Draft Post - August 2023

Can we improve Google Flights with Generative A.I.?

A hypothetical screenshot, combining SGE and Google Flights UI elements, of the checkout process.

TL/DR: I have been lucky to spend a few years in the travel industry and to have been part of building an award-winning booking application for the airline Flyr.¹ Drawing on that experience, this post explores how Google Flights, my favorite flight booking search service, can leverage generative A.I. to improve two key aspects: (i) flight bookings with more detailed search criteria, and (ii) the checkout process. It proposes the integration of a new module, "Flights SGE," into Google's Search Generative Experience (SGE). The objective is to improve the user experience of searching and booking flights on Google, thereby seeking to drive user engagement and satisfaction. This detailed, 15-minute post features an improvement case with storyboarding and design illustrations that mix Google SGE and Google Flights UI elements. If short on time, peruse the illustrations for the general concept.²

Limits of self-service in flight booking

Google Flights empowers any person to search millions of flights. This is a far cry from the pre-internet era, where securing a plane ticket was a protracted exchange of explanations, repetitions, and question-and-answer sessions, often conducted over a phone or across a counter. With its intuitive interface, powerful search algorithm, real-time price data, price tracking features, price graph, and more, Google Flights offers users a comprehensive, reliable, and user-friendly tool for flight comparison and booking.

However, even with powerful meta-search engines like Google Flights, searching and finding the right ticket can sometimes be time consuming and the one-click checkout experience is far from a reality. For instance, working with flexible dates and trip lengths within a specific time period, or comparing alternative routes for a vacation, still requires the extra time to actually do and compare each search result to find the flights and price you are most happy with. Users can find these processes tedious and overwhelming due to the multitude of options, filters, and screens they have to navigate through. Additionally, the repetitive input of passenger and payment details for each booking during checkout, adds to the frustration and complexity of the task. Indeed, while 95% of respondents love being on holiday, 43% of US respondents in Travelport's 2022 survey found the process of booking travel less enjoyable than other online experiences like booking restaurants or buying clothes.³

Thinking from first principles

To understand better why the booking process is how it is, how these commonplace but complex tasks remain arduous, and how we might improve them, it is helpful to think from first principles. To start, central to flight booking is the pursuit of a "good price for a good flight."⁴ This goal is typically achieved through three key steps:

• Discovery: Finding the right option.
• Selection: Choosing the desired option.
• Checkout: Providing necessary flight, passenger, and payment details.

Each of these stages is crucial, removing one would inhibit the user's agency and the gathering of essential information for the airline, yet there is ample scope for their enhancement.

Consider the Discovery step. At this step, leading flight search websites like Google Flights presuppose that users want and need to independently review and compare a myriad of flight options to secure the best deal. Accordingly, these sites are geared toward making this task—flight search and comparison—simpler and more efficient, but not toward eliminating it altogether. The arduousness of some flight searches arises because self-service inherently requires user action. Thus, despite easing the tasks, the user still must execute them. In scenarios with multiple tasks necessary to accomplish the goal, this then can become tedious. However, recent advancements in Generative A.I. have led to the ability to interact with computer systems using natural language as well as the creation of agents that are increasingly capable of executing tasks on our behalf. This raises an important question: Could it be more effective to use these conversational agents for certain tasks, such as finding a "good price for a good flight"? And if so, how? Moreover, we might question the conventional linear, static checkout process that requires sequential steps to finalize the booking, and envision a more dynamic, personalized process.

Business objective and solutions

With these insights in mind, this improvement case will explore an agent-based Google Flights module. The business objective will be to improve the user experience of searching and booking flights on Google, with the aim of boosting user engagement and satisfaction. This aligns with Google Travel and Google's broader product strategy, which emphasizes building a superior user experience for finding and booking flights.⁵

It is worth noting that several possible solutions could be considered towards this objective. Outside of generative AI, we could imagine using the user's previous patterns and behaviors to further personalize flight search results and recommendation. Further, we could introduce more nuanced filtering options (e.g. seating preference) or build the option to perform single searches that combine wanted flights, hotel, car, and experiences, thus significantly reducing time spent on overall travel planning. On the generative AI front, we could imagine adding an AI co-pilot-like feature on the current flights search dashboard or adding summarized real-time flight results in Google Search, linking directly to the specific flight options on google.com/travel/flights when pressed.

Yet, when weighing these options, "Flights SGE" stands out as a particularly promising solution. This hypothetical generative A.I. module is designed for seamless integration into Google’s newly launched Search Generative Experience (SGE), thus aligning well with Google's commitment to creating faster and easier search experiences.⁶ Moreover, due to its ad-free interface and free partner listings on google.com/travel/flights, Google Flights generates revenue when incorporated into core Search as a module.⁷ Similarly, Flights SGE, with its design tailored for SGE integration, holds promise for similar or improved ad revenue generation as an envisioned part of SGE (See how sponsored content can figure in SGE at Google's blog here.). Finally, the modular nature of Flights SGE makes future improvements, like combined flight, hotel, car, and experience searches, more easily integrated without adding further UX complexity.

Measuring success

Evaluating the success of this new module will require a comprehensive range of metrics, core of which are selected as follows:

• Number of flight searches: This is our North Star metric. The total number of flight searches made by users. Reflects user engagement and the module's reach, providing a high-level view of its activity and usage..
• Proportion of total usage: Percentage of searches on Flights SGE vs. Google Flights search dashboard and percentage of Flights SGE checkouts compared to on partner sites. Underscores the module's efficacy in the ecosystem.
• Conversion rates: Percentage of users who move from searching to checkout and purchasing. Gauges the module's effectiveness in converting interest into actionable outcomes.
• User satisfaction and flight suggestion reliance: Gauges satisfaction during and post-interaction, especially after complex bookings, and assesses the accuracy of flight suggestions through user interaction and feedback.

In addition to these main metrics, it can be helpful to monitor any reduction in the time it takes for users to find and book flights, a sign of an increasingly efficient and user-friendly search and checkout process. In addition to these direct measures, indirect metrics such as customer retention rates and the influence of the module on broader Google ecosystem engagement warrant close attention. Through regular review of these metrics, we can gather important insights and enable timely refinements, ensuring that Flights SGE will continue to align with Google's objectives of delivering value to users and partners.

Storyboarding the booking process

The following storyboard illustrates Flights SGE's main functionalities and the personalized booking experience. The storyboard relies on the SERVICE principles—an approach I derived for designing A.I. agent-user interactions grounded in customer service principles. In brief, the premise of the principles is that good A.I. agents can learn from good human agents. Highlighted words below (e.g. salient, indicative) are drawn from the principles.

A fictional photograph of our fictional persona, Lina.

Let's consider a hypothetical UX persona, Lina, a vibrant 26-year-old graphic designer from Austin, Texas. Lina embodies the archetype of our main user group for this improvement case: young, tech-savvy, frequent travelers with a keen sense of adventure and a careful eye on their budget.⁸ While our agent-based interface can potentially cater to various users, such as family travelers, business professionals, and those less tech-inclined, there are several reasons to focus on this young, Gen Z and Millennial, segment:

• Future of Travel: Millennials and Gen Z are leading the way in leisure travel trends, with 69% and 68% engagement rates, respectively, according to a June 2023 Bankrate survey.⁹

• Unique Requirements: Over 80% of those in these age groups actively seek distinctive experiences, as reported by Expedia, reflecting a shift towards personalized travel.¹⁰

• Budget-Conscious: 81% of Gen Z factors in budget when orchestrating their travels, which stands in contrast to the 57% of baby boomers, as indicated by Condor Ferries.¹¹

• Dissatisfaction with Current Booking systems: A quarter of Gen Z respondents share the sentiment that the complexity of searching, comparing, and booking travel offers is not fun, as reported by Travelboard.¹²

These points resonate with Lina's own travel experiences. Despite her love for exploring, she often finds the process of searching for, and comparing, the best flight deals tedious and time-consuming, especially given her specific requirements for affordable fare and comfortable seating. As she plans a weekend getaway to Miami to visit old friends, her search is flexible in regards to dates but firm on these two other criteria. Ultimately, Lina, a budget-conscious adventurer, seeks a more streamlined solution—one that can simplify her process of booking affordable flights and fulfill her specific needs with minimum work.

It is important to note that, while the storyboard offers valuable insight into one user's journey, it does not capture all possible user scenarios or fully mimic real-time interaction dynamics. For example, the unique needs of family travel or the quick adaptability required by business travelers necessitate future consideration and exploration. In addition, the incorporation of sponsored content within the design has been omitted for the sake of simplicity. However, SGE does offer an integrated ad experience,¹³ and in Flights SGE, this could manifest as sponsored recommendations (e.g. hotels and experiences) and upsells (e.g. upgrades and extras), all designed to align seamlessly with user queries.

Discovery

Scene 1: Flight Preferences and Discovery

Source: Mashup by author of hypothesized and existing Google Flights and SGE UI elements.

Lina initiates her interaction with Flights SGE, articulating her specific travel needs, "Hey, I need flight tickets to Miami in September. Which weekend is cheapest? I can leave Thursday or Friday afternoon and need to be back by Monday afternoon. I prefer a window seat, extra leg room, and no layovers."

In response, the Flights SGE agent presents a salient, focused selection of the most suitable pair of flights as interactive cards. These cards, which are clickable and expandable, offer Lina a choice to review alternative flights within Flights SGE or to shift to the Google Flights search dashboard. Prompts such as "Ask a follow-up" and "Book (only me)" are displayed in a manner that's indicative of the next steps, facilitating Lina's decision to refine her search or proceed to checkout. The agent presents all options in a clear, reviewable format, detailing that the suggested tickets meet her request for a nonstop flight with her preferred seating.

The Flights SGE agent also reassures Lina of its choices by showcasing the cost-effectiveness of the selected flights through Google Flights' Pricegraph, which provides an option to explore alternate dates. With the added explanatory information at her disposal, Lina feels relieved seeing all her requirements met in one go.

Selection

Scene 2: Memorized Selection

Source: Mashup by author of hypothesized and existing Google Flights and SGE UI elements.

After confirming her travel date with her friends, Lina asks the agent, "Is the September 15th flight still available?" The agent, understanding the context, resurfaces the earlier flight options, highlighting any changes in price or other details. If Lina's original flight selection becomes unavailable, the agent empathetically offers alternative options that closely align with her initial preferences, acknowledging the unavailability of her initial choices.

The agent customizes the suggested next steps based on Lina's booking history, offering options to book either for herself alone or for both herself and Eric, her regular travel companion.Lina feels excited that the price is still available and feels a subtle rush to continue. She proceeds by selecting "Book (me and Eric)."

Checkout

Scene 3: Seamless Review & Auto-fill

Source: Mashup by author of hypothesized and existing Google Flights and SGE UI elements.

Upon Lina's selection, the agent provides an overview of her flights and enquires about any additional changes or services. Using vaulted (securely stored) information, the agent auto-fills passenger details for both Lina and Eric, with an option to add more travelers. The agent further suggests a potential upgrade by demonstrating its benefits and costs. Relying on Lina's previous travel habits, it subtly highlights her commonly chosen extras ("Add checked bag") for consideration. Lina appreciates the effortless choices and controls.

Scene 4: Extras and adapting to selection

Source: Mashup by author of hypothesized and existing Google Flights and SGE UI elements.

When Lina requests an extra checked bag, the agent immediately incorporates it into the booking and displays an overview of other possible extras. The checkout journey remains flexible and streamlined, highlighting only the salient steps and adapting based on Lina's profile and actions.

Scene 5: Confirmation and bespoke recommendations

Source: Mashup by author of hypothesized and existing Google Flights and SGE UI elements.

Lina instructs the agent to finalize the booking, to which it responds, "OK, I’ll make the booking and email you the confirmation." The agent enhances Lina's booking experience by suggesting a list of popular activities and accommodations at the destination that align with Lina's preferences.¹⁴ Lina feels intrigued as she glances over the bespoke recommendations. Although Lina and Eric will be staying with a friend on this trip, she appreciates the agent's suggestions as a handy resource for future travel planning.

Improvements & optimizations

Flights SGE reimagines the flight booking process through the use of an A.I. agent within Google's SGE. Highlighted in the storyboards above are the following key improvements we proposed for Google Flights:

→ Unified, Direct Query

Flights SGE simplifies the discovery process, performing multiple queries on behalf of users like Lina, and offering the most suitable option according to their input and preferences. We can also envision Lina's initial query incorporating her hotel and experience preferences, with the agent then providing the most cost-effective combination and dates. Although the system can be made to handle a broad range of user queries, more complex ones may still pose challenges, making fallback options necessary.¹⁵

→ Simplified, Personalized Views

Flights SGE dispenses with the many buttons and filters previously required on self-service platforms for search customization. The interface now focuses on personalized, frequently used options, and includes a free-text input bar for precise requests. To tackle the challenge of ambiguous user requests and the potential restrictive feel of the interface, the agent should be designed to interpret context and provide easy corrective input avenues, ensuring the user always feels in control.

→ Reduced Page Navigation

Flights SGE tailors the ticket purchase process to individual users, replacing the fixed, multi-step user flow with flexible, multidirectional navigation. This provides a more engaging user experience with fewer repetitive steps. To aid users unfamiliar with the booking process, the system could offer relevant tips along the process and, if possible, an "Edit order" feature even after a booking is made.

→ Streamlined checkout and reduced input

Building on "Book on Google,"¹⁶ Flights SGE minimizes typing by auto-filling and storing Google Pay details and intelligently suggesting previously used data such as passenger details, commonly used ancillaries, and more. With user consent, customer data could be securely transferred to partner A.I. agents for a personalized checkout process on partner sites as well. Ensuring user trust in data safety and security is paramount, especially with the potential future data transfers to partner OTAs and airlines.

→Tailored Upsells

Traditionally, while the upselling of extras is crucial for airlines and OTAs' revenue, it often complicates the user experience by adding extra steps and pop-ups in the checkout process, presenting a delicate choice between driving ancillary sales and maintaining a streamlined user experience.¹⁷ Flights SGE addresses this issue by streamlining upselling and personalizing the list of suggested extras that are most likely to appeal to specific users like Lina. In theory, a better user experience and more relevant suggestions ("ads") for upsells should produce more sales,¹⁸ but this will need to be tested and optimized for our specific use case.

Limitations, potential, and next steps

The introduction of Flights SGE, a generative A.I. module of Google Flights, could offer substantial enhancements to the platform's user experience. By directly addressing user challenges—notably, the time-consuming nature of flight bookings with detailed requirements and the often tedious checkout process—Flights SGE could deliver a more personalized and efficient booking experience. Such an improvement aligns with Google's key business objectives of improving user experience and engagement. This approach would allow Google to further distinguish itself from competitors, potentially attracting a greater number of users to its platform for flight search and booking.

However, it's crucial to note the uncertainties and potential risks of this new approach. There are possible limitations of A.I. technology to consider, including its capability to handle complex user queries quickly and interpret ambiguous requests. Questions arise, such as whether the information presented will feel actionable, if users might prefer the current flight search dashboard to text and speech prompts, or if users will trust the A.I.'s recommendations without manually verifying the search results.

These issues call for further storyboarding, prototyping, user testing, and meticulous monitoring of the key metrics listed in this post. Nonetheless, a wealth of research indicates that these challenges can be overcome. Minimalist design principles suggest that a simplified interface can bolster user understanding and decision-making.¹⁹ Studies on conversational interfaces indicate that users can easily adapt to text and speech prompts, making the transition from conventional flight search dashboards seamless.²⁰ In terms of trust-building, maintaining transparency in A.I. systems and offering clear explanations can significantly boost user confidence.²¹ Research also underlines that users favor efficient, customized online experiences.²²

Next steps

While the storyboard and "Lina" represent a common user archetype, it's crucial to consider diverse user scenarios and pain points. Future storyboards could encompass more varied user types, helping to devise a comprehensive solution. It is worth noting that the agent-based interface might also be appealing to less tech-savvy users, potentially bridging the digital divide by making online booking, especially more detailed flight bookings, more intuitive and accessible.

Beyond immediate improvements to the user experience, the implementation of a Flights SGE like experience may also pave the way for additional revenue streams. These could be realized through the facilitation of related service upsells²³ and more seamlessly integrated ads. This approach has potential to not only refine the flight booking process but also amplify Google's value proposition for both users and its own revenue generation objectives, aligning with the company's broader product strategy.

As we look further into the future, it's hard not to get excited by the opportunities presented by evolving user behaviors, particularly the younger generation's inclination towards visual and video inspiration. There is big potential for seamless integration of inspiration and action in travel planning. Imagine a prospective traveler watches a YouTube video about a yoga retreat in Fiji, and from within that very video, an A.I. a travel agent helps plan their trip there and finds the best price and dates for the combination of flights, hotel, and the desired experience. This seamless integration of inspiration and action aligns with Google's mission to organize the world's information and make it universally accessible and useful, pointing to a promising future for travel planning.

Notes