There’s no denying that there’s a lot of hype around chatbots right now. Even though the rise of the chatbots is apparently in full swing, chatbots are not really new.
Surely, you’ve been on a website and attempted to converse with the automated customer service agent who pops up. Here I am talking to Julie, Amtrak’s virtual assistant/chatbot:
While chatbots aren’t necessarily new, they seemed to explode in popularity earlier this year when Facebook began allowing businesses and developers to build chatbots for Facebook Messenger that can provide customer support and other interactive experiences for users in a variety of industries.
Chatbots in the travel industry: Are they helpful?
One industry in which chatbot development is booming is the travel industry. Big Online Travel Agencies (OTAs), such as Expedia and Skyscanner, have developed travel bots that help users search for flights and hotels. The bots can answer simple customer service questions using conversational language. These bots, in many ways, simply provide users a different, more interactive experience—for performing the same task the users would be performing on an OTA’s site. Sure, there are a few benefits: If you want a more friendly experience or to avoid having to switch between 30 different travel apps, these bots might be appealing to you. But they have a long way to go before they are truly useful.
Many users report that these bots have difficulty understanding simple requests. Most of the bots require you to leave the bot’s chat window and visit the OTA website to actually book, which many users find annoying. Some also argue that bots are reactive, rather than proactive. This doesn’t separate bots from OTA websites, but human travel agents can anticipate concerns in a way bots can’t; so, for complex travel such as corporate bookings, some argue that we still need travel agents or travel management companies. Ultimately, while bots are still in their infancy, they don’t solve many problems right now.
Travel A.I. and Mobile Travel Chat (MTCs) may be the future
Despite some of the critiques of the early travel bots, there are companies creating technologies that are much smarter than some of the Slack and Messenger bots that seem to be garnering all the hype. Some of these technologies are being marketed as bots, but their capabilities go far beyond simple customer service bots. These tools are being marketed by the companies that have created them (or labeled by others in the industry) as virtual travel agents, virtual travel assistants, Travel A.I., mobile travel assistants, or—my favorite—Mobile Travel Chat (MTCs), a term coined by Travel Startups Incubator.
These MTCs—such as Lola, HelloGbye, Pana, and our own Claire—are, in some ways, bots. Yet, comparing these complex tools to some of the simpler customer service bots somewhat minimizes all they can do. While each of these MTCs differ, they all create a conversational, chat-based experience (like other bots) that helps users complete complex travel booking tasks. Some purely operate through text messaging and others allow you to talk to a virtual agent (like Siri). Some, like Claire, are primarily driven by A.I., whereas other MTCs like Tradeshift Go leverage the A.I. to connect travelers to human travel agents for more complicated tasks.
Whether your travel company wants to jump on the Facebook Messenger or Slack Bot bandwagon, or you’re an entrepreneur or travel company interested in creating a full-fledged MTC/virtual travel assistant, there are some important strategies that hold true for creating useful bots or MTCs that customers will truly want to use.
We’re not getting into the technical nitty-gritty here. We’ll leave that to the developer experts! Instead, we are sharing some useful strategies that anyone in your company can use to brainstorm and make plans for a bot’s development. In fact, a lot of the work and planning that goes into a travel bot’s development can be done by those who aren’t super tech-savvy. Your company can leverage the writing and marketing skills of others in the company in order to plan the right ways to implement and problem-solve the actual bot conversations.
1. Connect your purpose to your target market
The first step in your travel bot development process should be to determine your bot’s overall purpose. Travel bots can do many things. They can help users:
- search for travel
- book travel
- remedy travel disruptions
- learn the ins and outs of a new location by acting as a virtual travel guide
How do you determine your bot’s purpose? Figure out the problem your bot will solve for users. It may be a combination of the above factors, but you might need to prioritize these purposes.
Once you know your purpose, carefully consider your audience or your target market. Understanding how users typically use your website or product, the types of questions they might ask, their level of technological skill, and their travel preferences (to name a few factors) will help you make decisions in planning the bot’s functionality and personality.
2. Start small & focus on your travel bot’s unique value
The travel bots and MTCs that succeed will likely be the ones that try to avoid conquering every goal for the platform from the beginning. Even big-name companies like Expedia decided to start small. While the Expedia OTA allows you to book flights, cruises, cars, and more, the Facebook Messenger bot currently only searches for hotels. Paul Gray, the Director of Platform Services at Kik, tells the MIT Technology Review that in creating chatbots, “A pitfall is trying to do too many things at once.” According to Gray, “You should start off small and simple.”
Perhaps the best way to start off small and simple is to realize your bot’s unique value to your target market (tip 1!) and ensure the bot performs the task that separates it from the rest very well. Certain bots offer a particular, unique value. For example, HelloGbye—an app that can be used for personal and business travel—stands out because it can do multi-person itineraries. It can make flight and hotel reservations for up to eight people in a live itinerary, even searching for different preferences (i.e., one user wants a 3-star hotel and the other wants a 4-star).
The features that set one MTC apart from another in the travel bot space are what will help users decide which tool to use. Therefore, as you start small, focus on making your bot the best at performing that unique function.
3. Personalize, personalize, personalize
Good travel bots and MTCs need to allow users to enter a variety of traveler preferences. This is one of the ways that bots will become truly unique and useful for travelers. Mixing traveler preferences with data about a destination allows the bot to provide “relevant and contextualized information to its user,” as Dany Agostinho, Innovation Senior Specialist for Amadeus IT Group, argued on its blog.
Companies and developers should brainstorm the types of preferences that are most relevant to their product’s goals—and linking preferences to your unique value will help you tailor the preferences. Here’s an example:
Let’s say your bot specializes in hotel booking. You might consider asking if your travelers
- prefer a city or ocean view
- need a hotel that offers childcare
- want a smoking or non-smoking room
- prefer one queen or two twin beds
- value booking an eco-friendly hotel, boutique hotel, or corporate chain
- must have Wi-Fi
…You get the idea.
With corporate travel bots, these issues get even more complex. Yet, some MTCs tackle these issues well. Our A.I. travel assistant Claire, for instance, makes booking easy by allowing companies to input their corporate travel policies (and accommodating different policies for different groups of workers in a company), so that a traveler’s preference does not turn into extra money spent by the company due to out-of-policy spending.
As you consider personalization, you might also want to think about privacy. A Global Business Travel Association (GBTA) survey of 756 business travelers between March and April 2016 revealed that in North America, around 70% of travelers want options catered to their personal travel history and preferences. Yet, while travelers are willing to share some information (frequent flyer number, preferred hotel brands, preferred times to fly), they are not as willing to share their business calendar and appointments, their travel history, or preferred leisure activities while traveling. Companies will want to consider these preferences and think about how intrusive their bot will be. Creating a robust privacy policy may also help alleviate concerns.
4. Add some personality
A little personality can go a long way in bot development. The majority of media coverage on bots cites their ability to be “fun” and “friendly” as one of the major benefits to using a bot versus the standard blank text box and drop-down calendar of an OTA. Companies will therefore want to consider how to make the bot’s responses reflect an intriguing personality—and one that is consistent with the company’s brand.
One way to do this is to hire writers with creative backgrounds—people who usually may not be your top choice for working in product design. For instance, Howdy and X.ai hired writers with performing-arts expertise to create unique text for their bots. Writer and filmmaker John Pavlus notes that Howdy’s bot “affects a casual and lightly irreverent tone,” thanks in part to “novelist/satirist/former-improv-comedian Neal Pollack” being on the product-design team.
Not every company or startup can afford to hire someone new just for this purpose, so think about others in the company who might be a good fit. Your PR person might just become involved in UX design!
While adding a little humor creates a fun experience, be careful not to offend anyone. Pavlus reports that the co-founder of Howdy outlawed the bot from asking rhetorical questions because people expected a response, and X.ai is not allowed to use gendered pronouns. AI interaction designer Anna Kelsey tells Pavlus they made this decision because “You don’t want to burn a bridge with anyone by calling them a he when you’re talking to a woman, or vice versa.”
Excellent advice for those working on the bot’s personality is to create several paths for consumers. Joe Gagnon, Chief Customer Officer and GM of Aspect Software’s Cloud Solutions, suggests in Venture Beat that developers create a happy path for informational engagement; a “not-so-happy path for issue resolution”; a humorous path for teasers, and so on. Identify keywords associated with each path (example: “livid” goes to issue resolution) so humor only surfaces when the machine can be sure the user is happy.
5. Work towards being predictive
The fact that travel bots cannot anticipate users’ concerns in the same way as a human travel agent could is a potential problem. Your travel bot should be as predictive as possible in order to be useful for users. The bot should make suggestions that continually drive the conversation forward and even help users discover other functionalities of the bot. Maruti Tech Labs gives this example of a shopping bot conversation:
- “Hey bot, send flowers to my friend.”
- “Sure, which flowers would you prefer, Roses or Lilies? Would you like to accompany it with a card?”
Notice how the bot’s response moves the action forward by suggesting a flower type and even adding on card?
Travel bots should function similarly—particularly sophisticated MTCs. If a user has completed booking a flight, the bot could prompt the user to book a hotel in the area or offer to make reservations at a nearby restaurant. Some of the ways you accomplish this are quite technical and deal with machine learning, but being predictive also often comes down to having smart people—who know the company’s purpose and audience well—brainstorm a variety of scenarios and users’ possible responses/desires.
6. Prepare for managing complex user input
One of the biggest complaints about travel bots is that many of them cannot handle complex input. Some bots require you to enter each part of your query as a separate thread of the conversation. Ideally, the bots that travelers will actually use again and again will be able to handle complex input—for example, a user should be able to type, “I want a one-way ticket to Rome, Italy from JFK airport on January 17, 2017,” and the bot should be able to handle the requests all at once, rather than requiring the user to input the destination city in one chat bubble; the date in another; the one-way request in another…and so forth.
Want an example? I tried the above search with the Skyscanner Facebook Messenger bot. At first, it couldn’t understand that I want to go to Rome and made me type a specific airport in Rome. From there, it was able to figure out that I was heading to Rome on January 17th, but it still didn’t pick up my departure airport, JFK.
Starting from scratch, I had to input each element separately to actually accomplish my goal, as pictured below:
To resolve these concerns, you need to build in edge cases, as Carylyne Chan argues in “Why Does Your Chatbot Suck?”. According to Chan, your bot should be able to handle:
- Multiple questions. Users might want to ask all in one request, “Can I change planes in Charlotte? Is that more expensive?”
- Complex questions. Troubleshooting questions are difficult for bots. Chan says that “there needs to be an inbuilt decision tree that captures a range of potential solutions, and use the chatbot’s natural language understanding engine to figure out which steps the customer has already tried, and then filter them to the right direction.”
- Insults. Users may hurl insults if they get frustrated. Plan ahead by creating responses that handle insults in a positive manner and redirect travelers to other resources when possible.
- Chatter. Users may ask questions, such as “Where were you born?” that you simply haven’t planned for. As Chan notes, these kinds of questions are a great opportunity to display your brand personality and add in facts that might entertain your users.
While these tips start to veer us toward the “technical nitty gritty” we promised to leave to the tech experts, careful planning and user testing can be conducted by anyone in the company. Have someone in the company try out competitor bots and note their strengths and weaknesses in terms of how they handle user input. Ask them to brainstorm random questions users might ask or insults they could hurl, and ask them to create a list of tips for your developer during the planning process.
7. Have an exit strategy
It’s clear we’re not yet at the point where a travel bot can handle every situation as well as a human. (Although, sometimes they can. Ever had a customer service rep get sassy with you? Bots may not be smart, but they’re also typically not rude.) Because bots cannot handle all of the tasks we might want them to, your bot needs to be able to understand when users are getting frustrated or when it does not have an answer, and the bot then needs to be able to connect the user to other resources for resolving an issue.
So, have a clear exit strategy. Should your bot admit that it doesn’t understand users? Matt Schlitt conducted an informal Facebook poll in a popular Facebook Chatbots group, asking, “What should a bot do when it receives a message it doesn’t understand?” Ninety-eight percent of the 193 respondents said that yes, the bot should say that it doesn’t understand.
Yet, your bot needs to take things a step further. If it doesn’t understand, figure out how to get users the information they need. Programs like Wordhop have the potential to solve this problem. With Wordhop, you can set alerts for when users are having difficulties with the bot. Wordhop monitors your bot, alerts you when there are problems, and then allows humans to hop in and intervene.
Before you use a more complex system like Wordhop, though, there are little ways you can create an exit strategy or redirect customers. Josh Barkin, the CEO and CoFounder of Wordhop recommends creating text that refocuses users.
If the bot does not understand your travel request, create a more general statement, such as “I’ve never heard that before,” and then refocus the user with a reminder of the purpose of the contact: “Would you like to book your hotel?” With any responses to misunderstandings, consider at least offering a button with a link to the customer support website or listing a phone number. Ultimately, if you can’t redirect users’ focus, you need humans to intervene.
Conversational U.I.: A little personality, a little intelligence, and a lot of potential
The bots that solve real problems, have personalities that reflect the company’s brand, and can handle complex user input are likely to succeed. Bots have a lot of potential to revolutionize the travel industry–and it’s not all up to the people building the bots to make this happen. Your company’s stronger writers and best critical thinkers can help you create text for the bot that will drive user engagement and even a user guide that helps users understand how to interact with the bot.
If you want to learn more about the different types of travel bots in the market right now, check out our Ultimate Travel Bot List.