Adding an AI Feature into a Travel Booking App

This is a speculative/passion project based on Robert Redmond of IBM’s AI for Product Design program. During the progra, I was thinking about unique ways to incorporate AI into a travel booking site.

Background:

I have been playing around with Generative AI and LLM’s for several years now. I have a deep interest and excitement around AI but there were still a lot of open gaps in how I, as a product designer, could use it, both in my design process, but also how to incorporate AI features into an existing product.

When I found Robert Redmond’s course, it was exactly what I was looking for. Lecture-driven classes with hands-on practice. With the knowledge of the AI models that exist, and a strong foundation on how to integrate AI (considering ethics, evaluating the product, and creating the data model), it has opened my mind on different ideas and it has been so fun to explore these.

The Problem

TuCasa is an imaginary travel-booking site, inspired by Airbnb. It is a place to dream, explore homes, and book dream trips. But TuCasa has a problem. TuCasa cannot accurately predict it’s gross revenue because of cancellations that happen last minute. TuCasa started to implement surveys to understand why users cancel and they discovered the following reasons:

  • Users do not have all the information to accurately predict the best times to travel to a destination. Close to the date they found out that it will be rainy, or very expensive, or during a particularly hectic time in that country

  • When users get closer to the data, especially later in the year, they realize they don’t have enough vacation days left

  • As a result TuCasa cannot accurately predict gross revenue

Is there a way that AI could solve this problem for users and TuCasa?

Research

Research shows that the travel industry is adopting AI and other new technologies. While AI has already been used in many use cases, for example in flight forecasting, personalized recommendations, and of course, users are using it to create itineraries in ChatGPT, there are still so many ways that AI can be used. According to Brian Chesky, CEO of Airbnb, he called AI

“certainly the biggest revolution in travel — in tech — since I came to Silicon Valley.”

AI can certainly provide a lot of value on the consumer side of things. According to a study by Longwoods International, nearly a third of US travelers (32%) are likely to use ChatGPT to plan a trip.

of Americans are likely to use ChatGPT to plan a trip!

According to a travel insurance article, here are some reasons why travelers cancel trips:

  • Illness, injury, or death

  • Acts of Terrorism

  • Employment changes

  • Financial situation changes

  • Learning more about destination (aka traveling during high season, rainy season, etc)

Define & Ideate

What are we solving for?

Users do not have all the information to be able to accurately predict the best times to travel to certain places, the best places to travel based on weather, prices, etc, and do not maximize their allotted vacation days with US holidays. Often, users cancel their plans because when they get closer to the date, they realize they are traveling during very expensive times. The problem for the business is that they receive a lot of cancellations and cannot accurately predict profit. We think that AI can help reduce cancellations through providing accurate information around predictive pricing and best times to travel.

What is the user contribution?

Users are inputting their paid vacation days, their paid US holidays, their desired locations and ideal time to travel.

How will the product will solve the problem and create a unique value proposition for our user or the business? CasaBot will give users all the data they need to make informed decisions about when are where to travel to, to reduce cancellations. This allows the business to have better data on future bookings and allows the user better peace of mind when booking travel.

What is the AI’s contribution?

A combination of Time Series Analysis for forecasting and regression analysis to predict users behavior over time. The AI will be able to assess the following: - The first thing the AI will do is help the user maximize their vacation days using their US holidays. Based on these inputs our AI model will then provide suggestions on: - where and when the user should travel based on historical data around weather and travel prices, as well as previous user and cohort data As an example, when we first launch the app, we input our vacation days, US holidays, and desired travel locations and times. The AI will then give suggestions to maximize vacation days and also best times to travel based on weather and prices. For example, the product will show the suggested date to travel to, how long, and where to.

Reasons of “why” this is suggested might be:

  • Prices are 10% lower for hotels and flights

  • Weather is sunny and dry

  • Users are 10% less likely to cancel their vacations during this time

The CasaBot acts like an Advisor and Personal Assistant

What is the value to the user, to the business?

By providing users with accurate information on pricing, weather, and based on their input, they can feel more confident in their vacation planning. The business can reduce churn and cancellation rates because previous reasons for why users canceled are being taken care of by AI.

Although it is the job of a data engineer to understand how to machine will compute data, it is still important for designers to also understand what are expectations are of the machine. Breaking down the information architecture of the feature or product is important so that we can figure out blindspots or considerations before we go to deep.

It’s also important that we really think about the data we’re collecting, what is absolutely necessary and what is not. Below is a document to define when and where a user is interacting with technology, what is being processed, and the resulting possible outcomes in the UI. 

I also wanted us to think about what data we were using, what was vital for the feature and what was not. As designers working with AI, privacy and fairness is important and I wanted to make sure that we weren’t accidentally creating bias.

Design & Prototype

Below are high-fi wireframes of the design. I wanted to create an entry way into the feature through search. However, there are many ways that this feature could be integrated, besides a form. For example, instead of a full itinerary through a form you fill out, when a user searches and our AI think there are better times to travel, we could simply send that message with similar bookings during the more optimized dates.

As designers using AI, we have to be risk averse. We have to try to consider every issue, risk, privacy concern. A SWOT analysis is helpful to understand some of the risks and benefits and look at them next to eachother.

Ethics & Further Considerations

A booking site like TuCasa has a lot of powerful when it comes to influencing users and potentially manipulating the market. For example, if our machine is explaining to a user that (as an example) traveling to Rio during Carnival is more expensive than traveling during other times, could that have an affect on the local homeshare economy? What does that do between TuCasa and the hosts? Are we breaking trust with them? I don’t have the answer. I think that it’s important to consider this and also how we communicate these changes to the hosts as well.

Overall the feedback was positive about feature like this! I think there is a huge opportunity for travel companies to use AI to lead users in the right direction, maximize their vacation days, and provide valuable data on traveling during certain times.

Testing the Prototype

I tested my feature with 5 users. Here was some of the feedback:

  • Some elements are too small

  • Wish there was option to override AI (for example, don’t care if it’s too expensive during July)

  • Ability to pick & choose aspects of itinerary

  • Wish there was an option from regular search for AI to not just open form, but provide suggestion and you save that suggestion (not launch form every time) - what happens with repeat users?