Contact Press Team
The subtle, invisible AI that Indian unicorns have made a part of consumers’ lives
Published on June 27, 2017 • Quartz
Everyone’s talking about Artificial Intelligence (AI) and big data in India’s startup ecosystem.
Across sectors, startups are seeking out talent with AI expertise to analyse consumer data and offer personalised services to users. Meanwhile, global giants such as Apple have taken notice of Indian companies that help clients with data processing and image and voice recognition, and investors, too, are backing Indian AI startups.
Here’s how some of India’s unicorns—companies valued at more than $1 billion—are putting these technologies to use.
From being an e-wallet to selling movie or flight tickets, Paytm now delivers a diverse range of services and machine learning helps bring order to that chaos.
“You could Google and try to look for something. But a better world would be when Google could on its own figure out Charu is looking for ‘x’ at this time. That’s exactly what we’re doing at Paytm,” the company’s chief technology officer, Charumitra Pujari, who came to Paytm after working with the e-retail giant Amazon for nearly two years, said. “If you’ve come to buy a flight ticket, because I understand your purchase cycle, I show that instead of a movie ticket or transactions.”
Pujari said that “every pixel”—each icon, product row, column—on Paytm’s homepage is personalised and reordered differently for each of its 225 million users, and the platform makes 20,000 recommendations per second—each of them in under 20 milliseconds. Using machine learning, the platform also crafts promotions for serious buyers.
“If you wanted to buy a ceiling fan, which is not promoted on (the) home page, once you started looking, in real time, we can pick up your request and kind of put together a promotion for you,” said Pujari. However, these promotions are created only when Paytm detects an intent to buy. “If you’re just browsing, looking at ten different products, I won’t create the promotion.” That’s how they maintain a balance of increasing transactions without driving costs through the roof.
In order to detect and prevent fraud, machines are constantly vetting fraudulent accounts that solely signed up to take advantage of promo codes, or add money from stolen credit cards. The fraud-detection engine leaves little room for human error and speeds up the process, Pujari said. On the seller side, too, machine learning helps flag sellers who might start booking their own items, and take the cash back, to make their product demand look higher than it really is.
Backed by big names like Alibaba, its affiliate Ant Financial, and Japanese investing giant Softbank, Paytm is now taking the plunge into offering lending services and credit cards to customers, which will allow them to make transactions even if they don’t have funds readily available in their Paytm account.
“Not everyone in India has stable credit files, not everyone will be eligible for loans or have banks that are accessible,” said Pujari. “(A) lot of people have done transactions on Paytm, so they can get some sort of a (credit) score.”
To gauge how much a user can borrow, the company is using machine learning again, “capturing signals from the mobile app to find out who you are and what credit you should be eligible for,” Pujari said. The technology will draw up a variety of options it determines affordable, like an option to pay in 6 months with 0% interest or in 15 months with 5% interest.
The team behind these services comprises software engineers, machine learning engineers, and data scientists based in Toronto, Canada, as well as in Paytm’s headquarters in Noida, India, with roughly 60 people per location.
“We are always hiring, especially for these roles. We know the future is AI and we will need a lot more people,” said Pujari.
Online marketplace ShopClues wants to use machine learning to solve a pressing blind spot in e-retail: sizing.
In the case of small, unorganised brands, sizing varies and rarely matches the norms of bigger, established brands. For instance, a size small in one brand could actually be an extra small in another. So, ShopClues plans to use advanced technologies to make it easier for shoppers to find the right size when buying clothes online, according to Utkarsh Biradar, vice-president of product at the company.
“We will look at 360-degree views, we will look at trying to decipher sizes by using image technology,” Biradar said. The company plans to use standard-sized products as a reference—so the customer has a better idea of the size of the product they’re looking at—and plans to work towards standardising sizes across products and brands on its platform.
So far, ShopClues has been using such technology to personalise the shopping experience for customers. Based on information such as recent purchases, the frequency of purchases, and other spending habits, its algorithm tries to decipher what a customer may be interested in buying next. The company also uses machine learning to advise sellers on the most effective price for their products, for instance, or the most economical logistics partner to work with.
Mobile advertising platform InMobi is already using AI and machine-learning techniques for response prediction and targeting to find the right audience for specific ads. It’s also applying these technologies to help advertisers expand their reach effectively, using machine learning to identify “lookalike” targets that are similar to existing users as well as figuring out what kinds of ads users don’t want to see.
Next up for the company is going deeper into the image aspect of an ad.
“One of the biggest physical components of an ad interaction is the image itself: What the user sees determines largely what the user does with the ad—click on it or ignore it,” Rajiv Bhat, senior vice-president, data sciences and marketplace, told Quartz. “Historically it was difficult to capture that component because humans can see an image but algorithms could not, but now with things like deep learning, we can capture that aspect as well.”
Though many of these technologies have been in use at InMobi for a few years, it’s only now that the company is using programs such as MLlib, which help apply machine learning at scale. “While people are talking about a lot about these technologies it’s not that easy to implement it, and implement it at scale,” Bhat added.
Ola, one of India’s leading ride-hailing apps, is using data science and machine learning to track traffic, improve customer experience, understand driver habits and extend the life of a vehicle.
The company uses AI to understand day-in-day-out variations in demand on its platform, and to figure out how much supply is needed to cater to that demand, how traffic predictions vary, and even how external events such as rainfall affect the efficiency of vehicles.
“Ola Share is a classic example of leveraging big data to improve matching rates and minimise deviations,” co-founder and chief technology officer Ankit Bhati told Quartz. The algorithms behind the car-pooling service bring down travel times and distances by enabling riders with different pick-up points and destinations to share one vehicle, thereby reducing the overall fare paid by the customers.
Ola’s aim, Bhati said, is to “build more such solutions and add value (to) existing ones.”
The company is also deploying advanced technologies on the customer front with Ola Play, its personalised in-vehicle entertainment system, which allows users to select from a variety of entertainment options while travelling. Similar to Netflix and Spotify, Ola Play also uses machine learning to remember a passenger’s favourite music or movies, and play back content from their previous Ola rides the next time they travel with the company.
Machines are being used to manage drivers, too.
“AI is understanding what is the behavioural profile of a driver partner and, hence, in which way can we train him to be a better driver partner on (the) platform,” Bhati said, noting that the technology helps them identify how many customers are cancelling on certain routes and why, for instance.
Moreover, since Ola leases cars and partners with carmakers to offer discounted buying and maintenance plans to drivers, it has a vested interest in making sure the cars on its platform are operational for as long as possible. So, the ANI Technologies-owned taxi aggregator also creates machine learning models to determine the driving patterns that help maximise mileage and the life of a car.
E-commerce company Flipkart has already re-designed its app’s homescreen so that it’s personalised for each of its over 120 million shoppers, according to a report in Forbes. Machine learning models log each customer’s gender, brand affinity, store affinity, price preference, frequency, volume of purchases, and more, which become more accurate as the company collects more data. Then, they can make predictions even without the customer searching for specific products. Going forward, the company will also use machine learning to study when and why returns are made to figure out how to reduce this.
Flipkart wasn’t immediately available to comment on its use of AI.
In March, Flipkart-owned clothing e-tailer Myntra launched the country’s first fully-automated design collection, Moda Rapido.
“Earlier, the AI technology would figure out certain attributes like a placket with a contrast that is selling well, a Chinese collar that is very popular or a particular type of cuff design that works well; our team of designers would then take those attributes and design a shirt but now, we have graduated to zero human intervention,” Ambarish Kenghe, the head of product at Myntra, told the Hindu BusinessLine newspaper. By tapping into data from fashion websites, social media, and customer data, the AI designer creates “a TechPack” with design dimensions and specifications for manufacturing to produce clothing items.
A team of 25 data scientists at India’s largest domestic e-commerce player are also using AI to study past buyer behavior to predict their upcoming purchases.
“If a customer keys in a query for running shoes, we show only the category landing pages of the particular brand the customer wants to see, in the price point and styles that (are) preferred, as gauged by previous buying behaviour, therefore ensuring a faster, smoother checkout process,” Ram Papatla, the vice president of product management at Flipkart, told the Hindu BusinessLine.
As the company scales, it’s also adding chatbots to assist users by prompting them to refine their queries or helping them find an answer to a question faster, according to the Economic Times CIO. The bots use Natural Language Processing (NLP) technology that helps them improve with every interaction.
Flipkart believes that its focus on AI and machine learning will be an asset in chasing growth in small-town and eventually rural India.
“The onus of expanding the market and solving for difficult categories is what we have taken on as our responsibility,” CEO Kalyan Krishnamurthy told Forbes. “You have to cover, very deeply, categories which are related to their home; what they wear, which is the middle-of-the-market-fashion, the unbranded fashion market; what they eat, which is the groceries business.”
The other goal is to introduce a leasing model where customers can exchange their old phones for new models by paying an affordable monthly installment. To enable this, chief technology officer Ravi Garikipati and his team have been using machine learning technologies to build credit-scoring models for customers since the end of 2016.