Opportunities and challenges for travel brands using AI

Artificial intelligence, specifically machine learning, is being used across the travel industry to expedite and improve data processing, saving companies both time and money. But there is still work to be done to ensure executives understand the advantages, limitations and appropriate use cases of these technologies.

That was one of overall themes shared in comments from panelists during the travel and hospitality session of DataArt’s IT NonStop 2020 online conference.

DataArt senior vice president of travel and hospitality Greg Abbott began the discussion by asking for reaction to a statistic that 50% to 90% of all AI/machine learning projects fail.

Andy Owen Jones, CEO of BD4Travel, says his company has done between 30 and 40 implementations, and only three have failed, but, “I can imagine if you are starting off, you’ve got no clean data, you’ve got people that are skeptical, and you are trying for a moonshot, it’s extremely difficult to cross that chasm in one go. And because people are a little skeptical, they are not willing to let you,” he says.

Panelists say failure may be a result of a lack of specific business goals at the outset, a misunderstanding of what is even meant by AI and machine learning and a failure to determine if a more simple, and possibly cheaper, solution could be used.

“You have to have very clear reasonable targets, you have to have a reasonable size problem to solve and you have to have absolute buy-in from the teams that are going to be using it,” says Arnold Bramnick, CTO of Norwegian Cruise Line. 

Hudson Crossing co-founder and partner George Roukas says part of the confusion can be blamed on the technology industry, which has been allowed to put “comes with AI” on its products indiscriminately.

“We’ve really done a disservice to our clients in totally mucking up the definition of AI. What is it, what are you buying, what are you thinking you’re going to get out of it,” he says.

“There’s a lot of responsibility on our part to make sure that people understand what exactly this stuff is and how it’s going to give them some sort of an outcome, not just an output.”

Anna Jaffe, CEO of Mobi Systems, says her company considers AI to come in three forms: automation, which means digitizing tasks that are better handled by technology than humans; artificial intelligence, which includes tasks that humans do well and AI is still learning to do, such as natural language processing; and associative intelligence, which Jaffe defines as thing people cannot do, such as processing huge quantities of data in real time and in context. 

IDeaS chief evangelist Klaus Kohlmayr says his company uses machine learning in its pricing solutions for hotels, but ultimately clients care less about the technology and more about the results.

“When we have conversations with our conversation about new methodologies, we don’t talk about what we are using to achieve the goals, we talk what the ROI is and what the output is and how it's going to make their business better,” he says.

And Sabre’s vice president and head of research, Sergey Shebalov, says it is important to recognize AI is just one tool and may not be needed in every instance. Shebalov shared an example of an effective use of a machine learning algorithm to identify the optimal price for an airline ancillary product, such as a seat with extra leg room.

“Based on historical data and continuous learning of how customers respond to different prices, we are able to recommend optimal prices,” he says. “And once our recommendations went live, compared to manual practice, we saw an uptick in conversion rate and several-million-dollars-a-year benefit from just one ancillary that customers were buying.”

Shebalov says Sabre has also used machine learning to predict demand for flights between two cities during a specific time in the future.

“This is a very good use case because it takes a lot of different data to understand the volume of demand ... and we see where machine learning methods help improve the accuracy of these predictions compared to basic linear regressions, which we used before. And accurate knowledge of demand helps airlines plan their networks, helps hotels to create their offers and so on.”

AirHelp chief technology officer Jakub Dziwisz says when his company deploys automations, it begins by testing a small sample in comparison to results produced by humans looking at the same data. That is how AirHelp developed its AI-based airline claim processing system.

“We deployed a bot that assesses whether a case against an airline is strong enough to go to a court,” Dziwisz says.

“We wanted to make sure our way of predicting cases makes sense. We had Lara, our bot, and humans doing the same exercise ... and we compared the results, and this is how we proved the system is good. And now Lara is handling 95% of cases we are processing.”

Jaffe says Mobi Systems' core use case for AI is to help brands analyze data to develop an advanced understanding of their customers. 

“Not trying to demographically group people into static segments and not trying to build a single static profile of someone. [Rather] trying to have a have a spatially, temporally aware model of who your customer is that works in real time and is online so you can respond to them,” she says. 

Jones shares a similar example from BD4Travel’s work with AI.

“What is value of customer? What is their propensity to do something? And then working out how to help old travel systems reshape in real time to respect that,” he says.

“The prevailing non-AI approach, or prevailing digital approach, is to build the best possible site for the average user whereas I think ...  our vision is how can we design a site or a mobile channel that wraps itself around the user because we’ve accurately predicted what they are after in real time. You simply can’t do that with rules.”

Roukas says examples such as these are needed to move AI and machine learning farther along the “adoption curve” and eliminate existing confusion.

“We have to get to where we have enough small success stories everywhere ... those are the referenceable case studies that will push us across the chasm, get us to that next group of adopters and then allow this thing to really fade into the background completely and ... it will just be baked in and we’ll be reaping the benefits much more uniformly,” he says.