Measure the Profits of Online Sales with Data, Not GMV

Artificial Intelligence and machine learning helps building an enhanced customer experience. To do this successfully, huge amount of data is required from the users and that’s where bigger sales are generated fetching as much data in a single day.

"The big thing about a sale is that it helps you generate a lot of data," says Ajit Narayanan, Myntra's CTO. "You hear a lot about the sales amount, and the revenue, but the consumer data has so much real value, to build a better experience for our users," reports Gadgets 360, an NDTV venture.

Diwali was the biggest sale of the year for the e-commerce companies, which entailed immense advance planning to meet the rush of the people. Ultimately, this also serves as a proving ground for the companies’ technology and other innovative infrastructure.  ‘End of season' is extremely important especially when two major fashion e-commerce companies are now under a single umbrella, after Flipkart acquired Myntra in 2014, and then Myntra acquired Jabong this year.

According to the further reports of Gadgets 360, Myntra's CTO, Ajit Narayanan, talked about how the company is preparing for the next big sale, and the role artificial intelligence and machine learning plays in this. Myntra cleared that apart from being an e-commerce company, it basically sees itself as a fashion brand.

Like other fashion e-commerce giants Myntra must focus on its technology aspects so that it can proffer proper and enhanced service to the customers in spite of sudden spike in the traffic. "At the highest point, the number of orders per minute goes up by 350 times. This is not sustained through the entire sale of course, but even so, you see something like a 25x increase in traffic," added Narayanan.

It is basically a vast surge in evaluation, for example, the 10x surge Snapdeal witnessed on its current sale. Obviously, the comparative dimension of the base audience is a factor as well, but from a technology perspective, it is the unexpected escalation that's frequently more challenging than the entire volume.

"We've done this a few times now, and we realized that planning for this had to become a part of our process, it wasn't something that we could just start doing a couple of months in advance," says Narayanan. "The way we develop and test has changed. Today, we simulate this load using user agents, that are almost like bots. We take the user data we have, and use algorithms to work out how the expected volume of users will behave."

"So you can tell how many people will just browse around, how many people will add things to the cart but then stop, how many will go to the checkout," he adds, "you can chart out the funnel. Then we use the user agents to test out our compute power, our networking, and storage, among other thing, reports the NDTV.

This testing is done on Myntra's live servers even as customers are shopping, because according to Narayanan, "simulation will only take you so far." Some of this can cause servers to fail, but at this point, there's not enough actual traffic that it can cause a major issue, and much of the testing occurs late at night so as to be less of a disruption to day to day business.

Myntra is able to pinpoint weaknesses that need to be augmented, and areas where it needs to scale up - though Narayanan says that it's not about finding a brute force solution either. "Because afterwards, we would have to scale back," he says, adding that at the infrastructure level, Myntra will sometimes 'borrow' compute power from sister concern Flipkart, reports Gadgets 360.