Chatbot Development, Conversational Marketing

10 Dos and Don’ts of Chatbot Conversation Design

Effective chatbot conversation design builds trust and creates a successful, enjoyable user experience. Here are best practices to keep in mind when designing a chatbot.

Effective conversation design is what helps to create a successful, enjoyable user experience. It shows the user what is expected of them, keeps them on track toward the end goal, and helps them have fun so they want to come back and engage more. Most importantly, conversation design can be used to build trust. When a user enters a chatbot, they have little trust in the experience because they don’t know yet if the chatbot can resolve their issue, help them learn something, or accomplish whatever goals they have. By guiding the user and delivering on the expectations set, they will not only accomplish their goals but will also trust that your business—and your chatbot—has their back.

The purpose of conversation design is to make the experience feel more invisible. The benefit of chatbots is to automate conversations at scale and offload human tasks, but the user should still feel like they are having a one-to-one conversation with your business.

11 best practices to keep in mind when designing a chatbot

Let’s start with all of the things you should do.

1. DO keep it simple. The user does not need irrelevant details, overly wordy sentences, or unnecessary questions. Keep the list of asks for the user to the minimum amount of qualification questions necessary to deliver a result.

2. DO overexplain. Counter to the last point, it’s crucial in the beginning stages of the conversation to overexplain what the user needs to do. Unlike face-to-face conversations and ones with a live agent, the user will not assume what the chatbot means, because the bot is not human. If you are working on a rule-based or on-script conversation, the chatbot only “knows” the pathways that have been built and will not be able to offer more context to a user if they do not understand. In other words, you get one shot to communicate with the user, so it’s important to explain in the clearest way exactly what you mean.

3. DO lead by example. An easy way to improve the quality of a conversation is by telling and showing the user exactly what you want them to do—for example, using directions like, “Select from the options below,” and “Tap yes to continue.” You can also use videos to demonstrate more advanced actions, like uploading a photo or locating account information.

Memora Health COVID-19 SMS chatbot

4. DO use confirmation. An important part of earning trust is letting the user know the chatbot understands what they’re saying. Whether the bot is repeating their inputs directly (retyping their phone number and saying, “Got it”), using state-aware personalization (like saying their name after they provide it to you or remembering their favorite product), or letting them know there are only two more questions, confirming the details and purpose will build trust.

Delta Virtual Assistant

5. DO make hard things easier with structured, on-script questions. The best way to minimize errors, failure, and drop-off from the user is to decrease their need to type inputs. Quick replies should be used whenever possible. If there is a lengthier or more complicated question, break this down into several parts and offer options.

Mav class action qualification SMS chatbot

And now for a few don’ts—the things you’ll want to avoid

6. DON’T leave them hanging. A chatbot should not use the same casual filler words that many humans do in conversation, like starting a sentence with “so” and ending it with “right?” when a question is not being asked. Questions and statements should always be final. At the end of the conversation, the bot should have the last word, confirming to the user that the conversation is over and there is nothing more expected of them.

7. DON’T be too chatty. A chatbot does not need to ask how you are doing or know how your weekend was. The main reason a user opts to communicate with a chatbot instead of a human is because they want to reach their goal faster. Conversation for the sake of conversation should be avoided.

8. DON’T be too open. Allowing for open-ended intents from the beginning and expecting the user to figure out what they are supposed to do is a recipe for chatbot failure. Always state the obvious and guide the user where you want them to go. When a chatbot needs to be open, like to ask for a user’s email address or loyalty number, provide a formatted example so they know exactly how to reply.

9. DON’T fail without a plan. The fail state, or catch-all, is extremely important. When the natural language processing (NLP), the tech, the dialog, or the bounds of conversation fail, the chatbot should step in and save the day. Giving the user options to get back on track and continue towards their goal quickly creates an excellent user experience.

Catch-all example from Haven Life Insurance

And lastly, 10. DO follow up, but…

DON’T overwhelm. Sometimes life gets in the way and a user needs to come back later to complete a more complicated conversation. Use notifications wisely, and remind a user that you’ve saved their progress and they are still able to complete the goal they set out to. When sending a notification, ask a question that the user can easily act on (a “yes or no” question) to keep the user active within your 24+1 window (on Messenger).

Sephora Messenger chatbot

As you continue to build out the experience, you can always add options for the user to pick up the conversation at a later point through reminders, like, “Want us to message you again when a new eyeshadow palette comes out we think you might like?” This encourages a user to return and also illustrates that value will be provided to them through this chatbot in the future.

What’s next? Continuous improvement

In addition to updating your conversational dialog as the chatbot becomes more advanced, you should also consider the performance and result from user testing and iterate on the dialog as it's released to more users.

When you’re writing a chatbot script, the experience is flat. It is easy to
predict where the conversation will lead and to see all parts of it in a logical way. In
a live chatbot, the conversation is 3D. The pathways are not displayed for the user, and unlike the designer, the user does not already know where the conversation will lead. This is why testing and analyzing performance metrics and error logs will help you create a better chatbot and a better user experience.

Image via UX Writers Collective