But flight deals are more nuanced than price alone

“Saving money on flights allow you to stay at a nicer hotel or splurge somewhere else on your trip.” –Interviewee

Travelers are looking to maximize value-for-money. Despite the importance to our users, Expedia was often surfacing cheap flights on budget airlines with multiple long layovers. But based on user research, comfort matters in addition to price when evaluating flights.

Can rule-based logic derive flight deals?

Good flight deal = Minimal discomfort when flying + Good price

Making sure flights meet the minimum standard for quality is not complicated as we can define rules, but determining what constitutes a good price is a challenge. While we could analyze historical price data, organic search data is relatively sparse and programmatically collecting price information can be costly at scale. Moreover, flight prices typically increase as the departure date approaches, meaning recent averages may not accurately represent today’s typical prices.

So machine learning is actually better-suited to solve this than rule-based logic.

Thinking holistically, end-to-end

Thinking about our flight ecosystem, I mapped how users can come across our flight deals. There are many opportunities, whether with our outbound communications, top of the shopping funnel, or further down the path. For the first version, I decided to focus on the email to checkout segment as email was the best channel where we were reaching our users, giving us the traffic we would need to meaningfully test a new product.

Prototyping with real ML data

With the inputs from the studies, the machine learning team developed a price context model. Armed with an understanding of the model, it helped inform and drive my design decisions. I explored different ways to display deals in our flight shopping path, looking into badging and strike-through pricing on our flight cards. I landed on badging with the percentage savings spelt out, making the value-for-money easy to comprehend. The percentage also plays a role with managing the model’s false positives.

Impact: +17% conversion rate

I made sure to carry the badging consistently throughout the flight shopping path, so users can build confidence that the deal they selected is the one they're booking. We shipped the email to flight checkout flow, providing price transparency and supporting users’ confidence in their savings, resulting in an increase of +11% click-thru rate and a +17% conversion rate. The win also laid the foundation for other parts of the ecosystem, now also being used by Flight search.