What if hotel search felt made for you?

Personalizing Search Experiences with AI-Driven User Segmentation

DESCRIPTION

A mobile interface redesign for hotel search results that leverages AI-powered user segmentation to deliver personalized recommendations and reduce decision fatigue for travelers.

CONTEXT

As part of a 6-week capstone at tiket.com, our team explored how design could better surface AI-powered recommendations. We reimagined the hotel search results page to adapt dynamically to user spending behavior, booking patterns, and search intent, thus creating a more relevant and confidence-boosting experience.

MY ROLE

UX Researcher

Product Designer

TOOLS

Figma

Confluence

TIMELINE

6 Weeks [Summer 2023]

ℹ️ Problem

Our capstone team was assigned to come up with an explorative solution to the question:

How can the mobile interface be improved to maximize the capabilities of the existing machine learning / artificial intelligence systems at tiket.com?

🎯 Our Goals

BROADER GOALS

(1) Minimize decision fatigue and choice overload & (2) Minimize user dropoff after their initial product choice

CONTEXTUAL (NARROWED-DOWN) GOALS

(1) Improve user experience & (2) Retain user engagement

💡 Background 

Users are constantly looking for personalized content. (Bozhuk et al., 2020)

(This research studied the current conditions of the tourism industry from consumers’ behaviors.)

1. Understanding the capabilities of ML/AI in the OTA industry.

2. Understanding the existing ML/AI systems @ tiket.com

Out of the existing systems at tiket.com, we wanted to pick a system that hadn’t been translated in a design perspective for users to visually see and appreciate.

⚠️ Research

1. Understanding current problems @ tiket.com

According to an analysis on User Retention and Drop-off Rates from the Search Results Page (SRP) to Purchase Form for Hotels:

The biggest drop-off rate of users occured in the Room Listing Page.

2. Breaking down this information into user insights and translating them to tangible issues

Relevance Issue

Some users searched the words “child” or “family” on the search form before they dropped-off

= They likely didn’t find places appropriate for a family stay (quantity and qualities) or weren’t able to find the features they were looking for in the description page

Benefit Issue

Some users second-guess the benefits they’re receiving from the hotels they think about choosing and have the mindset that they might possibly find a better deal

= They feel unsure of committing to the wrong hotel decision, ultimately causing them to just drop-off overall.

Uncertainty Issue

Some users are uncertain about what they’re looking for and leave the Room Listing Page to change travel destinations on the Search Results Page

= They’re browsing with no specific intent to book a stay

3. The ML/AI system we chose: Cygnus

Cygnus - Hotel User Segmentation

  • Categorizes users from multiple factors such as (1) click behavior, (2) transactions, (3) geolocation, and (4) their hotel booking process

  • To ensure accuracy and relevancy, the user segmentation models are run monthly, updating user clusters accordingly.

Although not directly related to user-centered recommendations, we decided that this aspect of understanding our users’ behavior and personality was an area of untapped potential for developing an impressive recommendation system.

ℹ️ Refined Problem Statement

How can the mobile interface be designed to transparently surface and align AI-driven recommendations with user intent, ultimately building trust and optimizing recommendations?

🔍 A Look into Cygnus User Segments

Economy Spender (Cluster 0)

  1. Limited transaction frequency

  2. Budget-conscious

  3. Indifferent to Promotions

  4. Indifferent to Credit Cards

  5. Preference for Standard Hotels

  6. Use of Budget Hotel Filter

Frequent Saver (Cluster 1)

  1. Frequent transactions

  2. Budget-conscious

  3. Promotion-Aware

  4. Indifferent to Credit Cards

  5. Preference for Standard Hotels

  6. Use of Luxury Hotel Filter

High Spender (Cluster 2)

  1. High-Spending

  2. Promo Hunter

  3. Moderate Credit Card usage

  4. Preference for Luxury Hotels

💭 Our Proposal

Each user segment will have a different Search Results Page UI, with different recommendations specifically catered towards their interests and concerns.

Tailoring