Travel Tech Saber Ink deal with Google • The Register

characteristic The computing and travel industries have traveled hand-in-hand for decades. Perspective: American Airlines signed a contract with IBM in 1957, which developed the first computer reservation system based on two IBM 7090 mainframes in 1960.

Over time, the same booking system morphed into one of the top three global distribution systems (GDS) called Saber, which trades flights from many airlines and delivers them to travel agents and consumers.

But with that legacy comes a problem when it comes to catering to 21st-century consumers.

“The data formats that have been around in the travel industry have been around for a very long time, decades,” said Andrew Gasparovic, chief architect at Saber Labs, the technology division of GDS founded in 1996.

“They were designed at a time when not much thought had been given to what could be done with data The registry.

Gathering data about customer offers and the reservations they eventually make is common in e-commerce, helping sellers anticipate what customers are likely to buy next.

But according to Gasparovic, collating this data across more than 12 billion shopping queries and 1 billion travelers each year was no trivial task, and a task entrusted to a partnership with Google established in 2020.

Saber is in the process of migrating its IT infrastructure to Google Cloud. It also introduces operational data tools including Managed Systems Spanner, the distributed database powering GoogleAds, and BigTable, the wide-column and key-value NoSQL database.

When analyzing, it uses BigQuery, Google’s distributed data warehouse.

Other GDSs include Amadeus GDS – founded in 1987 by Air France, Lufthansa, Iberia and SAS Airlines as a Europe-based alternative to Saber – and UK-based Travelport (which includes Apollo, Worldspan and Galileo GDS). All of these networks started as live ticketing between airlines and travel agencies, but now also work with travel websites, car rental companies and hotels.

booking please?

Saber’s earliest travel distribution network predates the Internet. An airline’s reservation database stores a passenger’s booking information, seat selection, tickets, special requests, and other important information about their trip. Saber typically processes thousands of reservation updates per second on behalf of carrier customers. An airline’s reservations database must be served by many availability zones.

Meanwhile, the flight shopping system generates millions of travel plans per second on behalf of travelers using mobile apps, third-party travel websites and airline call centers. It manages 10 exabytes with Bigtable. Building a data solution to anticipate what customers might buy next builds on its existing data warehousing legacy, which includes Teradata, Oracle and IBM.

“We have pretty much everything you can think of in terms of data warehousing technologies and operational data storage. For all of these existing systems, we’re thinking about how we can get a feed in BigQuery first,” says Gasparovic.

“It was difficult to understand a traveler going from shopping to booking flights. Use this information to understand what these travelers are interested in, what their preferences are, and what type of product bundles they typically buy together. Something like that was difficult because that data existed in so many different systems in so many different forms,” ​​he adds.

Saber Labs maintains the traveler’s offer and creates an offer ID that is streamed into BigQuery along with the customer’s order. “It’s not just about putting it in the same place. It updates and modernizes the data model itself to be able to identify these things with a unique key,” he says.

Getting it all into one system, correlating it and understanding it as a whole was the first step. The next step was to create machine learning models that learn from the data. “That’s a really big deal for what we can offer our customers,” says Gasparovic.

But machine learning comes with a warning. It’s possible to play around endlessly with machine learning and not necessarily solve a real problem. Projects must start from practical problems that need to be solved, he says.

Saber Labs’ approach is to condense the training phase of an ML project by using a reinforcement learning technique to try something very quickly and see what value it brings. “You can think of it as a very fancy form of AB testing, where many different elements are tested at the same time,” says Gasparovic.

“The beauty of this is that we don’t have to train a lot in advance with models and try to play around with things to get accuracy. When we know that it actually provides a benefit, we can go back and do classic machine learning: build models and do a supervised learning process where you train a model on the data and then put that model into production and use his results to make predictions, but starting this experimentation has really been very helpful for us to know where to spend our time,” he says.

It has already deployed machine learning models to help its partner organization make better deals for travel customers.

“If we’re able to put that on a website and actually see the impact, that’s something customers expect, but it’s been very successful for hotels and airlines. That we made this possible for them is still something unique in the industry,” says Gasparović.

The Google Cloud solution builds on Saber’s existing data warehouses, but long-term plans include consolidating data onto one system, but Saber will likely run both worlds with the new world on top of the existing world for “some time,” says he.

Meanwhile, Gasparovic expects more wins from the existing line-up. “We’ve only just scratched the surface of what we could achieve with machine learning and experimentation,” says Gasparovic. ®

Leave a Comment