Hotels.com claims its decision to embrace the public cloud and big data technologies has helped convert more site visitors into holiday buyers over the past three years.
The Expedia-owned hotel bookings firm operates 89 websites in 68 countries, and started experimenting with the public cloud three years ago to give customers faster access to information about the holiday properties on its books.
As such, the cloud is used to power some of its sites’ smaller functions, such as the auto-suggest capabilities of its search boxes, while ensuring they have the overall capacity to cope with the seasonal peaks in demand for Hotels.com’s services.
“Demand starts to increase around summer, so that is when we would use the public infrastructure or expand our infrastructure,” said Bedos.
“Getting data close to customers is a key requirement. It’s what drives us to the public cloud, more so than the ability to scale,” he added.
The reason for this, he explained, is because web users typically have a low tolerance for slow-performing websites, so being able to respond quickly to online requests for information is essential.
The company’s online activities are also underpinned by a sizeable datacentre estate, although Bedos declined to confirm the number or location of these facilities, only that it doesn’t have one in every country it operates.
These sites are primarily used to carry out the heavy “number crunching” involved with the transactions that take place on the sites whenever someone books a holiday.
Big data insights drive customer conversions
Alongside its cloud experiments, the company also embarked on a move three years ago to step up its adoption of the NoSQL distributed database management system Cassandra to improve the overall resiliency of its systems.
“We started realising that if we wanted to move to the cloud and become more resilient, while dealing with the volume of data we receive, our traditional relational databases wouldn’t be able to cope. We started looking at NoSQL databases and selected Cassandra,” said Bedos.
“The reason we chose Cassandra is because it was very well-suited to our use cases and highly distributed, meaning data can be on multiple nodes. So, if some fail, that’s fine,” he said.
Thanks to some initial successes with the technology, Hotels.com has steadily started moving more workloads over to Cassandra to help it scale with greater ease and in a more resilient way, he added.
“Using distributed systems requires a shift in mentality compared with relational databases. The way the data store and schema is designed is very different, so we had to really get to know it. The use of it is now spreading to other parts of the organisation,” he said.
From a user point of view, Cassandra is used to feedback real-time information about the holiday properties people might be perusing on the site.
“If you go to a certain hotel, you might see a pop-up that says ‘five users are also looking at this particular hotel’ or you’ll see a pop-up that says ‘5,000 people have looked at this hotel in the past 24 hours’. I need to collect a lot of data and surface it to the users to perform this, so we use Cassandra for that,” said Bedos.
The company has also found the Hadoop big data analytics platform to be a good fit with Cassandra, he added. “Hadoop crunches a lot of data and gives a lot of insights that we can move back to the online world. This is stored in Cassandra, so it’s available the next time the customer needs it.”
For example, users need the data so they are provided with personalised insights about the properties they have previously viewed or suggestions for similar places that might be a good match.
According to Bedos, offering up these types of services can help users weigh up where to stay. It has also contributed to an increase in the number of visitors that go on to book hotel stays.
“We’re trying to get the best hotel relevant to users. If we achieve that using the power of big data and make it so compelling for users to do so, they will book through us. All these features have increased the conversion rate on the site and made us more relevant for consumers,” said Bedos.
“We’ve improved the performance of systems, reduced licensing costs, become more distributed and lowered the risk to the business. These could be considered intangible benefits, but they’re still important,” he added.
