Ravi Garikipati, the Chief Technology Officer of Flipkart talks to FactorDaily on the company’s engineering priorities, AI-first culture and Big Billion Day. Edited Excerpts.
On this year’s BBD
We have overarching strategic goals for the business that gets translated to a bunch of innovations on the business, product and tech front. Then we have a bunch of streams spawning from there. For some, it's all about bringing new products and features, for some, it's looking at our ability to scale. The fact that we have our own private cloud platform helps us significantly. Some of the big guys, like Alibaba, have gone on the same path as well. We work closely with various vendors, build custom hardware components and most systems are built in-house so we can control every line of code. We also look at performance engineering and we've come a long way in the last few years. The last piece we focus on is to have playbooks around unforeseen circumstances. The focus is largely about making sure that at any given point, our channels are up and running.
On the role of data
We also focus quite a bit on things like helping our business on real-time decision-making. We have close to 30 PetaBytes of data that we process every day and about 1100 counters that help business guys take decisions. Suppose a particular category has done well in terms of target and you need to shape the demand for some other category, that's not something we wait for hours-- that's real time. That translates to different merchandising and it is also personalised.
On user data
We have close to about 1,000 data points about any user. We use machine learning models to dole out insights about the user. We predict the user’s age, store affinity, brand affinity and price affinity and personalise the experience. So if you are a runner looking for shoes, you see a different product assortment. In some cases, where we also see certain user behaviour that might translate to some sort of fraudulent outcome, we are able to predict that real-time and take away certain options like Cash on Delivery.
The 1,000 data points are raw data and then you can use machine learning models to derive insights and put that for personalisation. Those data points are not of much use in the raw form. With user insights, we have a tiered customer experience. The premium customers will probably have a human touch, for others it could be virtual agents and chatbots and so on.
Can you handpick a bunch of innovations from a technology point of view?
The traction is largely between the app and the mobile site. This time around, the focus is to reach out to middle India where we typically run into bandwidth constraints. It's not very common for anyone to download an app or a flaky network so the fact that we invested quite a bit on a progressive web app held us in good stead. It's reflected in how the mix is changing. After app, it's largely m site and not as much on the desktop.
We've introduced a lot of payment modes that align with a lot of our affordability constructs. We also have new payment instruments like PhonePe.They all help us bring down the friction.
We built a sophisticated content syndication platform. The platform takes charge of organic and inorganic content addressing various partners including advertisers, category leaders and brands. It used to be a human-driven thing in the past but now it's all machine driven and highly personalised. The idea here is to be able to identify what content is more amenable for these users across buckets and auto-populate it. In the past, it used to be someone sitting there and pinning that content. But now it's all automated. We do it for better interaction, engagement with the user that eventually leads to better CTR. A few percentage points difference could make a huge difference to the GMV.
Priorities for engineering?
In the 3 year horizon, we are deeply focused on AI-first culture. Just like app first, that's going to be an integral part of what we do. Almost every problem that we solve, should be seen through an AI first lens. That includes search, discovery, product assortment, listing quality and other operational elements. That means quite a transformation.
We are also going to be looking at our computing environment to address AI/ML computing needs. We'll be working with the best of the partners in the ecosystem to be able to create our own computing infrastructure.
We're also 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.
The next wave of internet explosion will happen in the middle India and they may not be comfortable with conventional modalities such as text. It will be a combination of imagery, visual interfaces combined with speech. There's going to be a fair bit of investment there.
We're also going to be reaching out to academia.
Supply chain would be another interesting area. We want to redefine entire transportation industry as it relates to consumer internet opportunities.
Where do you get the conviction from that you can innovate in AI? Because talent is an issue, most PhDs work for large companies like Google and Facebook.
It's imperative. It's very existential. It's very realistic as well. But it requires tremendous conviction that things ought to be done with AI mindset. Firstly, you need to have a rich corpus of data and I think we have that. On a BBD like event, we collect few tens of petabytes of data. Once you have the data, most of the platforms that enable data goodness is out there as well. They demystify some of the complexity and help engineers bring data to play. Most of the algorithms are out there. I don't think you need 10 CMU professors or Stanford professors to come with new algos. But that said, I'm maintaining a strong channel with research and academia. We have our own data scientists. In some cases, we need to leverage that. But most problems we look at is about getting our act right. Today, even the most advanced companies struggle with data quality and completeness. We're investing in all these areas. Access to data is key here. That's one of the reasons professors at top universities drool at companies like Flipkart. We are not inventing new algorithms, applicability is important here.