October 2017, Las Vegas (USA)
How machine learning boosts personalisation in travel
Data is everywhere but how do you create value for individual users? It isn’t easy but in this case study we hear how data science can deliver more personal results
Consider this common hypothetical scenario. A user opens a browser and taps in ‘cheap flights from Boston to London’. Needless to say, what appears first in search are the ten blue links from Google. After some surfing, the user is then drawn to a metasearch engine like Skyscanner, Kayak or Momondo, where flights can be filtered by date and price. Here the cheapest deal is, let’s say, from Norwegian Airlines, and as the meta aggregates offers from various providers, the user again opens multiple links to deals from online travel agencies (OTAs). Time spent on each site is never much longer than around 30 seconds.
At one OTA where tech firm AltexSoft has been gathering user interaction data for a year, typically the user opens the flight details, hastily closes a pop-up window without reading the contents, and continues searching. Then within two days, the user returns to the OTA to close the top deal.
AltexSoft's data scientists call this type of ticket surfer a ‘economy buyer’. Accounting for around 40-50% of airfare searches, the ‘economy buyer’ looks for the most affordable deals, they don’t spend too much time exploring flight details and aren’t bothered about long layovers or seating.
Back in 2012, Amadeus published a research called Who Travels with You. The study outlined five main segments of travellers: digital natives, young adults, family travelers, empty nesters, and golden oldies. While digital natives and young adults combined are only 22% of the entire travel market, they are the most active web users who prefer booking flights separately from accommodation and leisure activities. But there’s a tangible difference in behaviour between digital natives and young adults. For instance, digital natives usually belong to the 'economy buyer' group, while young adults, aged 25-44, with no children, can afford to be more selective in choosing travel services.
The data science team from AltexSoft has proved that pattern by actively gathering user interaction data working with their OTA clients. The team tracks practically every little interaction: destination searches, clicks, dates, hover moves, and even the ways users examine travel service providers. While the dataset may not be as large as those collected by the big players in the travel market, it already lets engineers segment user types and launch personalisation features.
But how? To collect all records linked to user behaviour, the team devised a user behavior tracking (UBT) engine. This consolidates data, thus allowing data scientists to build prediction models around it. Using machine learning trained with data, the engine is able to predict the likelihood of a conversion after just a couple of clicks. Given that a high percentage of visitors are so-called ‘economy buyers’, new users are bundled into this main category by default. However, once they begin checking amenities, exploring layovers, legs, and airline details, the model gains confidence about whether the user is likely to buy and uncovers their more personal needs.
By segmenting all users into five likely ‘conversion’ groups, OTAs are able to optimise ad retargeting campaigns, and increase bids for the most promising users
The spectrum of this artificial intelligence (AI) application is broad. According to Alexander Konduforov, head of data science at AltexSoft "we can easily, for example, distinguish between business and leisure travellers". Now, it’s a matter of delivering differentiated value to these two groups of visitors.
Going back to the ‘how’, AltexSoft started with the low-hanging fruit, which Konduforov says is perhaps the smartest approach when you embark on something as complex as machine learning. In fact, EyeforTravel’s research in the run up to the Las Vegas event shows that to date this has been common practice. As Stuart Greif, a Day 2 keynoter a senior travel executive at Microsoft, pointed out in a recent interview: “The near-term promise is focused on more discreet areas and use cases,” because it’s still incredibly complicated to book travel”. No doubt we’ll be hearing more about that tomorrow!
Low-hanging fruit may be the focus, but by segmenting all users into five likely ‘conversion’ groups, OTAs are able to optimise ad retargeting campaigns, and increase bids for the most promising users.
Value in email and value buyers
Another finding is that value can also be created by email personalisation (More from EyeforTravel How email still has an edge). If an ‘economy buyer’ provides an email address to be notified when the destination price is the lowest, the algorithm will pick three flight offerings matching the cost preference and attach them to the email.
While economy buyers usually account for almost half of an OTA's audience, the other half - ‘value buyers’ - are more sophisticated in their flight preferences. For these buyers, suggesting the lowest price doesn’t cut it. According to Konduforov, this is when real personalisation must kick in, with the system accounting for things the person values most. “Some don’t like long layovers, some are picky about seating and meal options, and some are fans of particular airlines,” he says.
With this in mind, the best option is for the team to deliver a customised flight search engine that will filter flight alternatives that considers both cost and value priorities for everyone typing in their destinations.
The two main challenges today are data related – how to store it, and how to process it
While 79% of business executives surveyed by Forrester believe that personalisation will help them achieve marketing and customer experience goals, the practice is still an investment in data science and the underlying technology itself.
“The two main challenges we see today are data related. As we collect more data, we have to figure out how to efficiently store and further process it,” Konduforov explains.
Other problems include:
A lack of individual user data. Although the dataset has enough records to build accurate predictions about incoming visitors, the machine still needs users to stay on a website long enough. Only then, can it begin to accurately identify groups and tailor the offering. Undoubtedly, long-term users can provide better personalisation opportunities.
Cookie blocking. Saving cookies allows the algorithm to recognise visitors who have been visited before, and that does simplify things. But people tend to block cookies. Even if a user had purchased tickets before, and can be qualified as promising during the second or third visit, one browser cleanup rolls this person back to an unknown state. Now the system is dealing with a clean slate.
Most users aren’t registered. A registered customer who regularly uses the account on all devices allows the system to provide the best value options based on long-term and consistent data is the best-case scenario. But that’s not the reality. Just a fraction of users are registered. Some login only on desktops, and most don’t have accounts at all.
There are some ways to partially sidestep this problem. Although you may not have the behaviour data, you can make assumptions about users solely relying on metadata. For example, referral websites that users come from, and what devices and browsers they are using, can give some insight. For instance, users coming from Skyscanner are more likely to buy than those coming directly from Google. But there is still a tricky balance to be struck between user privacy and data collection.
AltexSoft is a sponsor of EyeforTravel North America 2017, which kicks off today in Las Vegas. Here CEO Oleksandr Medovoi and Head of Data Science Alexander Konduforov will be sharing more insights into the business opportunities for deeper personalisation that come from harnessing data.