Machine learning and artificial intelligence capabilities like natural language processing are gradually making their way into the managed travel booking experience. A maximally personalized shopping experience, however, relies on technologies beyond the reach of even the most sophisticated algorithms and bots. Plus, privacy regulations could slow progress.
Even so, chatbot-powered mobile booking assistants and AI-enhanced travel management companies and online booking tools are getting better at predicting what travelers want, weighing that against what their companies require and delivering balanced itinerary options. Adoption of such tools, however, at least among the managed travel-centric BTN audience has not been high. Just 7.6 percent of travel buyers said they provide AI-enhanced booking technologies to their travelers.
Mobile Travel Assistants
Chat-based mobile travel assistant technology has been a hotspot for managed and unmanaged business travel innovation. TripActions, 30SecondsToFly, HelloGbye, Hello Hipmunk and others are working on their algorithms. Even startups that have launched in the consumer travel space like Paul English’s Lola have pivoted to business travel. That’s largely because machine learning and algorithms require lots of data; they need to digest a lot of trips, ideally for the same individual, before they can personalize itinerary options.
Tech investors agree. Concur bought Hipmunk in 2016, but Concur, a traditional online corporate booking tool, has yet to capitalize on Hipmunk’s AI technologies for Concur Hipmunk, its new solution designed to help small businesses capture unmanaged travelers. Credit card network American Express acquired travel-oriented chatbot provider Mezi in January and is turning it into its proprietary AskAmex travel tool for card-holders. TripActions received $51 million in Series B funding last month and is looking to build infrastructure technology that will unleash the real power of AI for managed travel. Indeed, even with the excitement in the mobile travel assistant space, there’s a lot of work to be done—and traveler expectations are high.
Travelers expect chatbot-enabled mobile booking technology to respond quickly, to understand complex requests and to be flexible in how travelers search for flights, according to 30SecondsToFly CEO Felicia Schneiderhan. Hipmunk CEO Adam Goldstein said travelers want bots to be “explicit and helpful and not be too magical.” Yet, a little magic might be nice, according to TripActions chief technology officer Ilan Twig, who said today’s overall user experience is still frustrating and not traveler-centric enough.
Travel bots have gotten better at understanding what travelers want. Schneiderhan said 30Seconds-ToFly’s AI-travel assistant, Claire, can optimize a traveler’s options by considering both a traveler’s explicit flight request and the suggestions the traveler has chosen during previous searches and by then combining those with the corporation’s preferred supplier set and policy constraints. Claire then pushes four personalized flight options to the user, and 98 percent of the time, the travelers choose one of those four. Agents step in when flights need to be rescheduled or canceled.
Hipmunk can understand nuances in the language of traveler requests and the context in which a trip is booked, said Goldstein. It also can help with open-ended questions. “We’ve gone from [an experience in which] all you can do is chat to [an experience] where you can book inside a bot to [an experience] where you can actually authenticate and get specific trip information personalized for you,” said Goldstein.
Of the 10 options TripActions recommends at a time, users pick one 91 percent of the time, Twig said. The median time a user spends booking on the platform is six minutes. During travel disruptions, the TripActions bot scans for impacts on an itinerary and reaches out to the traveler. “The bot is putting you on the next flight, communicating to you that we noticed the delay, found an alternative and it is refundable, communicating the details of the flight and telling you that your itinerary has been updated,” said Twig. “It’s all automated.” The agent steps in if the traveler has a question or rejects the bot’s recommendation.
Now, 30SecondsToFly is “building more narrow bots that can coexist and then basically bounce the user between them depending on which context they need, type of intent they have,” Schneiderhan said. “That is the type of bot that will be speaking to them. [Using multiple bots increases] accuracy and speed for a broader variety of user requests or user intents.”
With Claire, Schneiderhan is trying to make travel processes and policies more dynamic and intuitive. The company is working to give Claire the ability to process trip approvals. Similarly, expense reporting technology providers are using AI to relieve supervisors of approval tasks. Adoption in the expense arena, however, has been challenging.
TripActions is experimenting with blending data from new sources. It would like to access details like employee seniority to provide more personalized features. It’s also working on reducing its 10 recommendations to three and reducing the median time users spend booking to one minute.
The Next-Gen TMC
Chatbot-enabled personal travel assistants need fulfillment partners in the background: TMCs. Christopherson Business Travel is an investor in and the fulfillment partner for 30SecondsToFly, at least when the company sells its product directly to corporations. Schneiderhan actually pivoted the company in March to serve tech to TMCs; in that model, fulfillment would shift to each client TMC rather than be provided by an umbrella partner TMC. Few travel chatbot providers have gone that route. The one that did, Mezi, is largely out of the picture now that Amex acquired it. Schneiderhan told BTN sister publication The Beat last month that 30SecondstoFly has contracted with three TMCs. That’s the same number Mezi had on board when Amex bought it.
Among Mezi’s three TMCs were Silicon Alley-based WTMC and Silicon Valley-based Casto Travel. Casto president and CEO Marc Casto decided to go with Mezi last year, based on demand from tech-savvy clients who were eager to have a sleeker, more consumer-grade technology experience for their corporate travelers. But not all clients are ready to jump on AI-enhanced travel tools. “Some people prefer to call an agent. Some people prefer to go to an online booking tool. Some people prefer to use a text-based messaging system like a chatbot,” said Casto. To satisfy various client needs, Casto Travel created specialized workflows and training through which a group of agents can work with AI-enabled agency desktop tools. The agency desktop tools take in and steer content from the AI-enhanced traveler-facing chatbot, Marco.
Marco provides travelers with three personalized options: “what is best based on the request, what is best based on the company policy and … what we feel is most appropriate based on their prior travel purchases, what they’ve always enjoyed in the past. So it’s very intuitive based upon prior purchasing,” said Casto. He said the agency is getting closer to finding the middle ground between the preferences of the traveler and the requirements of the company.
In terms of language recognition, Marco has the capabilities of a “third or fourth grader,” but with additional human inputs every day, the agent team is training Marco to understand conversational nuances. Casto hopes Marco will communicate at a high school senior level within two years.
WTMC CEO Sarosh Waghmar uses Mezi technology as part of a broader agency technology strategy he calls Bots & Beings. He told BTN Mezi can handle day-to-day traveler requests like flight searches, recognize a user’s loyalty status and deduce a traveler’s preferences from historical booking patterns to provide personalized travel options that are still within a corporate travel policy. Human agents intervene when a request requires lateral thinking, which happens often. WTMC is taking the pragmatic approach to enhance agent intelligence without trying to remove the human touch.
Waghmar, who was a Mezi board member prior to the chatbot’s sale to Amex, praised the technology’s user interface and its natural language processing, but he sees real limitations for personalization. It’s not exactly the chatbot’s fault. Instead, it’s the antiquated TMC infrastructure used to fulfill the requests, he said. Real personalization has to dig deeper, he told BTN; it has to be initiated at the TMC level by technology that uses the traveler profile as a dynamic filter for recommendations presented to the traveler. Right now, he said, that technology doesn’t exist.
“Today’s enhanced AI tools fail to provide personalized booking experiences to travelers [because] they don’t link up to the traveler profile. They don’t even know what’s in the profile because that information is dead; it’s just static. [All the profile information] has to be pasted [after the fact] at the time of completing the PNR,” he said. “When I start to search, [the booking tool] just shows me the cheapest. Then I have to filter it myself to show me just nonstop, then to show me by airline preference. Why am I doing that if I’ve already entered my profile? Why is the profile just lying there?”
WTMC champions cloud-based profiles that integrate with multiple technologies to accommodate whatever process the traveler chooses to initiate the booking. “If I need to push [the profile] out to any of the GDSs, I’ll be able to push it out. … If I need to populate it on Concur, I’ll be able to sync it to Concur. I’ll be able to push to whoever and whenever I want,” Waghmar said.
The profile is one side of the personalization coin, providing rich information for an AI-enhanced tool to understand the traveler. Delivering a tailored product, however, is the other side of that coin, and it’s also waiting for a solution.
The momentum building around New Distribution Capability is a major step in the right direction. It will make more-detailed air content choices available to both agent desktop tools and mobile booking tools. Waghmar’s WTMC was the first Level 3-certified NDC TMC.
Casto said NDC will have a major impact on AI’s ability to provide personalized experiences. “The ability to access very personalized fare requests for a traveler in a mechanism that is specific to their needs … well, that’s exactly what we are trying to achieve with our chatbots: How do we tailor this specific to the traveler’s desires? Bringing in [NDC] content will be extraordinarily helpful.”
Slow Going for Traditional Online Booking Tools
AI for managed travel booking
tools is clearly still in its infancy, which accounts for much of the low
adoption. Another contributing factor, however, is that travelers and travel
managers may benefit from AI enhancements without knowing it—particularly
those users in the vast empire of Concur Travel, which has been conservative with AI. The company, which has
captured 57.4 percent of the global T&E management software market,
according to IDC, has made a few machine learning enhancements to its
online booking tool particularly around hotel search. It currently uses machine learning to analyze users’ past booking patterns
and anticipate the hotels they will most likely book for their next trip. It also identifies similar searches for groups and suggesting the same property choices for participants,
according to a Concur spokesperson. Machine learning also powers some of TripIt’s key itinerary
management features.
It’s a big decision to choose a booking tool, and there may not be an option for a company accessing the booking tool through its TMC. AI and personalization, at this point, may not be a big decision driver; even more, said Concur Labs VP John Dietz, corporate clients may object to it.
He said Concur has been conservative with introducing personalization and predictive features due to privacy concerns, especially for global clients in the era of the European Union’s General Data Protection Regulation. “We as an industry are trying to work on how we preserve our consumer’s rights to be forgotten [while also] predicting what they want next. Balancing those [opposing options], we are going to come out on the side of what [corporate] clients want in terms of privacy first then prediction second.” Dietz added that when Concur is ready, its likely focus will be AI-enhanced trip-disruption recovery, which requires more nuance than bookings do because they require more context about the particular situation.
Not all booking tools have taken that tack, however. KDS made a big splash in 2013 with its Neo booking tool, which not only gives door-to-door itineraries in a few clicks but also applies traveler preferences to its recommendations. It bases those on the individual’s historical bookings and on various content inputs. It looks at more than just the obvious ones in order to optimize routes.
When then-CEO Dean Forbes debuted the tool in Paris, he said, “The algorithm is built so that it learns. We won’t be right 100 percent of the time. We have been testing Neo on ourselves. It suggested Oliver [Quayle, then KDS SVP for products] fly from his home to our Paris office via Southampton Airport. It was smart enough to propose something he hadn’t thought about before, and now he is using that route regularly.”
That was before the term machine learning had become an industry buzzword, but KDS has continued to build on the concept since American Express Global Business Travel acquired KDS in 2016.
Dietz believes the industry is close to fully understanding traveler requests, but like Trip- Actions, Casto Travel and WTMC executives, he questions how the industry will deliver personalized products, given that distribution technologies are antiquated.
KDS Neo director of global product marketing Amber Stauffer said the company is focusing its machine learning on enhancing the hotel amenity data coming from the global distribution systems. This would allow Neo to deliver on travelers’ personal preferences by presenting more consumer-style hotels for business travel.
“The next job for all of us in the industry is that once we understand the request, how are we going to fulfil that request?” said Dietz. “That is the kind of work that is happening in research labs.”
_____________________
Correction: The printed version of this article erroneously asserted that Concur Travel had not made machine learning enhancements to its booking platform. BTN regrets the error.