AI Case Study

Booking.com resolves half of customer queries to its text chatbot in five minutes and without human intervention using semi-supervised learning

Booking.com has implemented a Booking Assistant chatbot to answer customer questions via text. The chatbot uses semi-supervised learning to successfully handle more interactions on its own, now resolving 50% without human intervention.

Industry

Consumer Goods And Services

Travel And Leisure

Project Overview

"Indeed, Booking is continually making changes to improve the bot product — conducting over 1,000 A/B tests a day across its offerings... [the chatbot takes a] hybrid approach, which uses natural language processing to understand open-ended questions but also offers menu cards with common topics and frequently asked questions and answers to help guide the user. When it looks like the bot won’t be able to answer a question quickly, Booking routes the query to either its customer service agents or the host property — depending on who the bot thinks is best placed to answer. For example, a traveler who has already checked in might ask about how to restart a hotel’s internet router. The bot may decide to route the question immediately to the hotel owner, since it is so specific to the property.

Notably, the bot now answers about half of all queries by itself, without having to escalate the query... regardless of whether it escalates or not, it can now answer half of all questions within five minutes. The bot can handle 90 specific subtopics, including everything from payment to transportation, arrival and departure times, date changes, cancellation requests, parking information, extra bed requests, pet policies, and internet availability.

The bot has responded to more than a million queries so far, Waters said, and the company plans to increase access to 100 percent of English-language bookings this year. Booking says it has more than 30 million unique users, but it does’t break that down between English and non-English users."

Reported Results

"The company says its Booking Assistant bot now resolves half of the thousands of daily customer support queries it receives within five minutes. That’s up from 25 percent in July and 30 percent in December, when the company launched the bot out of beta.

It’s too early to say whether the Booking.com bot itself is driving increased booking revenue because it only deals with questions that arise after the booking process. For example, travelers may inquire about early check-ins, whether they can bring a pet, or the size of a bed."

Technology

"The bot uses Active Learning, a semi-supervised machine learning approach that allows the algorithm to learn from interactions with the user, says Adrienne Enggist, Booking’s director of product for the bot. The company built the logic and AI entirely in-house because the range of post-booking travel topics the bot deals with is narrow enough that Booking remains the expert."

Function

Customer Service

Contact Centre

Background

"Travel giant Booking.com says with 5 million 'reported listings' it now tops Airbnb’s reported 4.5 million... Until now, Booking.com has been perceived as mainly offering hotel listings, a perception it is aggressively trying to change."

Benefits

Data

Details undisclosed; internally sourced travel information theoretically.