Photo by SpaceX

A Review of Travel Chatbots

TL;DR

• Travel search is one of the most common use cases for chatbots, we reviewed the 5 main travel bots on Facebook Messenger
• Travel bots can differ significantly on capabilities (how many things the bot can do), and also on onboarding and helping users. One bot already implemented the “Link Account” function to send notifications after the booking
• Some bots understand complex text input, others require information provided one at a time in a ping ping conversation. The latter limits what a user can do and provide a suboptimal user experience
• Current bots have limited search capabilities, similar to website homepages, with no options to sort or filter results. With such limited features, will user choose them over the websites and apps?

Reviewed chatbots were tested on July 15th, 2016

Travel companies are among the most active in the chatbots arena, with many bots already released on Facebook Messenger Bot Store. In this post we look at the current state of the art reviewing Expedia, Hipmunk, Skyscanner, Cheapflights, and Kayak bots. For each bot, we analyse 6 features:

1. Capabilities: what can the bot do?
2. Onboarding: how does the bot introduce itself?
3. Help: how is help implemented?
4. Flow: is there a free flow conversation or a predefined list of questions (ping pong conversation)?
5. Results: how are search results presented?
6. Natural language understanding: how good is the bot at understanding user input?

Expedia

Onboarding, Help, Results

Link to chatbot

1. Capabilities: Basic hotel search
2. Onboarding: Simple message with a “Get Started” button
3. Help: Not much help on how to use the chatbot, just a link to customer support website and “Start Over” button
4. Flow: A ping pong conversation with a sequence of 3 questions (location, check-in date, number of nights). After each input, it gives a feedback on what it understood and some buttons for common values (ex. Tonight, Tomorrow). Overall, it feels like a fixed process implemented inside a chat UI
5. Results: 5 cards with images, prices and links to the site
6. Natural language understanding: By far the worst at understanding user input. It can understand information only in a predefined way (ex. you can’t reply “3 nights” to the question “how many nights?”, you must write “3” for the bot to understand). It also has a fairly aggressive auto-correction on city names, if you type “Hello” it will set the location to “Le Bec-Hellouin, France”

Hipmunk

Onboarding, Help, Results

Link to chatbot

• Capabilities: Basic hotel and flight search, create fare alerts and get travel advice (ex. “Best time to fly to SF”, “looking for a beach vacation from NY”)
• Onboarding: Card with 3 buttons explaining the main features (“Search Flights”, “Search Hotels”, “Travel Advice”), similar to help function
• Help: Similar content to onboarding, strangely no buttons this time but just text bubbles
• Flow: Free flow conversation, bots understands some complex inputs ans asks when it requires more information. Provides feedback on what it understood and hints at more options during the conversation (ex. It says that you can define specific airports after you provided the basic search details)
• Results: 9 cards with images, prices and links to the site. Flight cards include “Create a fare alert” and “Sort by cheapest”
• Natural language understanding: The bot can understand more complex user input, including getting all booking info in just one sentence (ex. “Flight London Milan today return tomorrow”) and understanding inputs like “next friday”

Skyscanner

Onboarding, Help, Results

Link to chatbot

• Features: Basic flight search, search flights by country or city in a given month and have suggestions on best weekends or weeks
• Onboarding: Simple message with a “Contact us” button with link to customer support website. After first user input, bot sends another message with a “Not here for the bot…” button, again with link to customer support website
• Help: Basic information and buttons to “Start over”, go to “Customer support” website or “Talk to a human”
• Flow: A ping pong conversation with a sequence of 4 questions. Provides feedback on what it understood and buttons for changing values
• Results: 3 cards with prices, buttons with link to the site and to show 10 results
• Natural language understanding: The bot can understand some user inputs like “next friday” or “today”, but cannot get all booking info in just one sentence

Cheapflights

Onboarding, Help, Results

Link to chatbot

• Features: Basic flight and hotel search, some discovery searches like “Inspire me!” or “Somewhere hot”, and a currency converter
• Onboarding: Simple message
• Help: By far the best help section, 3 cards with full explanations on all the features, plus buttons for “Customer Service”, “Download free app” and “Newsletter sign up”. Until July 22nd, there is also a 4th card with a competition to win free tickets by using and sharing the chatbot
• Flow: A ping pong conversation with a sequence of questions. After each input, it gives a feedback on what it understood. While the flow is not optimal due to the lack of understanding of user input, the tone and overall design is very well done
• Results: 1 text bubble with the cheapest price, with a “Book now” link to the site and “Search other dates” button
• Natural language understanding: Similar to Expedia, user has to write in a fixed format and provide information one at a time answering bot questions

Kayak

Onboarding with Help, Results

Link to chatbot

• Features: Basic hotel, flight and car hire search. You can also search for things to do and get travel advice (ex. “Where can I go for £500?”, “when is the best time to fly to New York”)
• Onboarding: It’s quite long and thorough, with 4 text bubbles and 4 buttons. The first half uses the new Facebook Messenger feature “Link Account”, where you can link you FB account to your website account to receive notifications (ex. Flight delays). The second part is the help section. Bonus point, it’s the only bot that greets you with your name using FB account information
• Help: Help is a bit counterintuitive, when you write “help”, the bot tells you to write “search” for getting help. After you type “search”, you get 3 buttons with explanations on “Flights”, “Hotels”, and “More options”
• Flow: Free flow conversation, bots asks when it requires more information. Provides feedback on what it understood, no buttons for changing values. For flights, it asks first to select the outgoing and then the return flight, other bots suggesting already the combination of the two feel easier to use
• Results: 10 cards with prices and button to the site
• Natural language understanding: Like Hipmunk and Skyscanner, the bot can understand more complex user input, including getting all booking info in just one sentence

Conclusions

Photo by SpaceX

In this blog post, we reviewed travel bots based on 6 features. There is more to consider while designing a chatbot, like disambiguation (ex. how they deal with city names that cold be in multiple countries), ambiguous input, updating values, managing failure, etc..

What we liked

• Bots are new and lots of praise goes to all these companies trying to figure out the best way to use them
• Cheapflights extensive help section and Hipmunk feature hints are a good way to guide new users
• Link Account, as introduced by Kayak, has the potential of engaging proactively the users without being annoying. Bots should be able to start the conversation again in the future, the key is not being spammy

What we didn’t like

• Having ping pong conversations due to lack of natural language understading
• Limited user personalization. For example, 2 bots out of 5 didn’t recognize we were in the UK and showed prices in US Dollars. Plus, there is no clear concept of user preferences to speed up future interactions
• All chatbots have only basic search capabilities, no one has result sorting (apart from a “Sort by cheapest” in Hipmunk) and filtering (ex. max price)

With limited search capabilities, travel bots seem a channel to drive traffic to apps and websites, rather than substituting them. Will users be interested in using chatbots until this gap is reduced?

Conversate is a new Artificial Intelligence startup. We built an innovative Natural Language Understanding engine, with state of the art algorithms and enterprise services. Currently in private beta, if you are interested you can sign up at http://www.conversate.eu or write us at info@conversate.eu