With the amount of data they handle, online retailers are studying if they can offer better services with analytics
New Delhi: Flipkart.com may soon be able to read your mind—well, almost. India’s largest online marketplace is working on a pilot project to see how analytics can be used to sense the anxiety level of a user.
“When a customer calls us, how do I predict on a scale of one to 10 how anxious the customer is?" asked Flipkart’s chief technology officer Amod Malviya. “If we can determine the anxiety level, we can automatically route them to our select set of people who have expertise in dealing with these situations," said Malviya, who is using machine learning to do the trick. He insists that this is just the tip of the iceberg.
Machine learning is a scientific discipline that deals with the construction and study of algorithms that can learn from data.
Snapdeal.com, the second largest online supermarket, predicts market trends based on user behaviour, click data and information from social media that it collects. It also uses an algorithm to rank sellers, said Amitabh Misra, senior director (technology). The company uses algorithms to see if loads need to be shifted from any one courier firm to another, and that too, in real time.
Of course, Amazon.com Inc., the world’s largest e-commerce company, besides using drones for delivery in the US, uses high-end analytics and algorithms in a number of areas such as recommending relevant products to users, showing users relevant search results, displaying ads to users that they are very likely to click on, predicting future demand for products, detecting spam reviews and detecting fraudulent orders, a company spokeswoman said.
The access and delivery bloomers that e-commerce companies such as Flipkart, Amazon and Snapdeal made during their major sales drives in the past couple of months in India may make it hard to believe in their tech prowess.
Nevertheless, the fact remains that these companies are pulling out all the stops to make analytics work for them. Especially, given that these companies churn out humungous amounts of data on a daily basis.
Flipkart segregates data into three segments, said Malviya. First is consumer behaviour data—what kind of things customers like or dislike, why do their preferences exist, how do consumers think about purchasing a product, what is important for a customer, value quotient or price quotient or lifestyle quotient and for which class of products.
The second segment deals with product behaviour data—why does a particular product sell higher than the others; which two products, in consumers’ minds, are substitutable; and which two products are complementary.
The third is the supply chain side of data points, gathered from vendors.
Flipkart also has a data scientist group focused around machine learning-oriented problems. It is also working on automatically detecting fraudulent reviews on products, aimed at artificially inflating the rating of a product.
The company set up a team of seven machine-learning experts in 2013. Some of these experts are also engaged with research teams in various academic institutions including the Indian Institute of Science, Bangalore; Indian Institute of Technology, Kharagpur; and Carnegie Mellon University, Pittsburgh, Pennsylvania, in the US.
“Early this year, we agreed on sharing data with them and identifying the problem statement that we will be working on with them," Malviya said.
Snapdeal, which drives over 40% of its order volume from recommendations and personalization, says data use has three elements. “One is how it can help buyers to buy. Second, how it can help sellers sell more effectively. And finally, how do we use data analytics to run the operations internally," said Misra.
To do all the number crunching and make a sense out of the mountains of data, Snapdeal uses a multi-tier system, which it calls a “Hadoop-based farm". “In this multi-tier structure, we have a dataware house, columnar database (a management system that stores data in columns instead of rows), Hadoop and a business intelligence layer," explained Misra.
Hadoop is an open-source data processing tool. The system can be used through dashboards, tables and application programming interfaces to determine anything.
Amazon, according to its spokeswoman, also leverages data about products that are purchased together to show users “people who bought product X also bought product Y".
“In product search, we try and show users the most relevant search results and in online advertising we show users ads they are most likely to click, based on past click history," she said.
All e-commerce firms are sharpening their technology focus on the fashion category.
Flipkart is keen on using machine learning in the fashion-shopping space. The company is exploring the image analysis side of things in order to make fashion shopping easier, said Malviya, without elaborating. Image analysis is a process of extracting meaningful information from digital images.
Snapdeal, too, is working on a project to make fashion shopping more personalized, based on a buyer’s persona. “When it comes to fashion, people broadly know what they are looking for, but we still need to show the choices which interest them. We want to broadly categorize our entire user base by persona, that is, on the basis of behaviour traits mixed with demographic traits," said Misra.
He added that the company is trying to integrate fashion trends across different geographies and “bringing those feeds into our data, and personalizing it on that basis. For example, south India may never need woollen clothes—we are building a lot of new stuff, especially on fashion—to match the taste", said Misra.
Amazon.com is planning to make an aggressive push into fashion and lifestyle products to take on Flipkart-Myntra, which dominate online sales of apparel, footwear and accessories, Mint reported on 5 November.
“The big retailers like Marks and Spencer or Walmart have been using it for quite a bit now to understand customer behaviour and product positioning. For example, chocolate and chewing gums are put generally on counters, since most of the customers buy them impulsively, while milk is available quite inside the store. You don’t buy milk impulsively, you go to buy milk specifically, and while you walk to that particular location, you are enticed to buy other products," said Sudarshan.
He, however, added that what is different between brick-and-mortar retailers and e-commerce companies is the huge amount of data that is collated at one place, ready for slice and dice.
“It is a first-hand data available on a tap, which can be used in a far more meaningful way. It is far easier to predict consumer behaviour for online retailers," he said.
According to Sudarshan, analytics is being used at three levels: “First, understanding what a consumer wants. For instance, if a person uses BookMyShow app to book a movie ticket every weekend and his first choice is a Telugu movie. If that is unavailable, then he chooses a Hindi movie. Now BookMyShow can easily predict what this particular customer wants. So, it can send a personalized advertisement on, say, Thursday, giving some offer or discount.
“Now if a user books a Telugu movie ticket every weekend and then goes to eat at different restaurants, which he explores through Zomato, then there is a chance to offer him packages, that is, a Telugu movie with his choice of restaurant. This is the second level, predictive analytics, to understand what a customer likes.
“The third level, which has not progressed yet, is giving a customer what he can afford or is likely to afford by analysing his buying behaviour, his quickness or flexibility in decision-making while going for an expensive purchase," Sudarshan explained.
There are challenges, though.
One of the big challenges while deploying analytics remains how to ensure systems are powerful enough to take decision on their own, instead of people taking decisions.
“That is the one class of systematic intelligence," Malviya said. “The second aspect is, deep personalization—how you understand customers much more deeply, in order to create much more significant value proposition for the customers."
For instance, at present, when you create a return for a particular product, there is a certain process that you have to go through in order to understand reasons behind it and it takes certain amount of time, according to Malviya.
His company is working on a pilot project that uses a wide variety of data points to create a model that determines whether it can predict the reason behind a call, and automatically initiate a response, eliminating the need and time taken for a conversation.
Flipkart is also working on another pilot that involves machine learning to understand the context behind a customer’s search query. “A phrase like cheap phones can have different meanings for different customers," explained Malviya.
Big Data, meanwhile, requires huge storage houses or data centres.
Flipkart, which raised $1 billion (around ₹ 6,180 crore today) in July, already has two data centres in Mumbai and Chennai, and expects to increase its network footprint across India.
“Across various cities in India, we will work with local Internet service providers to create network POP (point of presence). It will allow us to deliver our content to customers little more effectively," Malviya said.
Snapdeal, which also recently raised $627 million, plans to increase its investment in Big Data platforms and high-end analytics tenfold.
The company, at present, has 50 specialized developers including data engineers, architects, data miners, data scientists, and plans to open up an innovation centre in Bengaluru, taking the numbers of engineers to 1,000 from the current 350.
Snapdeal also plans to launch an advertisement platform that will be powered by analytics and machine-learning algorithms by early next quarter. The platform will help sellers disburse their marketing budget more effectively as well as project their return on investment, said Misra.
To be sure, Sudarshan of Deloitte has a word of caution. “While technology will be a major factor in determining the state of things for online retailers, it cannot surpass judicious decision-making," he said.