Booking.com Destinaion recommendation & Product match

Background

I worked for several teams as a UX designer during my time at Booking.com. Here I will share some of my most successful and interesting projects.

Destination Recommendations

Destination Recommendations was a team dedicated to personalizing the experience of searching for a travel destination. Different areas of the Booking website were dedicated to the goal of helping users find where they should travel to next, including the index page, which contained a grid component offering suggestions based on past bookings and searches, and an MLT (Multi-Legged Trips) bar, which offered suggestions for trip enhancements and gap closers.

Customer Journey Mapping

The first step I took in order to assist with understanding the user and generate ideas in brainstorming sessions, was to create a mapping of the relevant stages in the costumer journey where destination recommendations have a real impact. This mapping was based on gathered research material from the Booking research team, and different teams who worked on adjacent projects, as well as personal insights from user testings. It included Jobs-to-be-done, information needed by the user in every stage, user pains, and product touch points.

Index Page Destination Recommendations

One of our first projects was the index destination recommendations grid, for which we wanted to introduce customization options in order to incearse user interatcion, and in the future gather usage data. Our success metrics for testing were:
The initial thought was to create filters for users to choose from in order to refine the offered suggestions.
After exploring some research and perfoming user testing for different designs, we continued with the following features, which had some advantages to offer in terms of UX and ease of implementation.
A combobox search field- Since offering filters implies multiple selection, a complexed faceting system would have been required in order to execute this kind of feature. Therefor, a first iteration feature was suggested in the form of a search combobox field, populated by existing user endorsements, allowing only one pick at a time.
A 2-step country picker - As user research revealed that focusing on a specific country is one of the most sought after features, a country picker was suggested on top of the endorsement filter. The design included a 2-step picker, where the continent is chosen first, leading to a list of relevant countries. This design was offered for its aesthetic appeal, as well as its value in reducing cognitive load.

Redesigning the MLT Bar

The MLT (Multi-Legged Trip) bar was another challange taken by the team. The goal here was to redesign the bar in way that would allow offering users more suggestions for enhancing their existing trips, and closing gaps (nights that are missing accommodation bookings).
The suggested design offered just that (4 suggestions instead of 1), plus a better indication of over-lapping bookings (several accommodations booked for the same dates), in order to encourage users to promptly cancel any unneeded bookings they might have.

Before

After

Product Match - A Personalized Match Score Feature for Accommodations

The Product Match team was given the task to create a peersonalized match score between users and accommodations. The initial plan was to use a collaborative filtering ML model to genertae the score, but the UX research made by myself, plus the joint work with the team's data scientist, brought forth a solution that reached that desired outcome in terms of its success metrics, and allowed for the feature to pass A/B testing and go live.

Designing a Personalized Match Score

The initial idea of a match score asked for a simple score indication on the UI, that would not take too much real-estate, and reflect the score generated by the ML model developed in favor of this feature. User testing pointed out a visibility issue due to the fact that most pages were already pretty cluttered, and so a short animation was created to make the score pop.
With that, the more challanging aspect of this feature had to do with the reliability of the score. Most of my users testings' subjects raised a problem with trusting the score without knowing what it's based on, and without being able to control it by customizing their preferences. These issues were discussed with the team's product manager in order to find a solution. A simple tooltip explaining what the score is based on was tested, but the explaination did not resonate well with users as they percieved the model to be too general, not taking into account their specific needs which change each time they travel.
After a sit-down with the team's data scientist, another solution was fashioned, in which another model would be added to the score calculation, taking into account the users' preferences based on previous trips. On top of that, a feedback loop mechanism was added in order to give the user more control over the score, enhancing the capabilities of the ML model in the future. A/B testing began with the original version showing only the score and tooltip, and did not meet the benchmark needed to go live. After adding what was referred to as the "explainability model", plus the feedback feature, A/B testing was passed successfully, with an increase in overall conversion, coupled with bookings made for accommodations with high scores.