Designing Conversations: How I learned to stop worrying and embraced our new chatbot overlords

This talk was originally presented at Carbon Five's May 2016 Talk Night in Santa Monica, CA. It's about designing conversational UI in the form of chatbots. The first part is an overview of the current landscape of bot platforms, from SMS to native apps. I review how each platform affects UX and product strategy, and I offer advice on choosing which platform is best to use when launching your bot. The second half covers the best practices of designing a conversational UI – from making sure your bot is asking the right questions, to getting your user to answer in the right way, to organizing your chat paths.

Transcript

1. Designing conversations How I learned to stop worrying and embraced our new chatbot overlords Justin Barr Young @theJayBeeWhy Carbon Five Talk Night, May 18 2016
2. Agenda 1. Defining Conversational UI and Bots 2. Bot platforms 3. Tips for designing bot conversations
3. Conversational UI: Interacting with a computer through conversation, versus a Graphical User Interface (GUI). Chatbot: A computer program designed to simulate conversation with human users through chat. Bots are a form of conversational UI.
4. Slack
5. Carbon Five Slackbots
6. Platforms Current bot landscape breaks down into 3 separate strains: 1. SMS-based bots 2. Over the Top (OTT) bots, built on existing platforms 3. Dedicated native apps
7. 1. SMS-based bots Bots like Assist and Prompt communicate through texting Largest app platform pre-installed on every device Light-weight, low-cost way to launch and get feedback No app store review! Limited types of content, no extensibility
8. Assist
9. 2. Over the Top (OTT) bots OTT bots use platforms like Facebook Messenger or WhatsApp Use a larger array of content like photos, GIFs, stickers Users can easily find and befriend your bot Platforms are becoming more robust with GUI features and toolkits for developers Platforms have rules and limits you have to abide by
10. Uber for Messenger
11. 3. Dedicated native apps Bots like Operator or Quartz live entirely in their own app Freedom to use any content you want in messages App “wrapper” can organize conversation threads or incorporate traditional GUI features Opens the door for “atomic applications” High implementation costs
12. Operator
13. GoButler
14. Atomic Apps Each message is its own mini application Use text, media, and interface components in conversation Leverages conversation and traditional GUI for tasks when appropriate
15. Operator
16. Introducing the bot When the user first talks to the bot: Be clear about what the bot does and doesn’t do Don’t be afraid to to glamorize the bot’s mission! Offer a “help” command
17. Hey! I’m BurritoBot. I can order burritos, tacos, and bowls for you! BurritoBot If you ever need help, just say “help” and I’ll offer a list of all the commands you can use.
18. Asking questions Do ask leading questions Don’t ask open ended questions Demonstrate the type of answer the bot is looking for If the number of possible answers is small, just list them
19. What do you want to order? 1. Burrito 2. Taco 3. Bowl BurritoBot You can say the name of the item or its number, like “1” or “burrito”.
20. Break things down “Talk” in multiple paragraphs at a comfortable cadence Ask complicated questions as separate, discrete questions Confirm an answer before asking the next question
21. Great! Let’s order your burrito. 1. White 2. Wheat 3. Lettuce BurritoBot First, what kind of tortilla do you want?
22. Operator
23. Quartz
24. Be flexible Consider different ways to say something. Consider the different ways a user might semantically answer a question Consider how a user might format an answer Offer syntax examples when asking a question
25. Assist
26. Task paths Without visuals, user flows might seem confusing. Over-communicate task paths When a user wants to switch paths, be explicit about pausing or ending the current path. Start a new path after a few hours or a day of no response
27. Operator
28. Modals Sometimes a bot has to insist that a user does something Don’t repeat the same question over and over – ask it in a variety of ways for the user
29. You want a burrito with a wheat tortilla, pinto beans, carne asada, cheese, medium salsa, and guacamole. BurritoBot Is that right? Are you ready to order your burrito? I can’t order your burrito until you confirm the order above. Is that order correct? (Answer “Y” or “N”)
30. Responding to the bot Terminal commands are hard to remember, but natural language is inefficient Create short, memorable, and vernacular commands
31. Slack
32. Input types Typing in natural language can take a while… Give users an easy way to references items or answers Assign list items a number or letter Break down questions into Y / N answers Think about the user’s device
33. Assist
34. Fill in the _____ Consider a “madlibs” model to reduce answer options and move conversations along Give users meaningful choices, not false choices Make sure the users have agency in the conversation
35. We have two vegetarian options: “Veggie” has grilled zucchinis, onions, and corn. “Soyrizo” is soy-based chorizo with medium spice. BurritoBot What sounds good? Soyrizo, please Gimme the Veggie
36. Good example: Lifeline
37. Bad example: Quartz
38. Terrible example: Lark
39. Organizing chat threads Scenarios Script Responses
40. Conclusion Pick the right platform: SMS, OTT, or dedicated app Leverage both conversation and GUI components (when available) Break tasks down and ask leading questions Use the right input type for your device and context
41. Thanks! Justin Barr Young @theJayBeeWhy Carbon Five Talk Night, May 18 2016