Travel investor Brad Gerstner recently said that the vast majority of value created from the last ten years has been wiring up the world’s accommodation and distributing it online.
And for the next ten years, he told the PhoCusWright Conference, it’s going to be proactively recommending and automating the booking process for consumers.
Whether or not this is true is up for debate, but there is no denying the power of data to create a more fine-tuned travel booking process.
Travel is a great use case for recommendations and while we have ambitious goals, we must build products that complement how people shop for travel.
And we must do so through new inputs, not just a more refined view into the current data. Let's face it, travel search has to improve.
From filtering to data
Travel recommendations have been around for a while, we just call it by a different name: filtering.
Every time we use an OTA, metasearch or supplier site to look for travel, we immediately adjust the filters to our preferences to narrow down results.
While some may find this annoying, it takes all of three to five seconds and is extremely effective at helping us find what we want.
Recommendations of the future have to be more than pre-filtered results. three to five seconds of our life is not painful enough to risk giving us wrong results.
All data is not equal
The challenge to move beyond filtering is in the data sets we use. Filtering by its very nature is using any easily identifiable attribute ensconced in the GDS or APIs that are feeding the inventory.
Pool? Check. WiFi? Check. 8:15am departure? Check.
This is basic level data and is our starting point. To give us utility beyond this, we need much deeper data.
Things like:
- Good nightlife as long as you have a room on the 5th floor or above.
- Pool is great for kids, but awful for quietly reading.
- Flat bed seat is really one of those angled beds that make you slide down every 5 minutes.
We need advanced algorithms to find this data, make sense of it, and then productize it.
This involves wading through millions of reviews, data attribute sets, booking history and preferences, and new sets of data we don’t yet even know about, and deriving tangible, useful tags that can be used not just for better filtering, but proactive recommendations.
Preferences today not equal to preferences tomorrow
Finally, we need to find a way to make these attributes accurate and relevant indicators of what we’re looking for. It’s one thing for a very set use case such as business travel to a major city.
It’s another when you’re trying to do it for vacation travel, which can have it’s own set of varying preferences.
In the first five minutes on the phone with a travel agent, you impart a summary of what you’re looking for. From there, he/she can help you find what’s best.
But yet recommendation engines tend to think that by hooking into your Twitter or Facebook account, they can magically predict the best hotel for you.
Nine out of ten times, they’d be wrong.
The path forward
A new crop of recommendation engines are entering the market and starting to get some of these things right.
Just last month at the Travel Innovation Summit, we saw Ostrovok, Olset (TLabs here) and TravelShark present their takes on personalization in travel.
Each of them is taking some novel dataset and inserting it into a user flow where it adds value by reducing options, shortening the path from search to transaction.
Other startups are also focused on generating data in specific areas, such as RouteHappy (TLabs here). By collecting and making sense of flight attributes, they’ve created a new dataset to be used in the booking process.
This kind of complex yet normalized data is perfect for enabling powerful recommendation applications and is increasingly important as suppliers add more ancillaries and thus complexity to the booking process.
Beyond getting the data right, these companies have a challenge of making the personalization accurate and effective, even as the use cases go up and complexity increases exponentially.
Olset does this by tapping into booking history and networks, and by asking users. TravelShark relates properties to each other so you can discover by finding hotels similar to ones you already like.
Comparisons
Travel should look to other industries for advice on how to do this well. Netflix uses collaborative filtering algorithms – helping you discover things that other people with similar tastes have liked also.
In music, Pandora, Spotify, and now iTunes Radio all build playlists based on preferences and musicians sharing similar attributes.
Although travel is different – you’re paying much larger sums (and perhaps spending your valuable vacation time) on a big choice -- the lessons still apply.
Just as we would never expect Pandora to know if you’re in a Modest Mouse mood or a Katy Perry mood, we can’t expect travel sites to automatically know what’s best.
We can, however, hope they make discovery easier, decision-making better, and the process a whole lot more fun.
That’s a worthwhile goal that will increase conversion and justify investment in these start-ups. And it just might be the biggest frontier in online travel for the next decade.
NB: Disclosure - author is an advisor to Olset.
NB2:Travel discovery image via Shutterstock.