Better bookings through analytics and agile - an interview with Hotels.com CTO Thierry Bedos

Bedos tells Computing how big data and flexible development help the booking service provide the best experience for its customers

They say a week is a long time in politics, but it's arguably more so in the world of technology where things move at a increasingly fast pace.

Two years ago, the Apple Watch was no more than a concept and driverless cars weren't much more than an experiment by Google, not the potentially lucrative market that all the automobile manufacturers are eyeing now.

Back in the summer of 2014, accommodation booking website Hotels.com was shifting towards a mobile-based, personalised service for its customers using big data and analytics. But in 2016, as CTO Thierry Bedos explains, the company is looking to go much deeper with its big data strategy.

"Since then we've continued making progress on building our platform. We've been consciously migrating off of SQL servers for all the use cases which aren't really required on a regular basis. So we've chosen Cassandra as the platform of choice," Bedos, who ranked 38th in Computing's IT Leaders 100 list 2015, says.

The project involved migrating a lot of data stores, so Hotels.com - part of the larger Expedia group - turned to DataStax, a big data software company specialising in Apache Cassandra, to help manage the transformation.

"They provided a really good solution", Bedos tells Computing.

"We've moved a number of applications into Cassandra. It's been very helpful to move to this stack, because as we start to move to the cloud, then it becomes a lot easier to scale."

Bedos goes on to describe how the DataStax deployment fits in with Hotels.com's overall big data transformation strategy.

"There are all of our online use cases, so for that we've been using DataStax. Not only for just basic storage, but we've also combined with other tools in use cases which allow us to really expand the online datasets and not degrade the online response time," he says.

However, it's Hadoop distributor Hortonworks that is used when it comes to interrogating offline data, across the whole of the Expedia group, not just Hotels.com.

"We're using a large cluster that is used across the back end, there's a lot of datasets that are common to all the different brands, so we're working with one platform on Hortonworks for that," says Bedos. "DataStax and Hortonworks work very well hand in hand."

All these datasets - including customer preferences - are combined and then analysed to provide the best recommendations for customers when booking accommodation.

"If you go to the booking page and look at your hotel, then we'll use information on hotels stayed at by people with similar interests and behaviours by using the power of big data to make recommendations," Bedos explains.

That aspect of personalisation is integral to providing a reliable, quick and targeted service to Hotels.com customers, Bedos adds.

"On a site like ours, there are hundreds of thousands of properties, but it's not the number of properties which is important, but how relevant can we be as quickly as possible? We owe it to our customers to help them make this choice about properties," he says.

"So it's absolutely about personalisation and trying to make a tailored experience and that's what I'm interested in."

[Please turn to page 2]

Better bookings through analytics and agile - an interview with Hotels.com CTO Thierry Bedos

Bedos tells Computing how big data and flexible development help the booking service provide the best experience for its customers

Bedos says his team are constantly tweaking the Hotels.com website to further improve the user experience.

"We do a lot of feature testing; we do over 100 tests every quarter on the site, making sure what works and what doesn't work, and if we find a feature doesn't work we try to fix it very quickly," he says.

‘Pushing the limit'

"We're accelerating the pace we're able to deploy features for customers. We've been agile for a long time, shrinking the window between two releases, and now we're getting to the point where three times a day we can have a feature live after being started only four hours previously," says Bedos, adding that his developers are "really trying to push the limit".

"This gives amazing tools to the product team," he says. "The quick release mechanism means it's really easy for our engineers to deliver features incrementally. The risk of introducing a bad feature is lower and the agility goes up. It really helps."

Bedos and his team apply the same mentality to security.

"All our engineers go through security training to understand all the threats and we do a whole range of activity around security to make sure the data the customers give us is 100 per cent secure," he says.

That, Bedos tells Computing, is something that any organisation can't afford to take for granted, due to the damage a breach can cause, as has been seen with incidents ranging from TalkTalk to Ashley Madison. "Cyber attacks have been in the news and it can really make or break a brand, so we take it extremely carefully," he says.

To further safeguard customer privacy, Bedos is keen to ensure that only data that absolutely needs to be stored is stored.

"We don't keep personal information if it's not required; we're just interested in understanding what makes a choice relevant to our customer.

"If we know where you've stayed, we can get insights about what you like. That's it, the rest we don't need to keep, because that's private to you."

[Please turn to page 3]

Better bookings through analytics and agile - an interview with Hotels.com CTO Thierry Bedos

Bedos tells Computing how big data and flexible development help the booking service provide the best experience for its customers

Nonetheless, that data is worthless to Hotels.com if the company isn't able to mine it in order to learn about customer preferences. But while the website has been able to hire data scientists to do this, finding people with the right mix of skills isn't easy, Bedos says.

"What you need is somebody who has a data science background, but also someone who is a good engineer, which is very difficult," he says.

"If you concentrate on the data science, you get a lot of academics who'll be amazing at handling data, but don't have an engineering background.Then vice versa, you can find engineers with an interest in data science, but who might not have yet experienced anything around algorithms and how to manipulate data," says Bedos.

The answer, at least for Hotels.com, is to train employees to become data scientists by encouraging staff from different fields to work together and learn new skills.

Cross-pollination

"What we do is either take someone with a data science background and pair them with someone with more of an engineering background for cross-pollination. Or train engineers and get them to brush up their skills on data science," says Bedos.

"It's one of those when you're at the forefront of what's happening in the market, you have to be able to invest in building the skills of your people, because you're never going to be able to find somebody with the right skills," he says.

It isn't only the Hotels.com data science team that is expanding; Bedos is keen to grow his team in every area as the website becomes more popular, especially when it comes to mobile traffic.

"We've continued significantly growing the team building these features because we're moving into a world where half of our traffic is now mobile and we have about a quarter of our bookings via that platform," he explains.

He says the firm has moved to a "mobile first" design - ensuring that the site is responsive by having the same coding behind everything. Every feature is designed for mobile devices first and is then extended to the desktop version.

"It used to be the other way round but now mobile is so important in our industry that we're doing mobile first," says Bedos.