60px

451px

Amadeus

Blog

Making the AI journey from theory to practice

Martin Cowen

Contributing Editor

English
English
This content is only available in this language.

Martin Cowen is a contributing editor to the Amadeus Blog. He is a freelance writer, editor and moderator with a global perspective on B2B travel technology and B2C trends.

Artificial intelligence needs no introduction, although it might need a definition. Amazon is as good a place as any to start. It definesAI as “the field of computer science dedicated to solving cognitive problems commonly associated with human intelligence, such as learning, problem solving, and pattern recognition.”

Amazon’s definition also categorises machine learning and deep learning as “derived from the discipline of AI” which synchs nicely with IBM’s framing of AI as the “general concept that machines can be ‘taught’ to mimic human decision-making”.

Other subsetsof AI include logic, natural language processing and robotics.

AI and its subsets are all fed by data, a rich resource within the travel, tourism and hospitality sectors. McKinsey believes that “the potential AI impact [in travel] can more than double what is achievable using traditional analytic methods, amounting for between 7 – 12% percent of total revenue for the industry,” the highest revenue benefit of the nineteen sectors covered in its research.

 

Amadeus gets in early

While many businesses are now starting to think about how to operationalise AI, Amadeus has a number of AI-driven implementations live across its product and services, ranging from airline revenue managementto network planning. And they are already generating some impressive data-driven proof points.  The ability to execute is not the result of a reactive last-minute FOMO spending frenzy, it is a result of decisions taken more than a decade ago to move off monolithic TPF mainframes and onto open systems.

Denis Lacroix is SVP of Core Shared Services, taking on the role in 2017 after nearly thirty years with Amadeus. He oversaw the final decommissioning of the mainframes and the seamless transition to open systems. 

“We saw that technology, travelers and our business, were changing, so we took the decision late 2006/early 2007 to get out of TPF mainframes and into the new world of open systems. That decision meant that when the cloud, APIs, platforms and infrastructure as a service came along, we were in prime position to move, we’d already done the groundwork.” - Denis Lacroix, SVP Core Shared Services, Amadeus

The benefits of the move to the cloud in the specific context of operationalizing AI is that Amadeus can source, sort and streamline vast amounts of internal and external data upon which the AI, machine learning and deep learning applications are based. Test and learn is more effective, and any AI-based products or services can be deployed at scale. of this community.

Machine learning is the AI subset which can deliver the value to our customers,” Lacroix explained. “We are pretty close to having a million or so machine learning models in production.”

Amadeus has recently made another big bet on the technology after pioneering a set of Artificial Intelligence APIs  on its Amadeus for Developers portal. These APIs will “democratize” AI for startups, independent developers and emerging businesses, giving them the right tools to develop sophisticated solutions that could predict travel intent, traveler behavior, and flight delays, amongst others. Innovators can start testing and collaborating in just minutes - with no prior background in AI or data necessary. Initial feedback  has been more than promising at a recent Amadeus hackathon, and the Innovation Launch stage at the Phocuswright Conference.

Searching for a use case

If AI is the “general concept that machines can be taught to mimic human decision-making”, many travel industry processes can become AI use cases. One clear example of where AI processes are now fully embedded into Amadeus is in some of its flight search products.

Every traveller is familiar with the frustration of trying to find the right flight when pages and pages of results are presented. That frustration is compounded when the results take a long time to load, hence the strong correlation between quick response times and high conversion rates.

The AI compatible systems in the background can curate which options appear in front of the traveler, having learnt from previous interactions, knowing which options usually convert for this particular flight. 

Amadeus research shows that people choose the cheapest option 57% of the time , so the system is taught that this is how the results must be displayed. But before this, the system must learn how to identify the cheapest option for a particular city pair and to ensure that the ticket is available, at the price, and at the time of the search. 

The AI discipline in play here is “supervised” machine learning – supervised in that the algorithms learn from prescribed data sources to achieve a pre-defined result. Unsupervised machine learning is more maverick, with systems given free rein to access and interpret data sets with a view to discovering something which, if viable, can then be supervised into a pilot or production.

The Amadeus in-house data sources used to bring up the cheapest available fares include the itinerary builder, the availability server and the fares engine. All these insights combine to create what is known as “findability” – the ability to find the cheapest, available price. The results so far are impressive. OTAs using the AI-backed search tools can increase findability by an average of 5%, while metas can see a 20% increase in traffic. The resulting increase in gross bookings across the ecosystem is around 10%. 

Connecting with a use case

“Virtual interlining” is another text book example of AI in action. Creating a trip involving a connection between two unconnected carriers is a growing area of interest for airlines, travelers and tech suppliers.

The claim from some green screen cryptic command experts – that this has always been possible – is not without merit. An old-school retail travel agent with an encyclopaedic knowledge of airline schedules, airport layout and baggage restrictions could assemble an itinerary, as two separate transactions. 

Technology has made the manual process better, but artificial intelligence is the only way to make virtual interlining become seamless and a viable, monetizable product.

For virtual interlining to work, the AI needs access to every airline’s schedule, availability and pricing engines, including those not held within Amadeus. And there are other components to a successful virtual interline booking which need to be integrated before the results page appear. Connection times at the airport are critical – is there enough time to disembark, collect and re-check-in hold luggage (if relevant), and get to the correct departure gate? 

Even once the AI has learnt which connections are viable and which are not, it’s work is not yet done. Are the results displayed according to total trip time or the cost? Which converts better? Can the trip be insured, connections guaranteed, and at what premium? Can disruption be handled seamlessly?

Fully automated virtual interlining is one for the future, but Amadeus believes that cracking the codes will deliver a great result for airlines because it delivers a great result for travellers. 

Amadeus is currently working with four OTAs on pilot versions of virtual interlining. Results so far show that virtual interlining is coming out more than 25% cheaper than traditional interline, codeshare or partnerships in more than one-in-five itineraries. But findings also show that AI-generated virtual interlining pricing is more expensive for nearly four out of ten itineraries. 

Conclusion

AI is not yet the answer to everything, but having an AI-friendly corporate and IT infrastructure in place makes business sense nonetheless. Companies need to be able to attract top talent to their tech teams and have a business mindset which will allow this talent to thrive. AI and its subsets are built and developed using data, of which there is no shortage in travel. The biggest challenge facing the travel industry and its tech partners is which use case to go for.


Tags

Airline IT, Traveler Experience, Travel Search, Artificial Intelligence, API, Guest Post


Do you want us to email you the latest insights?

Subscribe to updates