3 ways that data science is delivering for Expedia

Across Expedia Group’s 20 brands, there is scope for sharing data learning points, infrastructure and algorithms. Expedia Partner Solutions’ director of data science explains how this is driving business value

If Expedia Group has one thing it has data, a lot of data. And with 150 data scientists sitting within the organisation, it takes it seriously. So does Nuno Castro who has 13 years experience in the field and is now leading the data science team at Expedia Partner Solutions (EPS). EPS is the B2B arm of Expedia and counts OTAs, major airlines, retail travel agencies, loyalty and membership organisations, among others, as its customers.

In the five and a half years that Castro has been in the role a lot has changed. “Back then the landscape was very different. We were more focused on building the infrastructure and gathering the data,” he explains.

Today, it is quite a different scene and Castro sounds almost evangelical. “Expedia Group is committed to the power of data science across the business. We are in a really good place. All the data is all there, the infrastructure is there. Now we can spend a lot more time doing more cutting-edge machine learning to develop and evolve APIs for our partners,” he says.

EPS say they help to unlock the power of Expedia Group for the travel partners in three ways:

  • Data learnings: Within the Expedia Group there are around 20 different brands with different use cases, and one of the advantages of this is the ability to share dozens of different data learning points. “Learnings from across Expedia Group are used to build the API that we give to our partners. We cannot send the data directly but we can send the intelligence gathered for consumer flight and hotel packages,” explains Castro, adding that this ability to share data learnings means that we can build cross-sell APIs. This means that an airline, for example, can get the exactly right hotel product in front its customers.
  • Infrastructure: By leveraging the group’s cloud infrastructure and accessing huge amounts of data across the business, Expedia is able to, in industry jargon, ‘spin up a cluster’ and train machine-learning models quickly and at scale. “We can reuse all that for those brands that don’t have their own independent infrastructure,” Castro explains.
  • Algorithms: The ability to share algorithms is another advantage. For example, hotel rankings are a major problem as 70% of bookings are made from the top 10 links. Expedia and hotels.com have been working on hotel rankings for five years to make it relevant to both corporate partners and travellers, and these learnings are fed through to EPS’ travel partners.

As one example of the work EPS is doing, Expedia Group adds 15,000 properties to its inventory every month. On Expedia.com and Hotels.com this can be immediately updated but for some partners loading a property is a complex process and still needs to be done manually. What this means is that there is a limit to the number of properties they can add at any one time. To solve the problem, EPS has applied deep learning to prioritise which properties are most relevant for each partner. For example, a low-cost airline focused on city breaks can ensure that the right hotels for their specific traveller base are prioritised in the list.

The technology has now been tested with some 40 partners with notable results. According to Castro, recently a partner applied EPS deep learning recommendations to a property and saw a +-40% year-on-year increase in conversions. “Many data scientists are tempted to work on deep learning because it is cool but to see impact on a business is very rewarding,” he says. “It is also great example of continued collaboration between data science and sales teams.”

Success doesn’t always come easily, however, with eight out of 10 ideas failing. Castro stresses that it’s important to iterate quickly and have good processes in place to test machine learning algorithms offline. The idea is to use historical data to simulate various scenarios and discard poorly performing ideas as quickly as possible. “So many things can go wrong…such as missing values in data, and the key is to detect those before they go live.”

Another challenge is keeping pace with the rapid changes afoot in this field. “Data scientists must always be learning, and they must have the ability to solve problems, big and complex problems that have never been tackled before and especially at this scale. It’s also important to be able to transfer learning and adapt solutions to an existing problem,” Castro adds.

On the plus side, however, while five years ago training resources were scarce, now there are many. The trick is to filter what is relevant. In-house, EPS does a combination of on the job training, and employees are also given a chance to study data science theory and fundamentals in company hours.

The crystal ball

In the past 20 years, the travel experience has remained largely unchanged. Broadly speaking, you enter check in and check out dates, and then click search; the experience is not personalised or relevant.

But machine learning is changing that and Castro believes that it is putting the agency feel back into travel. He strongly believes that systems can and will become more personal and conversational potentially using voice or chatbots. “Most of the time I don’t care where I go or exactly where I want to go. I want to be recommended something. This is where artificial intelligence and machine learning can play a role and Expedia, and its B2B brand, EPS, will be right at the centre of that innovation,” he says.

Ultimately it is about knowing the customer and a company like Expedia that has a good understanding of supply and demand believes it is in a strong position to innovate. “Expect to see some new innovation very soon,” Castro says.

Join us at EyeforTravel's Revenue and Optimisation and Marketing Summit (Nov 26-27) to hear more about revenue boosting business from top travel brands

Related Reads

comments powered by Disqus