After my last article on what you need to have before developing a chatbot, this second post (in a three-part series) discusses 5 key things you must have within your chatbot.
At HiJiffy we offer hotels a conversational booking engine within their Facebook Messenger branded page. Since the booking process happens in a messaging app, it’s a new channel where guests can check prices & availability and book their stays. We already launched chatbots for more than 100 hotels where we learned some tricks on how to increase the engagement and bookings through Messenger.
1. Offer the key feature of your chatbot right away
When launching a chatbot for the first time, it’s difficult to think about all the different combinations of words people can use to say something as simple as “I would like to book a room”. Until we have more evolved Natural Language Processing (NLP) systems, the wiser thing to do is to start with pre-built responses, which provides useful shortcuts to tireless typing.
Opting for this approach, instead of having an open chatbot with no decision trees shortcuts, users immediately know that they are not talking with a human agent — setting the expectations off right since the beginning. As your NLP based model evolves and learns with users interactions, your chatbot will get smarter each day to reply to frequently asked questions in the near future.
Best Practice
You should think about the primary goal of your chatbot and give that option right after someone starts a dialogue with you. When thinking about a hotel, the most important feature would be having direct bookings. We recommend to start the flow by dividing the users into two major groups: 1) people who want to check prices and availability and 2) people with other purposes.
2. Make a seamless hand-off to a human agent when necessary
One of the most important things to realise when launching a chatbot is that it will not solve each and every problem your users may have. This happens due to limitations of the underlying NLP technology (the system is not yet fully trained), but also because sometimes the chatbot simply cannot have the expected answer — imagine someone asking if there’s something left in the lost and found after staying in a hotel.
In this cases, you need to have a human agent prepared to step in the conversation and the chatbot to human handoff needs to be a smooth transition to ensure the user knows he’s no longer talking with a chatbot.
Best Practice
The chatbot needs to acknowledge to the user that has run out of options and cannot help any further. Instead of escalating to a human agent immediately, you should ask the user to do it, as some people feel uncomfortable bothering a human agent and prefer to try a few different sentences. If the user replies yes, handoff to a human agent and let the user know someone will reply soon.
3. Make it visual – use emojis! 👍
When creating a chatbot, you should create a user experience that is easy, pleasant and natural for users to be able to achieve a given task. It’s important not to forget that all the old user experience rules still apply. For example, you must be consistent among all flow, the conversation must be enjoyable and the call to actions should be clear.
One simple way to make your chatbot more enjoyable and clear is by adding emojis. Whether you want to use emojis extensively or not depends on the overall communication strategy of your business, as emojis reflect how you want your communication to be perceived by users.
Best Practice
One of the best places to add emojis is in UI elements such as menus or buttons. Whenever there’s the need to click a button add an emoji to reinforce what it is about — this may increase click-through rates up to 50%.
4. Webviews will greatly improve your chatbot
The first chatbots were composed mainly of bubble chains with simple text. This was OK for simple use cases, but when someone wants to book a room in a hotel this interface isn’t enough. As chatbots try to solve more complex use cases it became clear that bubble chains with simple text cannot be the answer to everything.
There’s more to bots than NLP and artificial intelligence, the solution relies too on external web pages. For more complex use cases, a webview opening inside Messenger will be more engaging, showing more structured data and additional scrolling if needed.
Best Practices
There are two cases that are best solved by webviews. When there are multiple pieces of data to enter in the chatbot (such as check-in and check-out dates or number of adults and children), it’s far easier to change these data in a form than in a conversation. The second use case is when there’s a need for richer browsing experiences, like seeing room photos, details and amenities.
5. Re-engage with users the right way
By now you should now that travellers visit 38 websites before booking a vacation. This statistic needs to be revised: nowadays travellers visit 37 websites and your chatbot. On the day of the booking itself, travellers visit about 16 websites overall.
It’s very likely that users are using your chatbot in the process of deciding the best hotel to stay when they are still comparing prices and destinations. This way, it’s very unlikely they will book right away. The great thing about a chatbot is that, contrary to a website, you have the possibility to re-engage with the user some hours or even days after the first search — without any costs.
Best Practice
Send a notification to your users remembering them they can book with you, and show a clear call to action so that they can access the most updated prices easily. The best time to send a notification depends on the kind of property you have. If you have a luxury vacation resort you can send it a few days later, as people will take some time to decide their vacations. If you have an urban hotel targeting business travelers, you should opt to notify them just a few hours later.
This article originally appeared on Pedro Gonçalves’s Medium account here and is reproduced with his permission.