On the surface, online travel companies haven’t really changed much over the years. The way you find and book tickets remains mostly unchanged from how it was five years ago. Search, browse the listings, find the best deals, and make the booking. Under the hood, there are a few areas that have been a focus, explains Vikalp Sahni, CTO at Goibibo. «There are three problems that we solve as an e-commerce company,» says Sahni. «One is search and discovery. The second is reservations or booking or purchase. Third is the post-booking experience.»
Of these, the second is becoming easier and easier, as payments systems get more robust. But for the first and the third, artificial intelligence is playing a major role in improving how the sites function, even if it’s not very visible to the customer.
«Search and discovery is a part where you need a lot of data, you need to see what all hotels are available, what all flights are available,» says Sahni, explaining how a signed in user’s search results are ordered. «So a recommendation uses the [customer’s] search pattern, his booking history, the current inventory and rate thresholds, location preferences, of the user, of the hotel he bounced off, and many more. The more we go ahead and create an algorithm which can extract information from various data-sets, the more we become accurate.»
«Very simple could be, you go to a search results page, you look for a hotel in Goa, you pick a hotel, but then you bounce off,» he continues. «Now, you come back and search for Goa again. Now, we would have looked at what kind of hotels you looked at, and are interested in, and we create a new carousel based on that information. And hotel booking is a tedious process, especially when you are doing leisure, and the metric that we at times see, that the overall time taken for users to book reduces.»
According to Sahni, GoIbibo has seen a 15 to 20 percent improvement in terms of booking time reduced. «This is something that — a number that has to keep growing as long as we are doing the right thing,» says Sahni, adding that the bounce rate and exit rate that the company sees on pages also went down with results being customised for users.
Computer vision and conversational platforms
That’s not the only place in which AI now plays a role for the company either. Computer vision is another area of AI that has been very useful, Sahni tells Gadgets 360. Companies like Myntra and Flipkart use AI to improve the metadata on their photo catalogues, and Goibibo is doing something similar for the photos travellers upload as part of reviews.
«So when you go to the reviews and rating system, when you upload images, these images are automatically tagged, we automatically tag that if it’s a pool image, a beach, etcetera,» says Sahni. This has also helped in one more way we weren’t expecting — preventing porn from showing up on the review pages. «When [users] upload the traveler photos — it can be anything, it can be a porn photo as well, so we need to have a moderation team. Now, we’ve built this capability where we’re able to identify these images much better than earlier, and it doesn’t require a manual intervention. It reduced the manual workflow by 60 percent. We have a confidence score, and out of 100 images, 60 images will be extremely confidently tagged. Rest of the images we extract tags from it, but it still goes into the manual moderation.» It’s not foolproof but it speeds up the process of moderation, which means that more reviews are available for customers looking to make a booking, which can increase the likelihood of a customer finding the right option.
Beyond this, conversational interfaces are the order of the day for Goibibo. Like Ixigo, Goibibo has also built its own smart assistant, though it’s more of an adaptive layer that can be used to connect Goibibo’s data to various conversational platforms, unlike the former’s vision for a standalone app.
This enables the company to use a number of different platforms, ranging from WhatsApp, and Facebook Messenger, to Amazon’s Alexa on the Echo speaker. «It does NLP, entity recognition, it is an adaptive layer which allows us to build interfaces on our own deck, which is the Goibibo app,» explains Sahni. According to him, conversational platforms are well suited for the post-purchase experience, when customers want support, information on status, and cancellations. This is something that is traditionally run using call centres, but it’s possible to reduce that requirement through the use of AI.
«You can ask Goibibo on WhatsApp what are the cancellation charges of my hotel or get me by booking or what is the flight status, all commonly asked queries from the consumers, you can send to Goibibo on WhatsApp and we will start sending you responses,» he says. «There are still scenarios when our NLP engine will not understand because it’s still nascent, but we will try and send you a link to a different type of workflow, and there you can actually go ahead and chat with our manual call centre folks as well, in case you have a very random problem we have not been able to program for yet.»
Between an app on Google Play and the App Store, there are some common denominators, such as displays and touch interfaces, Sahni explains. However, with the different chat platforms, there are a lot of differences. «In Alexa you have voice, while in WhatsApp you have a visual feedback, but you don’t have action control. In Facebook Messenger, you have visual, voice, and action control,» he says. «We have built a thin adapter layer, that is common across all these apps. Think of that, a framework that we have built, with the core NLP, the core NLP, and the core entity recognition, and then we have a layer that understands what capabilities the interface offers and give the response accordingly.»
«It’s getting more and more complex as we get multiple conversational interfaces, but it’s an interesting development,» Sahni adds.