AI-Driven Hotel
Recommendation
role
Product Design Intern
Project Lead
timeline
July-Sept 2023
impact
Inc. engagement rate
Boost CTR & CVR
stakeholders
Business Intelligence
Design Managers
PM
Key Summary
We are aiming to leverage AI and delivering tailored recommendations to users based on their clusterings / segmentation to reduce drop-offs and boost conversion.
Results Highlight
The project in numbers
Interview with major stakeholders
Segmented Users Type
Tailored User Interfaces for each segment
Visual Glimpse
The Visual Output of This Project
Problem
Enhancing user experience goes beyond design; it demands strategic collaboration and leveraging advanced technology.
Nowadays, Artificial Inteligence is very accessible for most of people. Our company's strength lies in our dedicated Business Intelligence team harnessing AI and machine learning for insightful data analytics, refining decision-making in business processes.
But our design is not fully integrated with our BI team, so we need to step-up the collaboration game to improve our performance.
Search Result Page
Select Room
User Drop Off
On other note, there's a drop-off on our funnel (exact number is confidential), so we need to retain our funnel, and we were challenged to improve it using AI / Machine Learning.
Goals / Objective
The project in numbers
Make the design team aware about the use of ML/AI in the development of a feature.
Utilize existing ML/AI engine to enhance user experience.
Insights - Understand AI / ML it Self
Ingredients that will form the solution
What is AI ?
Artificial Intelligence is a computer system that is capable of performing tasks that typically require human intelligence.
Machine Learning (ML) is a type of computer science where computers learn to do difficult tasks without being explicitly told how to do each step. Instead, they rely on patterns and making guesses based on the information they have. Computers use ML algorithms to look at lots of past data and find patterns in it.
Even though machine learning is a part of Artificial Intelligence (AI), not everything AI does involves machine learning.
What's the Opportunity of AI in Tourism?
Users are constantly looking for relevant personalized contents.
AI in tourism offers a significant opportunity by meeting the increasing demand for personalized content. Using AI algorithms, businesses analyze user data to provide tailored travel recommendations, improving user satisfaction and engagement.
This technology streamlines travel planning and fosters deeper connections by offering relevant, customized content aligned with user preferences.
Understanding Data Science Team in Tiket.com
How is the team workflow cycle look-like
How do they work
AI in tourism offers a significant opportunity by meeting the increasing demand for personalized content. Using AI algorithms, businesses analyze user data to provide tailored travel recommendations, improving user satisfaction and engagement.
This technology streamlines travel planning and fosters deeper connections by offering relevant, customized content aligned with user preferences.
DS = Data Science | MLE = Machine Learning Engineer
Source: Data Science Lead at Tiket
Where Do Designer Stand?
Now, collaboration is the key and a pivotal aspect of a company. Designers play a crucial role in the data science workflow, merging complex data insights with user-friendly experiences.
We collaborate using our design skills to convert intricate data into visually engaging representations.
As you can see under the figure below, designer stand in the early process of the understanding in the problem definition together with Data Science team.
Source: Data Science Lead at Tiket
Data Science Engines at Tiket
Detection
Lacerta (Hotel Room Inventory Improvement)
Aurora (Image Scoring System)
Natural Language Processing
Pavo (Review Beautifier)
Apus (NLP Engine)
Corvus (Automated Email Parser)
Pricing
Orion (Dynamic Pricing)
Prediction
Serpens (Paid Hotel)
Why Cygnus ?
Cygnus focuses on user segmentation, aimed at categorizing users from multiple angles such as click behavior, transactions, and geolocation on the hotel booking process.
This segmentation offers valuable insights into user behavior, aiding the development of recommendation systems and dynamic pricing models.
To ensure accuracy and relevance, the user segmentation models are run monthly, updating user clusters accordingly.
Cygnus Hotel Customer Segmentation
Cluster 0
Cluster 1
Cluster 2
So, How Might We Improve User Experience by Utilizing Customer Segmentation Engine ?
Understanding Pain Points
The largest drop-off step: Select Room in Room Listing Page.
Search Result Page
Select Room
User Drop Off
Source: Amplitude of Tiket.com
The reason behind this are relevance issue, uncertainty issue, and benefit issue for new users.
Search Result Page
Select Room
60.2%
Back to SRP
Source: Amplitude of Tiket.com
The reason of this phenomenon is due to "Decision Paralysis"
The Decision Paralysis
Decision Paralysis is the lack of ability to decide out of fear of making the wrong choice, and 60.2% of customers are experiencing this.
This happens due to the overwhelming travel booking process caused by the abundance of choices (Choice Overload).
How we handle current situation
We currently use "Fornax", a ranking engine. Currently it offers users random suggestions. This random recommendation gives user little to no control over the product they’re seeing.
The ranking engine are very broad & wide, calculating all of the user's behaviour. Leading to a randomized recommendations.
The way to improve it
To improve the current experience, we will use 2 methods, the 1st is to leverage the data provided by our Data Science Segmentation Engine.
Then onto the 2nd step, we'll extract the data and translate them into visuals with appealing content based on their segmentations.
Leverage Data
Cygnus provides insights into 3 main user segmentation groups. We analyze each group’s behavior and translate them to contents and visuals.
Visual + Content
Give users relevant content based on their segmentation group, then alter the UI visual appearance for each group to provide a more personalized experience and attract users from specific segment.
Leverage Data:
Chunking
The ranking engine are very broad & wide, calculating all of the user's behaviour. Leading to a randomized recommendations.
We do this by labelling & grouping our recommendations in SRP.
Illustration of Chunking
Tailoring
Each customer segmentation group has different behavior.
To cater this different preferences, we tailor our UI visual appearance to match their behavior
Cluster 0
There will be 3 types of recommendations for cluster 0: Best Deal, Best Hotel Under 500k, and Most Affordable, Here's the breakdown.
Cluster 1
There will be 3 types of recommendations for cluster 1:
Hidden Gem, Best Deal, and Best Hotel Under 600k
Cluster2
There will be 3 types of recommendations for cluster 2:
Luxurious Premium Hotel, Promo with CC, and Affordable Luxury.
Solution
Visualss & Content Strategy
UI Visual for Economic Spender (Cluster 0 & 1)
Price related nudge
Show facilities, but less prominent than price
Highlight price for budget user
UI Visuals for Economic Spender (Cluster 0 & 1)
How we attract economic spender:
•
Rating & review for social proof
•
nudge scarcity to give sense of urgency
•
price related nudges to attract economic spender
•
highlight the budget-friendly price
•
show facilities as USP
UI Visual for Economic Spender (Cluster 0 & 1)
UI Visuals for The High Spender (Cluster 2)
How we attract economic spender:
Segmented SRP (Search Result Page)
Each Category will receive different recommendations
High spender will receive :
•
luxurious hotel,
•
promo that use credit card,
•
hotel with higher rating
Frequent buyer will receive :
•
luxurious but affordable hotel
•
promo
•
average to mid-high rating
Economic spender will receive :
•
budget hotel,
•
hotel under 500k,
•
standard hotel rating
Each Category will receive different product card
High spender Card Strategy:
•
focus on visual imagery & facilities
•
show price as installment
•
Higher rated hotel
Economic & Frequent buyer Card Straegy :
•
more compact card
•
highlighted price, with price related nudges
•
Average rated hotel
Retrospective
What could we improve next
In reflection, this project has been a remarkable journey despite facing bandwidth constraints that hindered our ability to conduct usability testing. Looking ahead, our potential for improvement lies in expanding variables within our interface, encompassing factors like location preferences, traveler types, and facilities. Integrating these variables could significantly enhance our user experience, catering to a wider range of user needs and preferences.
As we conclude this retrospective, this serves as a stepping stone toward refining and evolving our project, aiming to create a more comprehensive and user-centric solution in the future.