Check out the on-demand sessions from the Low-Code/No-Code Summit to learn how to successfully innovate and achieve efficiency by upskilling and scaling citizen developers. Watch now.
Hopper, a travel app that tells you the right time to book your flight to get the lowest price, says it has leveraged AI to drive significant growth over the last two years.
Already, its AI powered notifications drive a quarter of its overall sales, and the company says that portion will grow. More significantly, those AI-driven sales are to customers who were originally looking at traveling to a different destination or on a different date, but who decided to change their mind upon Hopper’s recommendation.
“What we found is that users converted 2.5 times better on those recommendations than for the trips for which they had actually searched,” said Dakota Smith, Hopper’s head of growth and business. For example, a customer in New York looking to buy a ticket to Hawaii might be lured to buy a significantly less expensive ticket to another desirable beach location in the Caribbean or Miami, Smith said in an interview with VentureBeat. “Reaching out to the right user, at the right time, with the right recommendation is our key to success.”
The app’s growing reputation, and catchy features like “secret fares,” has created strong download momentum. According to Smith, Hopper just achieved 30 million in total app installs since its 2015 launch. That’s up 50 percent from February, when the company first announced it had passed 20 million installs. In April, Hopper shared that AI drives 25 percent of ticket sales, a number that is apparently growing (although Smith refused to share an updated figure).
Intelligent Security Summit
Learn the critical role of AI & ML in cybersecurity and industry specific case studies on December 8. Register for your free pass today.
The company is an upstart in a flight ticket industry predicted to be $800 billion by 2020, faced with numerous competitors. The only way it was able to differentiate itself was through data. It has a relatively small team of 10 data scientists (out of 170 employees total) driving its prediction and notification engine, according to Smith.
Hopper engineered the app’s prediction algorithm to track 1 trillion price points per month from around the web. It stores five years of historical data and many trillions of archived prices. It then tracks feedback from user behavior to offer the most accurate predictions. These predictions drive the app’s notifications, says Smith, which in turn drive 90 percent of sales.
On top of that core platform, Hopper began testing a new AI-driven recommendation algorithm, which sent users notifications about cheaper deals to alternate origins, destinations, months, or weekends. The conversion data further strengthened the algorithm, making future recommendations even more relevant.
Soon, Hopper was selling $1 million a week of airfare for flights that users had never asked for.
The company said it had $15 million in revenue last year. At an average booking price of $500 (customers often book more than one ticket at a time), it’s generating about $1.5 million to $2 million in sales every day. In addition to a $5 fee per booking consumer charge, Hopper makes money from airline commissions, incentives from the industry’s Global Distribution System (GDS) companies, and hotel bookings.
My interview with Smith is embedded below. It’s one a series of articles we’re publishing in the run-up to our Transform event on August 21-22 in San Francisco, about how brands are using AI to drive growth. Our motto is: “You can do it too!” Hopper’s Smith will be coming to Transform and talking in more detail about how to build a consumer app with AI.
Here are my five takeaways from the interview:
1. Bite the bullet and collect user data from the start. Most websites deal with a “cold-start” problem, since they don’t require a user to register or log in. In travel, for example, users might search for flights from New York to Honolulu, but it’s one search, and the user remains anonymous to that website. They might leave, come back three weeks later, and do another search. That website will have a difficult time correlating those two searches. Hopper decided to go mobile only, and since a user has to log into its app and Hopper knows every device ID, it collects a huge amount of personalized information. About 70 percent of Hopper’s users sign up for push notifications and alerts for trips, allowing Hopper to take otherwise small correlations and feed them into its algorithm.
2. Don’t be afraid to start small. From Hopper’s Smith: Users may only start with one search, but Hopper will push dozens of notifications about it, and “start recommending alternatives and things to users and consumers and training that algorithm from there. We realized somewhat early on that it’s very hard to get a user to put their entire consideration set in a search box. They’ll give you one destination, maybe one pair of dates, but it’s very difficult to get them to sit there for an hour to tell you every place they’re considering traveling to. But we knew we had this way to communicate with them over several months via the push notifications. It started off really slow. We took baby steps into the AI, into the machine learning. Looking at users who were watching flights from places like New York to Honolulu, which just objectively, we know that’s a 10-hour flight. We know that’s probably about $1,000. We know it’s a beach destination. We started saying things like, ‘You’re not buying this flight; have you thought about flying to Miami? That’s about $200 and two hours away.’ We sold more flights to Miami than to Honolulu when we did that.”
3. It pays to be patient. From Smith: “To think that we just got a user to install our app from a digital marketing campaign, and all we ask them to do is watch the trip and sign up for push notifications.” More than half the time, Hopper tells the user to wait for a better price. “We’re talking about month three, month six, users started booking the trips they watch with us. … It took six months for us, to trusting our own product,” he said. It was a painfully slow start, and conversion was low, but it was worth it.
4. If what you have doesn’t appear unique, start somewhere. Hopper started purchasing data that was otherwise available publicly from the industry’s GDS. But Hopper started saving it for five years to track patterns, and then on top of that reengaged users through push notifications to start getting superior data. Smith’s advice for another company: Look for where “you have enough interactions with enough users to have an algorithm that becomes sufficiently intelligent to [impact your] business.”
5. Personalize engagement, and learn from there. Hopper can track whether you prefer more or less communication, and will adjust notifications accordingly. Also, according to Smith: “Some users give you lots of indications that they’re flexible by watching and searching for multiple things. Other users don’t, and you’re taking these small correlations you’ve seen from other flexible users and sending push notifications and applying that to these users who, from all indications, are not flexible.”
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.