Travel Bots: Are They Here to Stay? - Glass House Relations
reg

10

Apr

Travel Bots: Are They Here to Stay?

Artificial Intelligence is the buzzword for 2017. AI has infiltrated industries as diverse as personal shopping, executive recruitment and medical diagnostics. Transformational fervor around AI has fizzled in the past, but AI-powered messaging platforms have energized innovation of late.

Enter the chatbot travel assistant. Startups have thrown down the gauntlet, enabling “virtual travel assistants” as early as 2015; Pana’s irreverent pitch to enterprises in October 2015 began: “Friends Don’t Let Friends Use Concur.” Several more have followed suit: HelloGbye, Mezi, TripActions, 30 Seconds to Fly and others have introduced mobile travel assistants, touting booking capabilities, with or without integrated corporate policies; offering proactive alerts and disruption support; and even expanding recommendation capabilities to local restaurants and entertainment, which is atypical for corporate travel agency services.

So far, the startups have targeted individual corporate travelers and smaller enterprises that have lightly managed programs, minimal preferred supplier agreements and straightforward travel itineraries. But it gets sticky for a bot to digest data and make judgments for negotiated, global managed travel programs that contend with policy parameters, frequent trip disruptions and traveler-initiated changes.

That hasn’t stopped established managed travel players from Concur to a number of travel management companies from pursuing similar AI strategies. After it acquired Hipmunk last October, Concur was particularly excited about the Hello Hipmunk AI tool, Concur EVP of platform and data services Tim MacDonald told BTN. He said Concur planned to act as an incubator for Hipmunk to work on bots to benefit the Concur enterprise product suite. On the TMC front, Carlson Wagonlit Travel’s Carla avatar has been in beta for several years, while Australia-based FCM Travel Solutions introduced its message-based mobile travel assistant Sam, which stands for SmartAssist Mobile, last July in the U.S. FCM plans to roll Sam out to Europe and Asia this year. American Express Global Business Travel has not announced a chatbot, but Oliver Quayle, VP of product marketing and innovation for Amex GBT and KDS now that the two have merged, cited GBT’s database as central to KDS’s post-merger innovation map, which looks like it will favor AI and machine learning and may build on the predictive and personalized approach of KDS’s Neo booking tool.

What’s the Goal?

“When I talk internally, I always tell the story of the great travel agent during my time as a road warrior at IBM when I traveled 150 nights a year,” said Travelport senior director of product innovation Nathan Bobbin. “She knew everything that I wanted in my constantly changing schedule. She’d say, ‘Hey, I know you’re supposed to go home, but instead, you are going to Tel Aviv. So here’s the trip I booked for you. You’re at the Marriott because I know you love Marriott; I didn’t take the early flight because there was no aisle seat. There’s an extra connection, but I knew you’d be OK with that …’ And she was always right because she knew all that stuff about me. That’s the experience powering the dream of AI, machine learning, personalization and mobile, where all those capabilities come together so that everyone can have one of those amazing travel agents in their pocket.”

Microsoft global director of travel, venue sourcing and payment Eric Bailey painted a comparison that makes today’s standard corporate online booking tool experience seem absurd. “You can almost go through the insane conversation you would have with an OBT: ‘Please give me an hour with a plus- or minus-three-hour range as to when you want to leave. In return, I’ll give you the lowest price without any reference to what your preferences are, and I don’t care if it’s a dollar cheaper or $1,000 cheaper.'”

Personalization and predictive results are just two features chatbot travel assistants aim to provide. Amid 24/7, globalized business and travel disruptions, imagine round-the-clock alerts and predictive rebooking support. Also within reach are expanded services that can recommend local restaurants and make reservations or that can identify entertainment and fitness options that appeal to the individual traveler. Indeed, some apps like Mezi and Pana claim they already can handle some of these details.

How Do We Get There?

The short answer is data—and lots of it. Co-founder and CEO Swapnil Shinde described Mezi’s behind-the-scenes structure: “When a user sends a message to Mezi, several different chatbots begin collaborating with one another. If AI detects that the user’s intent is to look and book flights, a flight chatbot will start talking to the customer. If the intent is hotel, a hotel bot will talk to the customer. So then we have a bot for dining and one for payments; we even have a bot for reminders and marketing.”

All these bots are powered by data feeds. Mezi uses global distribution system content, Expedia content, Priceline content. TripActions, for another example, has agreements with Sabre, Booking.com and Priceline. But travel content is just one data set. Chatbots also look at flight schedules, delays and cancellations. Data on weather or traffic conditions may power alerts, and historical conversations and bookings that the traveler has made within the tool can power recommendations. Mezi uses only historical data, while other tools begin with a profile and add to it through usage.

Other apps have forged data partnerships to bolster personalization. HelloGbye accesses American Express card data for hotel transactions to deliver personalized hotel results to users even if they don’t have a long history or a complete traveler profile stored with the HelloGbye tool. Juiced up with Amex card data, the bot will digest the individual’s historic bookings but also mine the transactions of other travelers to return “Amazon-like” recommendations. Making sense of these large volumes of data and turning them into relevant recommendations for individuals requires machine learning.

Machine Learning

Managed travel technologies have, for a long time, relied on rules engines to drive automation. TMCs use “if this, then that” scripts in mid-office systems to run quality-control and quality-assurance routines on trip reservations. Configuring an online tool to bias preferred partners in search results is another example of a rules engine’s work. All of that is familiar territory for agents and travel buyers.

Re-shopping tools like Yapta and Tripbam are based on automated mid-office routines that TMCs have performed for more than a decade, but with a twist. They define a cohort of comparable hotels or flights instead of shopping the same hotel or flight. That change—knowing the right cohort to re-shop—introduced one of the first “machine learning” innovations for managed travel.

“As those kinds of innovations expand outward from basic rules engines to more and more complex rules, you get to a point where you can’t manually configure all of the rules because there are just too many scenarios or you don’t know what the if/then [action] should be,” said Evan Konwiser, digital traveler VP for Amex GBT. “You get to a point where you have to create more mechanized solutions for a rules-engine effect. That’s where machine learning comes in, where you allow the algorithms to effectively set the rules themselves based on the data that is available. The enhancement of machine learning is that those algorithms self-evolve, meaning they can take feedback from a user from any number of [data] sources and they can improve the algorithms over time without somebody going in and manually configuring changes.” In reality, though, these algorithms are tweaked manually all the time as R&D pursues increased accuracy and better insights.

Algorithm-based machine learning has to happen in the background of chatbot apps—or of any travel booking tool—to increase personalization for the user. If built from historic data, that could include everything the traveler booked, plus details on what they did not choose. Or a bot could start with a rich data profile with all of the user’s declared personal information, travel preferences and loyalty alliances and then layer on subsequent usage data.

Machine learning also can help the tools themselves perform better as trips get more complex. When trip complexity makes it impossible for Mezi’s bots to collaborate effectively, the request transfers to a human travel agent. “The travel expert will step in, our chatbot will learn from that interaction, and next time [the bots] will be smarter in handling all Mezi users. The platform is designed in a way that it becomes smarter with every conversation,” said Shinde.

Natural Language Processing

Chatbots’ success relies on broad access to data, elegant user interfaces and personalized results, but natural language processing has been a game changer in making chatbots usable. It also has pushed innovation hard toward perfecting messaging platforms and delivery.

Online demos for HelloGbye and Pana, for example, show moderately complex trip requests going through the chatbots. Though chatbots work with both written and spoken communication, these demos showed spoken queries: Someone describes the trip aloud—complete with idioms—including class-of-service and style preferences like “modern” or “four-star,” and the tools return personalized results nearly instantly. HelloGbye also demonstrated a “conversation” in which the user was booking two travelers. After the bot returned search results, the booker realized he’d forgotten to say that only one traveler would be flying in business class and that the travelers would be staying in different hotels. He corrected himself, and the bot processed the revision and both bookings without a problem.

That’s the kind of advancement BCD Travel director of emerging technologies Miriam Moscovici said is critical for message-based apps to succeed. “Realizing the ability to have a cumulative conversation has been a dramatic change for natural language [processing]. Eight or nine years ago, you could ask a chatbot what the flight is to Los Angeles on a given day, and it would respond with some options and some prices, but that was it,” Moscovici said. “Now, you can respond again to the chatbot and say, ‘What about a day earlier?’ Entire lengths of conversations can be harmonized into one understanding of what the question is and be able to execute commands.”

 

Article taken from Business  News.

Image source: Chatbot Icon, Creativeworkline