PremiumAnalysts say that travel sites, including Yatra and Cleartrip, are increasingly using machine learning (ML) models to predict customer behaviour. User data is also being used for product development
New Delhi: As the online travel market in India continues its upward trajectory—it is growing at about 11% annually and is expected to touch $48 billion in revenue by 2020, according to a June 2017 report by the Boston Consulting Group and Google Inc.—firms in this segment are ramping up the use of cutting-edge tech tools to boost their business.
Yatra Online Pvt. Ltd and Cleartrip Pvt. Ltd, for instance, are focusing heavily on bots, machine learning (ML) algorithms, caching servers and other technologies to convert more users coming on their online and mobile platforms into paying customers, as well as to offer them value-added services.
One of the trends that could soon pick up pace among users is voice search capability, believes Manish Amin, co-founder and chief information officer of Yatra, which has more than five million customers, out of which 82% are repeat customers. He thinks that the time may now be “ripe" for voice.
“Even though we launched voice as a feature two years back, it did not really take off because devices like the Google Home or Amazon Echo had not achieved mass usage in the market," he says. Starting from plain flight search that landed customers “straight onto the results page", voice is more of “an interactive conversation now" wherein users are prompted or guided to multiple specific options.
“People can also say, ‘Only show me Indigo flights’, or ‘Only show me early-morning flights’," he adds.
To enable voice bookings on Yatra.com, the company took a “hybrid approach" in which the users can “switch seamlessly" between voice and text chat, according to Amin.
The company has created what it calls a “knowledge base" containing the most frequently asked questions by customers dialling into its call centre. This is run by bots which keep improving as “different people ask the same question differently", says Amin. For this purpose, Yatra developed its own natural language processing engine. Explaining the rationale, he says, “Since you need a lot of data for the system to be trained on, we thought it would be better if we do it internally so that we could understand what people were looking for."
Cleartrip.com, on its part, has built a recommendation engine using ML—it has used the technology in its “Local" product feature (internally at the company, it is called the “activities" business). Activities ranging from day outings and camping trips to parasailing and cultural walks—conducted through partnerships with multiple agencies—are offered as value-added services to customers. “We have sold 7,000-8,000 activities on our platforms since we started this business. We are present in around 50 cities in India, besides in Dubai and Abu Dhabi," says Suman De, director of products at Cleartrip.
Over the past one-and-a-half years or so since the activities business was launched, Cleartrip has gathered “enough data for our ML algorithm to predict what people of a certain profile may like to do next" in terms of activities, according to De.
And what if there is an altogether new activity not listed in the system? “This does not happen that often now," he says. “But we have people who enter such activities into the system and curate them for users." Such a dual approach, says De, helps the company pick the best of options from a list generated automatically by the algorithm.
“Another area where we are using ML is in what we call ‘sort order’ in accommodation so that the search could turn up the best order of hotels on the results page," says De, adding that this is still “a work in progress".
In the airline booking business, De says that ML is used in two products, “fare alert" and “price lock". “In fare alert, you can choose to be alerted on airline fares going down on a particular date or period. What the algorithm does is monitor the fares for the specified flights or airlines and automatically send out the alerts when the prices go below the set threshold value, allowing you to book at just the right time," he explains. The company is also trying to predict, through ML, when the prices are “likely" to come down—and that is a challenge, he says, because there are so many factors affecting prices. “It is not an exact science yet," says De. As such, the company is running tests and trying to figure out how to improve it further.
A very critical factor for any online business is how fast the search results appear—something that can be quite tricky for a site like Yatra.com which, according to Amin, has additional data on “thousands of travel destinations". Its “explore" feature, for instance, allows one to search the cheapest cities to travel to in Europe, Africa or Asia in different origin-destination combinations. “The cheapest prices are shown from the day of search to the next 12 months—and the results take only a couple of seconds to appear," he avers.
“Most of these things we have achieved by using the caching technology. For this we use a content delivery network (CDN) such as the one from Akamai," says Amin. A CDN speeds up the process of providing search results, and is especially useful for caching images for faster load times.
Analysts opine that travel sites are increasingly using ML models to predict customer behaviour. “Employing ML models on the past transaction data, online travel companies are able to make predictions as to when it is their customers are likely to travel next. They are using these insights to develop customer-specific marketing campaigns," says Alok Shende, managing director of Ascentius Insights, a research and analytics consulting firm.
User data is also being employed, according to him, for “new product development". Some of the new products being offered by online travel firms are “very niche" and localized. For example, he says tour operators have started offering 10-day treks to the Sierra Nevada mountains in the US and other niche destinations such as the national parks.
“Unlike the standard fare of pan-US tours that would take tourists to a dozen destinations, tour operators are discovering—through Big Data projects—which of their customers can be targeted for these unique product offerings," concludes Shende.
