Subscribe to Win!

Q&A: Can Artificial Intelligence book a Oneworld round-the-world airfare for you?


Traveltalk finds out how Elemental Cognition’s Product Manager, Greg Burnham, built Oneworld’s new Artificial Intelligence interface.

Elemental Cognition Product Manager, Greg Burnham.

Any travel advisor worth their salt will tell you round-the-world airfares are the most complex products in the aviation landscape to book. 

Navigating the already restrictive rules of multiple airlines and juggling availability on desired travel days to meet the traveller’s needs is no easy feat, even with the benefit of flexibility in the client’s schedule.

On top of that, hundreds of possible routings and millions of potential flight combinations make round-the-world journeys no easy feat for any advisor to manage.

So how could Artificial Intelligence be expected to complete this task instead?

That’s the task airline alliance Oneworld set for Artificial Intelligence (AI) firm Elemental Cognition, which recently completed initial trials of a beta solution it says “combines the power and flexibility of the best human agents” into a self-serve technology solution.

Operating largely as a chat-bot, the AI interface employs logical reasoning, natural language and dialog detection to understand the needs of a customer and navigating product rules and availability across Oneworld’s 13 members (14 if you count Affiliate Fiji Airways).

To understand the system better and see if AI can match it with human agents, Traveltalk dug deeper with Elemental Cognition’s Product Manager, Greg Burnham, to find out what went into developing such a complex piece of technology.

CLICK HERE to watch a demonstration of the EC Artificial Intelligence tool in action or CLICK HERE to give it a try for yourself.

 

What were the major goals Oneworld challenged EC to meet with this AI interface specifically geared to round-the-world airfares?

Oneworld’s round-the-world itineraries make for wonderful trips at great prices, but booking them is challenging. Travellers must comply with dozens of fare rules as well as navigate ever-changing flight availability.

Oneworld engaged EC to help customers navigate this complexity, allowing customers to personalise their itineraries without hitting frustrating dead ends. EC’s virtual agent lets customers take the lead in booking their dream trip while doing the hard work of plotting the optimal itinerary given customer’s preferences. When there are problems or trade-offs, the agent explains the situation in clear terms and presents the customer options so they can get back on track.

Qantas is a founding member of the Oneworld alliance.

Can you provide some detail on the voluminous data that would have been input into this AI system in its formation?

The core “problem-solving” that the virtual agent performs is driven by a relatively small set of data: a description of the fare rules and the different preferences customers may express. EC’s proprietary Reasoning Engine uses this understanding to dynamically guide the conversation with a customer to help them book their trip, all while following the fare rules, plotting a course with good flight availability, and optimising for shorter layovers and travel times.

For the virtual agent to engage in a conversational dialog with customers, it must interpret a wide variety of ways customers phrase their requests. The system is trained on examples of such customer expressions. While we include thousands of examples to fine-tune the system, one hallmark of EC’s technology is that it can achieve high accuracy at language interpretation right out of the box with very few examples.

Beyond that, the virtual agent ingests historical and current flight availability data for every city served by the alliance airlines.

Oneworld's RTW Artificial Intelligence tool is designed to problem solve complex itineraries.

What sort of problems were encountered in the development of this interface and how were these resolved?

We learned from early testing about the sorts of problems customers encounter. One common request being to visit a city that isn’t connected via alliance airlines to other cities in the customer’s itinerary.

Luckily, fixing this sort of problem was easy: we taught our Reasoning Engine that it could consider suggesting that the customer add an interim city that the alliance airlines do serve. EC’s platform makes the addition of these problem-solving strategies very simple, and the Reasoning Engine figures out on the fly what strategy it needs to apply in any particular situation, while still considering the fare rules, customer preferences, and travel distance. 

How long does it take for a Oneworld customer to book a round-the-world airfare on its website, from planning to purchase, and how is AI tipped to bring this down while still retaining confidence?

Anecdotally, customers could spend hours using the old system and often had to engage travel agents. However, using EC’s virtual agent, customers who reach a bookable itinerary do so in an average of 30 minutes.

British Airways is another longstanding member of Oneworld.

Does this AI remember previous interactions with a consumer and how does it inform the user about changes (new fares, new rules etc) that have been introduced between interactions? How does it navigate instances where more than one person might use the same computer?

Customers can begin to book their trip without logging in but have the option of creating a login so they can save their itinerary and return to it later. Itineraries are associated with these logins, so there is no issue of multiple customers on the same computer.

If a customer has an in-progress itinerary and returns to it, the virtual agent automatically checks whether there are any issues around changes to flight availability or fare rules. If there are any, the agent explains these to the customer and also presents the customer with automatically generated options to resolve these issues. 

What sort of "issues" can the AI anticipate as it relates to a user's RTW airfare queries?

The virtual agent tries to plot an itinerary that complies with all the fare rules. If this is not possible, it will inform the customer in real-time as fare rule violations arise and offer them options for resolving the problem.

For example, the fare rules allow only a certain number of flight segments within a given continent, including connecting flights. The virtual agent will try all possibilities for following this rule, such as ordering the cities in that continent to minimise connecting flights.

If this is not possible, it will explain the rule to the customer and offer them the option of dropping a minimal number of cities from their itinerary.

Another possible issue is flight availability. Not every city is connected to every other city on every day. The virtual agent automatically tries a variety of city orderings and arrival/departure dates to find an itinerary that works, all while following the fare rules and the customer’s stated preferences. If none of these possibilities work, the system will explain the problem to the customer and offer different ways they can modify their itinerary to get to a bookable itinerary.

Finally, customers sometimes make ambiguous requests. For example, “San Jose” may mean California or Costa Rica. The virtual agent spots when this occurs and asks for clarification before acting on the user’s request. 

What is the rollout timeframe for this AI onto the Oneworld website? When will everyday consumers see it live?

We are currently in a public beta on the Oneworld website. Anyone can use the virtual agent at https://rtw-va.oneworld.com/.


Click here to read the latest issue of traveltalk Click here to read the latest issue of traveltalk
Written by: Matt Lennon
Published: 13 July 2023

comments powered by Disqus