Amadeus builds custom big data platform

Five years' work to display fewer, more relevant flights.

Travel IT giant Amadeus is beta testing new flight search technology after spending the past five years building a custom big data platform to power it.

The new ‘Featured Results’ product will present travel agents and their customers with four targeted flight search results while waiting for a more comprehensive list of options to load.

It will be offered to travel agencies worldwide by March, with Amadeus’ travel technology consulting VP Sebastien Gibergues expecting it to be particularly useful to mobile website operators.

“The fact is, we process more data than ever to return only four results,” Gibergues told iTnews, describing them as the fastest, cheapest, most popular, and ‘sponsored’ options.

“The results are instant and relevant, and relevance is really key here … It’s proven from the retail world that this actually facilitates the decision-making process.”

Gibergues highlighted American psychologist Barry Schwartz’s “paradox of choice” – the concept that limiting consumer choices facilitated decision-making.

He said Amadeus’ beta testing partner, online travel agency Vayama, had reported a 16 percent growth in its ratio of sales to searches in the first week of testing.

“[The concept] applies to Google, it applies to supermarkets and it applies to travel as well,” he said. “Very few people – I think it’s 8 percent on Google – actually go to the second page of results.

“This is really technology for the online travel world – the B2C [business-to-consumer] world.”

Powering fewer results

Gibergues said Featured Results was the result of about 100 developers’ work over the past five years – accounting for “a big chunk” of Amadeus’ yearly development budget.

The product relies on three backend systems:

• a massive computation platform that computes prices and routes daily;
• a massive search platform that performs user queries; and
• a cache manager that constantly monitors airline data and forces any necessary recalculations by the massive computation platform.

Amadeus provides each of its travel agency customers with an in-memory cache of “several hundreds of millions” of live airline prices and half-a-billion reservation records that could be up to a year old.

Customer caches are populated by the shared massive computation platform, which runs on Amadeus’ virtualised Linux infrastructure and proprietary application stack.

The platform takes into account current and historical flight prices and availability, as well as historical search information, to determine the most popular option for any given search.

“With the technology that we need to make [search results] relevant, compute the prices and keep prices accurate, there were a lot of pieces that needed to come together to deliver this very simple and very efficient user experience,” Gibergues said.

“We’ve built our own in-memory storage platform with travel-specific indexes. Data models and indexes are proprietary to Amadeus … we built our own big data platform.”

Amadeus has faced ballooning IT demands due a growing ratio of consumer queries to bookings – from 27 queries for every booking in 1996 to 270 per booking today.

Gibergues highlighted the cache manager as an example of how Amadeus optimised transactions to lower IT costs for the travel supply chain.

“We’ve actually built an intelligent and self-learning system in the cache manager that guarantees a certain level of freshness in the cache,” he said.

“We tell the cache manager that we need this level of accuracy in the data and the cache manager will self-learn and force computation to maintain this level of quality.

“Look to book ratios are going through the roof; people are searching more and more online before they buy,” he said. “If we are not able to optimise transactions, the industry as a whole may not be able to sustain the traffic.”