more like the Googles of the world out there," says Datar, an ex-Googler who joined Flipkart in 2015. The search revamp project was codenamed Sherlock and Datar’s team sits out of a bay called Baker Street — the home of fictional detective Sherlock Holmes.
Datar and his team made searching a more “distributed affair”. Earlier, one machine used to hold all products and listings. When a query came in, it went to this machine and got the results. To scale, many replicas of this machine were built to parallelly process search queries. With the new architecture, even for a single query, many machines that hold parts of the search result are hit. This is a faster and fault-tolerant way of doing things. Engineers call it “sharding”, a process by which large databases are partitioned into smaller shards that are easier to manage.
The scale at which Flipkart operates now, every bit of processing power or storage that's saved can make a difference. "Search is one of the biggest hoggers of CPUs," says Datar, who manages a team of nearly 70 Flipkart employees – mostly data scientists, product managers and engineers. Every percentage point efficiency he squeezes out of the process has an impact on Flipkart's margins. "There's a lot of onus on my team to make sure we use fewer machine resources," says Datar, who reckons that Flipkart has nearly double the number of features as compared to last year same time.
Flipkart’s search engine is built on top of Solr, an open source enterprise search platform.
For broader queries, like "sarees" or "shoes," the search engine also needs to personalise the result for a user. Searching through one giant index wasn't going to work. Sharding along with the ability to score results has helped Flipkart deliver better search results, says Datar. This isn’t an easy thing to do though. Scores can be different for each user as things like geography, availability of goods and many such aspects come into play. Some of these factors are dynamic (for instance, an item could go out of stock) and scores need to adjust themselves.
Make them click
With festive marketing kicking in, Flipkart expects to get between four billion and six billion eyeballs in the month of September. That makes Flipkart's homepage premium real estate. A merchandising council which meets every week and lays out the business objectives that week.
Then there's a bit of competition among categories and other functions to figure out who gets to be on the homepage. These meetings can get intense at times. But the loudest voice doesn't always win.
Prashant Kumar and his team run a system which decides what content gets on the homepage. How is that done?
There are two cuts to it, Kumar explains. First, you divide your users into many different buckets based on past behaviour. Then comes the second cut. Based on a customer's current behaviour on Flipkart, a machine learning model predicts their propensity to buy certain items on the site. "When a new click comes, the model has to be recomputed to come up with the change," says Kumar. Flipkart's content management system, codenamed Neo, serves up content that has a higher probability of attracting a buyer in real time. "Now each time a user clicks, we are able to serve him with better content," he says. Updating content on the site used to be a manual process. That's not the case anymore.
For the last 4 months or so, Flipkart has also been building out its homepage 24 hours in advance every day and having it tested internally before it's shown to users. "Given that we are now fairly sophisticated in terms of how all this content comes together, we can get into cohorts of users and get very granular," says Ram Papatla, Vice President, Product Management at Flipkart.
Fixing the app, drip by drip
FactorDaily readers will be familiar with Amar Nagaram, from our coverage of Big Billion Days in 2016. He handles customer experience for Flipkart. This time around, Nagaram wants to make sure that Flipkart is 99.9% crash free to beat the industry best of 99.7%. The last two releases have been running 99.8% crash free, he says. He's been dropping updates for this year's Big Billion Day into Flipkart's apps right from April. "Our apps actually started getting chunks of goodness since April," he says. The idea is to make sure that all releases are tested to work fine even during the Big Billion Day.
Flipkart has a new religion: AI
Flipkart also wants to make a bigger bet on artificial intelligence and machine learning. The next 36 months, the company wants to become an AI first company. Which means, going at all the problems that Flipkart is trying to solve with an AI-lens. “That includes search, discovery, product assortment, listing quality and other operational elements. That means quite a transformation,” says Garikipati. “We're going to be spending a good chunk of time building AI and ML capabilities. In some cases such as image processing, text processing, speech recognition, we can use deep learning very well,” he adds.
Internally, Sachin Bansal has been pushing a lot for AI, a top source aware of the company’s thinking told FactorDaily. The belief here is that the next wave of internet users will come from middle India and they may not be comfortable using text. Most companies targeting this market use images and voice to create interfaces and that’s where AI comes in handy.
Flipkart is pretty well established in online retailing. But is it the right approach for Flipkart to become an AI-first company? Is it missing out on building products such as Prime, Alexa and Kindle that Amazon uses to drive repeated use of their platform which in turn trains its artificial intelligence algorithm?
These questions still linger. But what is clear though, is that the battle for supremacy in Indian retail as it ramps up to a trillion dollars by 2020 is far from over.
Lead Visual: Rajesh Subramanian
Photographs: Rajesh Subramanian, Shamsheer Yousaf
Infographics: Nikhil Raj