The Case for Personalized Travel Discovery

Andrei Faji
Director of Marketing and Customer Experience
Photo of a hand holding a globe

It’s time for online travel discovery to evolve, and we’re doing it.

Over 100 million users login to their Spotify account every day to listen to Discover Weekly. It’s a custom playlist of artists and tracks they’ve never heard before, personalized to their taste, and curated with such thoughtfulness and hidden gems that even music snobs (like myself) pine over the selection.

Discover Weekly on Spotify

Think of Discover Weekly as your personal connoisseur of music, saving you the hours of time you’d need to spend searching for new music and contextual to the mood you’re in — and it’s entirely machine driven.

Spotify has created something remarkable for helping people easily discover music that they probably wouldn’t have found on their own. We’re on a mission to do the same for travel.

Travel is about going somewhere, for a wide range of reasons, in order to create an experience.

There are few things more complicated than recommending an experience to someone. Unless you really know that person (that is, you’ve traveled with them before and you’re acutely aware of their preferences, interests, and tolerance level to handle unknown variables) you probably won’t be able to give them a great recommendation. In fact, even if you know them, you can still make a poor recommendation because the bias you have from a positive or negative experience can be persuasive.

(Getty Images/Tom Kelley Archive)

At the same time, when you have an incredible trip experience it can be life changing. You’ll share photos and memories. You’ll talk about it for years. You’ll even go back to that experience frequently in your day dreams.

The power to help recommend, or serendipitously discover, relevant experiences for travelers is a challenge. Other than seasoned travelers, no one is doing it well.

The (very) brief overview of online travel…

For the sake of context, let’s take a quick look at what’s changed in travel over the last 20 years or so.

In 1996, a subsidiary of Sabre Corporation launched a website that allowed travelers to search and book hotels, flights, and cars. The website was called Travelocity and it was one the first sites to offer DIY travelers the ability to search and compare travel options on a single webpage and book trips online rather than over the phone.

:align="left" circa 1996

The first wave of dot coms in the late 90s and early 2000s made way for even more travel websites. Some were niche sites while others offered a wide range of information or services: online travel agencies, travel blogs, forums, and travel review websites like TripAdvisor. These sites not only provided relevant information for travel planning, they also created new means for travelers to connect — people were sharing photos and stories of the places they stayed, what they enjoyed or hated, and their recommendations to other travelers.

By the mid 2000s social networks created platforms that enabled this type of digital “conversation” between travelers to happen faster and more regularly. Travelers began sharing photos and reviews during their trips rather than after it.

Travel brands started realizing that these social platforms were becoming permanent channels in culture and the people who were using them were talking about them (the brands). A good customer experience could pay dividends, and a negative one could cost a fortune, as these social influencers would share their stories with the world.

This created a new world for travel brands to think about. One where the customer, rather than the brand, is at the center.

The world of online travel has become a highly competitive space where brands try to capture your attention (and wallet) with searching, booking, documenting, recommending, sharing, and comparing.

It’s also a place where customers and brands are much closer to each other, which has fostered brand awareness and loyalty across various stages in the traveler’s online shopping journey.

These opportunities have also created much higher demand by travelers for customer-first travel companies. For example: do you find it acceptable today for a brand to take 48 hours to respond to your complaint? How do you feel about hotels that still charge for in-room wifi? What about airlines that don’t let you download a mobile boarding pass?

Innovation in technology has amplified our expectations. This is why companies, across all industries, are placing customer experience at the center of their business strategy. Whether it’s Marketing, Engineering, or Operations, each department is now beginning to measure customer-centric KPI’s.

What that actually means for brands is still being defined, but what we know is that 92% of travelers expect their trip experience to be personalized.

Making online travel discovery more personal

Through Discover Weekly, Spotify makes users feel special and catered.

As travelers, we want to feel the same. When we receive a promotion email from a brand or when we book with a hotel we’re loyal to, we want to feel like we’re known. We want travel discovery to feel inspirational and enchanting. We want our wanderlust to be fed.

This is why our company exists.

When you land on any travel site, you can discover places based on when, where, and how long you’ll be traveling. The most relevant information, which is your why (the intent of your trip), is missing.

It’s your responsibility to navigate through endless offers, reviews, and other variables to find the best place to stay. It requires you to weigh every option against a mental checklist of contextually specific needs and preferences.

For the past three years, we’ve been working to design technology that enhances the online discovery experience, to make it easier for brands and travelers to connect in a more engaging way, by offering contextual recommendations for where to go and where to stay.

Do you require a pet-friendly hotel? How much extra is that fee?

How about a resort that’s walking distance to the ocean and a room with great view to surprise your spouse for your anniversary?

What about a place that’s perfect for a surf trip reunion with your college friends and their husbands and young children?

We believe travel companies, specifically in the online shopping journey, should answer these questions like an expert travel agent, making trip inspiration and discovery a valuable competitive differentiator.

If they want to earn your loyalty, these brands are responsible for ensuring you have a great time when you’re at their property. They’re incentivized to ensure a travelers experience is exceptional when discovering, booking, in their trip, and even afterwards.

Using artificial intelligence to develop a platform that gives personalized travel recommendations

So, how are we making this possible? Here’s a high-level overview of what we’ve created to help travel brands provide this experience. In future posts, we’ll get into more of the technical mechanics of these, including how we use neural network models for aesthetic scoring and content classification, as well as how we understand and organize our various data sources.

Recommendation Engine: our travel graph and machine learning models

We know that travel language and travel product data can be specific to each brand and their customers.

Since most AI solutions are too general to be useful for this type of granular complexity, we designed a knowledge graph built on travel data, which for lack of better words, is the brain of our recommendation engine.

We process unstructured and structured data through our graph (data from our customers as well as public and third-party licensed sources) such as reviews, descriptions, images, points of interest, destinations, hotels, vacation rentals, attractions, and more.

For training and modeling, our travel graph extracts valuable insights from that data, which we call concepts, and then scores and weighs these concepts against each other.

For example, if we process an image of a couple admiring a beach sunset, the graph not only looks at its metadata, it derives one or several concepts such as “romantic” and scores the image based on how closely it relates to a user searching for a romantic experience such as a honeymoon or anniversary.

Scoring, weighing, and connecting concepts in our travel graph

In order for AI to be “intelligent” it needs large volumes of data to continually learn from. We’re using machine learning models to train and improve it. Each user session is tracked so the data we get on user behavior and explicit data on how they search, is used to train the graph to get smarter.

Converse: dialoguing and natural language to extract user inputs and capture travel intent

We know that basic form fields and keywords only allow travelers to express so much about an experience they want to have. At the same time, people are becoming more accustomed (even preferential in some cases) to interact on messaging platforms as well as with chatbots. We’ll get into this more in a future post, but for now I’d recommend starting here if you’re interested in learning more about conversational commerce.

In order to capture the ‘what’ and ‘why’ of the traveler we designed a chatbot using Natural Language Processing (NLP) — which is the ability for a computer to understand human speech — for parsing intent.

NLP enables us to train our graph with any text about travel and figure out what people are saying about destinations and specific places. Entities need to be extracted, semantics need to be understood within context, and then used to identify intents.

Discerning intent for travel queries looks different than general queries

For users, it enables them to search for destinations and places using natural language rather than keywords. A question or statement is submitted by the user and then parsed using NLP to understand the intent, which is categorized as either a destination or product search.

From this intent, the chatbot prompts the user with follow up questions and ultimately returns relevant search results based on the criteria it gathers. Of course, the key is ensuring that the conversation design of the bot is tailored towards extracting as much useful information from the user as possible, which is something that we’re still working through and getting feedback on.

We’re using reflexive dialoging (who, what, where , when, and why with an emphasis on the what and why) rather than decision trees to extract as much information about the traveler’s trip criteria. In the case with one of our partners, The Leading Hotels of the World, natural language search was able to improve their users’ search results quality by a staggering 81%.

Natural language search for Leading Hotels of the World

Discover: matching destinations with experiences and attributes

A seasoned travel agent will intuitively know that if you say you want to go somewhere to lounge on the beach in January, not all beaches are equal for lounging, and not all beaches are lounge-able (definitely not a word) in January.

We think about this as destination intelligence — specifically, what makes a certain destination unique for a particular experience, and what important attributes (such as seasons) will be most helpful to a traveler.

In Discover, we currently have about 50,000 destinations and 3,500,000 specific points of interests that we’ve processed through our graph. When a traveler uses natural language to describe their dream trip experience, our recommendation engine is able to parse through their intent and match it with the concepts our graph derived from destinations and points of interest.

Discover — our destination intelligence

Engage: selecting and sorting travel recommendations with contextually relevant information

The limitation with most personalized result sets is users aren’t always sure why they’re getting a particular recommendation. It’s hard to trust an answer when you have no context why it’s the right answer.

Psychologically, we know people make purchase decisions based on their feelings, outside sources of influence, and experience. As such, we’ve thought through these behavioral considerations for how to display recommendations for the traveler.

We begin with the concepts and intent we’ve extracted from the dialogue. Our recommendation engine selects and sorts the best available options along with the most relevant information attached to that option such as images, descriptions, and reviews. This is called dynamic merchandising. It’s basically the same experience as walking into a retail store and having a sales associate help you find the apparel that’s most relevant to you. You want to try it on, in your size and preferred color, before you make a decision to buy it.

When you’re online, especially when searching for the best vacation rental for a family trip, there is no such thing as “trying it.” Instead, there are other reviews and images that resonate with what you care about. We’re focused on showing the traveler that information first.

When a traveler is searching for a hotel on a brand’s website for a two-day business trip, the information that contextually matters to her is different than a user searching for a hotel that’s best for a weekend getaway with his family.

We’ve seen significant improvements in engagement and conversion with Engage. By making contextually relevant hotel recommendations and showing the best images, one of our partners saw a 10% lift in their booking conversion. In a time where personalization is an obsession, this seems like a no-brainer, but many travel brands today still struggle with offering a generic, one-size-fits-all approach to search results.

This covers the basics of our recommendation engine and the products we’ve designed to elevate the discovery experience for travelers. Consider this your primer on what we do. :)

Just yesterday, in a product meeting, we looked at our roadmap and I said, “We’re only 1% finished.” That’s the sobering truth.

This is the challenging part of our work but we’re seeing it bring value for our partners and their users. That’s what drives us.

What we’ve learned over the last couple years has taught us a great amount about traveler behavior (what travelers resonate with most), market readiness (chatbots, personalization, and deep learning), and various pain points of the travel brands (legacy IT systems, lack of prioritization around the customer experience, and structure inefficiencies that make it hard to keep up with new technology).

The good news, however, is that the technology has evolved exponentially to drive measurable value and the mentality of executives is finally changing.

This leaves us with vast unchartered territory to create something meaningful for the travel world, and we couldn’t be more excited to do it.

Related Posts

  Photo of a business man inside a hotel room

AI Strategy for Hotels goes Beyond Alexa

Proposals for new technologies are hitting hoteliers’ desks at breakneck speed: mobile kiosks, mobile keys and concierges, artificial intelligence (AI), virtual reality, even robotic staff. It’s a lot to digest when you’re tasked with figuring out wh…

The Intersection of AI and Travel
  Screenshot of traveling website

Chatbot Ads: What They Are & Why Travel Companies Should Care

In 2015, Expedia reported that travelers visited 38 websites before booking a trip. In 2016, Google said that travelers were researching in (sometimes hundreds of) micro-moments, short snippets of time spent on a variety of mediums. In 2017, a Sojern…

The Intersection of AI and Travel