Project Overview

To establish ourselves as the leading digital airline globally, Singapore Airlines aims to simplify trip planning through cutting-edge technology. We are proud to be among the first airlines to integrate generative AI (GenAI) into our products and services, launching the Smart Flight Recommender and GenAI Search on our website.
The Smart Flight Recommender aids in trip inspiration by helping customers find fares, recommendations, and comparisons easily. Our GenAI Search tool has revamped website search with GenAI capabilities, providing single-answer responses to customer queries.

These innovations originated from an internal hackathon where employees identified key pain points in the trip planning process. The solution was to leverage AI technology for a more efficient flight and information searches.

Our tools align with new user behaviours influenced by ChatGPT, where users prefer natural language input over traditional keyword searches or website navigation. This approach fosters a more personal and intuitive interaction with our airline, delivering precise results that meet direct requirements. The Smart Flight Recommender and GenAI Search have significantly streamlined tasks that previously required multiple searches and browsing into just two steps.

Customer satisfaction with our GenAI Search tool has doubled, and conversion rates from our Smart Flight Recommender have shown promising growth from browsing to booking. As one customer noted, “Love it! I can compare flights without opening different screens.”

These initiatives ensure that our customers experience exceptional service throughout their travels with Singapore Airlines, beginning from booking flights on their digital devices.

Organisation

Singapore Airlines Ltd

Team

Product Owner:
SIA E-Commerce & Distribution - Andy Koh, Vivienne Khoo, Leslie Goh

Product Development:
SIA Information Technology (Developers, Data scientist) - Zhibo Xiao, Lauren Goh, Rajasekar Venkatesan
SIA Information Technology (Products Delivery) - Elizabeth Wong, Jolene Aw, Vetriventhan V
SIA Information Technology (UX) - Jowin Tam, Yasintha Perera, Lloyd Loh, Naweera Sidik

Supporting Business Units:
SIA Customer Affairs, SIA Customer Contact Services, SIA Ground Experience Development

Project Brief

The project was to address a common frustration among our customers: the time and effort required to research while planning for their trip. This includes flight search and information in preparation for their flight.

Travellers with specific requirements often spend considerable time searching and comparing flights and prices, frequently navigating back and forth between search results pages. This challenge is compounded by various considerations such as budget, cabin class, and travel period.

Additionally, we have a segment of customers who are seeking inspiration for destinations we fly to. They are keen to travel, are flexible to travel anywhere, but may have a budget to keep.

We also needed to address another pain point for our customers which is streamline the research required in preparation for a flight. One example would be finding information on baggage restrictions which often involves navigating through multiple website pages or indexed search results.

Our goal was to reduce the effort associated with search during the trip planning stage. We aimed to help users get the answers they need more efficiently by interpreting users’ intent and crawling over 800 content pages and consolidating information into a single source of truth.

The Smart Flight Recommender and Gen AI Search were designed to streamline the user journey, making it easier and more intuitive for users to find and compare flights, as well as access relevant information.

Project Innovation/Need

The primary objective was to reduce the time customers spend searching for flights. Statistics reveal that users spend over 303 minutes on travel content 45 days before booking a trip, with 128 of those minutes dedicated to airline websites for fare comparisons [Source: Expedia Path to Purchase 2023].

When searching for flights, there are many parameters that a customer considers which include destination, dates of travel, cabin class, budget, and availability of flights. Our goal was to create a tool that was easier to use than the current practices of searching and comparing flights on our main booking page.

We wanted the tool to inspire them during the discovery phase of trip planning. There are customers who wants to travel but did not have fixed destinations in mind.

The second objective was to address the frustration our customers face when searching for relevant information in preparation for their trip. They may have questions regarding the places we go to, the flight amenities available, and travel restrictions which are commonplace in the aviation industry.

Customers spend a lot of time looking for answers through traditional website navigation, frequently asked questions or traditional keyword search that gives indexed pages as answers.

To meet these needs, we wondered whether technology could reduce this by a substantial amount of time.

Our design aimed to minimise the number of steps from inputting search criteria to receiving results while maintaining accuracy and usefulness.

Design Challenge

One of the challenges with generative AI is the issue of hallucination. With the Smart Flight Recommender, we mitigated this risk by ensuring the customers’ intent are translated into actionable queries in our backend booking system which would then return accurate and relevant responses that are aligned with what the customer needs. We also implemented measures to detect and appropriately respond to inappropriate or irrelevant queries. We do not collect personal information on our site, reducing security risks and allowing users to freely utilise the tool.

However, for GenAI Search, hallucination was a more valid concern as the answers are generated by the pre-trained transformer models that we use, and varying search results that are inaccurate or harmful could negatively impact our business and customers. To address this, we conducted rigorous testing with over 1,500 internal staff, involving all Business Units. Multiple rounds of user testing helped us identify and correct inaccuracies. From the user testing, we also developed a rigorous set of question-answers that allow us to automate regression testing when we change or upgrade our transformer models.

Once confident, we expanded our testing company-wide to increase awareness of GenAI’s capabilities and to perform stress tests. These steps allowed us to launch the tool to our customers with confidence in its ability to accurately answer customer queries.

Future Impact

With the Smart Flight Recommender, customers can describe their desired trip by specifying various parameters, including travel intent, with a few words, which the recommender translates into possible destinations. For example, a customer might input “romantic getaway under $2000 in December” and receive a list of destinations to compare, with descriptions of each city’s unique features and best prices available. It recognises dates for holidays of all countries, eliminating the need for customers to consult a calendar. The Smart Flight Recommender combines the emotional aspect of trip planning with the logistical aspect of flight search.

Customers can continue to search information through our GenAI Search that allows natural language queries, making it more intuitive than traditional keyword searches. Responses are summarised, reducing the cognitive load of sifting through text-heavy information.

Month-on-month, we see an increase in number of customers who have rated the feature ‘useful’. Out of which, 49% found it ‘accurate or complete’ and 23% found it ‘easy to understand’. These statistics grow as we continually improve the capabilities of AI. Combined, the Smart Flight Recommender and the GenAI search are useful, complementary tools for our users to plan their trip.

Since its launch in Apr 2024, the tool has averaged 2,000 queries per day. Click-through rates from the tool to our flight search page have increased from 23.8% in April to 34.4% in July. Conversion rates have also been promising, with a healthy number of users booking flights originating from the Smart Flight Recommender.