[Design Dive] Chatbot Interaction: Design & Content Conference 2019

[Design Dive] Chatbot Interaction: Design & Content Conference 2019
Mako Mizuno

Mako Mizuno

Nov 8, 2019

This July, Convergence had the opportunity to be a part of the Design & Content Conference 2019 at the SFU Goldcorp Center in Vancouver. We had a chance to attend this amazing event and now share some of our experience and insights in this new Design Dive series post.


What is DCC?


DCC, or the Design & Content Conference, is an annual event encompassing new design and digital content trends, and this year marked its fifth edition. DCC provides an opportunity to hear from industry leaders about crafting experiences, telling stories and shaping the future of the web. This year the conference received roughly 300 attendees, most of whom were designers or content strategists. During the 3-day event, 15 speakers got on stage to share their insights and stories about all aspects of the industry. 2019’s main sponsors were Telus, AbstractMedia temple, and BrainStation.


In this edition of Design Dive we’ll start by sharing our insights on Melanie Seibert’s talk about designing content for conversational interfaces and some tips on improving human communication with personal assistants and chat bots.


Content Design for the Conversational Interface by Melanie Seibert


Melanie is a senior content strategist at WillowTree, a US-based mobile app & web application development company. As recounted by Melanie, a December 2018 Juniper Research study estimated that 2.5 billion digital voice assistants were being used around the world by end consumers. That total comes from both smart speakers and smart devices, and it is worth noting that most devices nowadays – Android and iOS phones, for example – have digital voice assistants integrated into their native platforms. This helped popularize the adoption and usage of voice assistants over the past few years and decreased the negative stigma surrounding devices “hearing” their conversations.


Three possible formats for conversational interfaces were laid out. They are: textvoice, and multimodal.


Text modals are exemplified by platforms such as Facebook Messenger or Slack, and have been a part of our daily digital communications since the beginning.


Voice platforms that we’re most familiar with include AlexaGoogle HomeSiri, and Cortana. These are newer and more direct forms for humans to integrate with the digital world.


Finally, multimodal involves both text and voice interfaces. For example, Alexa – Amazon’s interface – is coming out with new Echo devices that have screens on them, in order to display text or video messages, and Google Home is following suit, with the new Google Nest product line released earlier this year. These multimodal interfaces are becoming increasingly common since they mix the best of two modals, reducing their respective limitations.


But why would multimodal interfaces be better than just text or voice? Due to the way humans communicate – depending on the activity – some modes of communication are more efficient than others. It can be an efficiency gain in either receiving the information, or in transmitting it.


Humans type at about an average of 40 words per minute (WPM), but we speak a request to a system at about 130 WPM.



When it comes to absorbing a response, we read at about an average of 300 WPM – but only listen at an average of 130 words per minute. This means the most efficient combination of information exchange methods is speaking the request and reading the response.



In her talk, Melanie also introduced the seven steps involved in the design process for creating a good chatbot.


The 7 Steps of the Chatbot Design Process


1. Start with data To create a chatbot, always start with data. For instance, in a software development company, the initial data set might be customer service logs. Those logs would be composed of lists of questions asked by users/clients and would serve as a foundation upon which to start developing a useful chatbot.


2. Write a user story for each intent “As a certain type of user, I want to do something, so that I can achieve this goal“. For example: as a product user, I want to know whether I can replace this handle on my bag so that I can avoid having to buy a whole new product in case the handle breaks. Once you have basic user stories you can start to prototype conversations.



3. Prototype the conversation Start by prototyping the hypothetical chatbot conversations on Slack. Melanie nicknamed this “Wizard of Oz role-playing” because one person plays the chatbot, and that person is essentially the “man behind the curtain” – so to speak. The other person in this role-playing conversation is acting as the user. This process helps to discover how the conversation may branch and unfold. The Slack team itself, it was noted, takes a similar approach when testing their Slackbot.


4. Design the flow There are multiple ways to design your workflow, but one of the best ways to do it is called the golden path – also referred to as the happy path. This means defining the primary intent or goal of your chatbot and designing the workflow around it. Designing the ways by which your chatbot can be invoked is also important. Different platforms do it in different ways; for example, Google Assistant is invoked in a different way than Alexa, but they both have the same primary intent.


Another key aspect to have in mind while designing your flow is your error path. What happens when the chatbot does not understand the command? How many times should it be allowed to get it wrong before escalating it to an actual human?


Melanie mentions the importance of integrating with other services. If a chatbot is going to enable users to buy something, call someone, etc., then it’s going to be quite important to plan out the seamless integration with third party platforms and services.


5. Train the bot Bots don’t understand us very well – we have to tell them what we mean. A practical method of gathering training data is to do so through user interviews. Another way: jump on Slack. Create a Slack group with four or five people, and brainstorm different ways of saying the same thing.


6. Design the personality This is the fun part of the chatbot design process: thinking about the emotion you want to convey to your users. One way to decide upon what personality would best fit your company is explained in Margot Bloomstein’s messaging architecture methodology. Margot wrote a book called Content Strategy at Work, in which she demonstrates how to use a card sorting method to determine a brand personality. This same methodology can be used to design a chatbot’s personality, thus guaranteeing a concise brand image catered to clients’ expectations.


Let’s say we chose the following five attributes for a chatbot’s personality. You can think about “personality” as different tones in the chatbot’s voice.



In this example, there are different tones for the golden path, as well as tones for errors, information entries and exits.


If the flow is following the golden path, the chatbot’s tone will be a little bit more motivated, optimistic and less humorous because people are trying to get their questions answered.


In case the chatbot hits an error the path, in this example the tone becomes even less humorous, because normally people don’t like to joke around when they’re having problems; they just want to solve their issues and get support.


For entries and exits, however, a little more humour could be added so that users will leave your platform with a positive impression and memories of a fun experience.


7. Create and refine scripts


A great way to refine your scripts and manage content density, according to Alexa guidelines, is by using what is known as “landmarking”. What these guidelines mean by landmarking is the action of repeating the user’s request back to them. This can be implemented in all modals; for example, a calendar management chatbot may show you what meetings you have planned for the day, and an assistant may reply to your voice command by confirming that it will execute said action. A practical example can be seen below:


The user might say, “What’s happening at 10 am?”



A good response from the chatbot might be “At 10 am, you have the weekly status meeting with Rachel and Natasha”. This is considered a good response because the chatbot is confirming the request at the same time that it is providing information by repeating back what the user said. If it said, “You have a weekly status meeting with Rachel and Natasha at 9 am.”, the user would have to wait through the entire sentence to find out that it actually didn’t understand what you were saying, causing the experience to be less efficient.



Make Your Own Chatbot!


It is possible to create a chatbot without needing to write any code. Google Assistant provides you with ready-to-go templates, while Alexa Skills offers Blueprints, which allows you to fill out a spreadsheet and create your very own chatbot experience. Of course, out-of-the-box services like these do have some limitations, but they are a great way to start learning about chatbot development and functionality.


If you are interested in learning more about chatbots – including exploring how AI chatbots may be integrated into an existing platform – please reach out to us.


Stay tuned for our next Design Dive post, in which we will visit Sophia V Prater’s “UX For Lizard Brains“.