Sorry, I can’t help with that: Content strategy for chatbots

It feels like everywhere you look, someone is spinning up a chatbot team. They’re captivating executives with stories of how they’ll use AI and machine learning to deliver cost savings and better customer service.

Despite all the excitement, building a good one is harder than it looks. Figuring out what types of problems your chatbot can help with is the most important part of building one, but teams often rush through that process.

Learn how to make sure your chatbot meets user and business needs. Don’t build an expensive FAQ.

In this session, you'll:
- Understand how to approach research and discovery for chatbots, based on experience leading conversation design teams for a large insurance company.
- Get frameworks and templates that will help you evaluate whether a chatbot can help in the first place.
- Learn methods for creating a chatbot strategy that will lead to a useful, usable product.

Chatbot evaluator: https://docs.google.com/spreadsheets/d/1shjUnx-xvGZ6vara8yq3B8wSTS70GVfdb65ORNXWd3w/edit?usp=sharing

Transcript

1. "Sorry, I can’t help with that." Content strategy for chatbots Michael J. Metts | he/him | Co-author, Writing is Designing | @mjmetts | mjmetts.com
2. Where do chatbots come from?
3. People love AI. Business people love being able to say they work on AI.
4. BotBuilder 12:42 83% Hi, I'm an automated assistant. Why would you like to build a bot today? Amazon makes Alexa and Amazon is very successful! I'm sorry. It doesn't work like that.
5. 12:42 83% BotBuilder Hi, I'm an automated assistant. Why would you like to build a bot today? Google makes conversational AI and Google is very successful! I'm sorry, I don't understand. You can say things like "cost savings". 12:42 83%
6. 12:42 83% BotBuilder Hi, I'm an automated assistant. Why would you like to build a bot today? We need to meet consumer expectations. I'm sorry, I can't help with that. Would you like to take a short survey about your experience? 12:42 83%
7. 12:42 83% BotBuilder Hi, I'm an automated assistant. Why would you like to build a bot today? Lots of these customer service tasks seem easy. Can't we just automate them? I can help with that. I'll need to ask you a few questions to get started. 12:42 83%
8. “Content strategy plans for the creation, publication, and governance of useful, usable content.” - Kristina Halvorson Source: https://alistapart.com/article/thedisciplineofcontentstrategy/
9. "I'm having some trouble right now." Deciding whether a chatbot will help.
10. What are our goals? Writing and design decisions are inseparable from business goals.
11. Customer service Types of chatbots Example: T-Mobile's Virtual Assistant • Responds to any kind of natural language • Helps with customer service requests • Can escalate to a human
12. Customer service Types of chatbots Possible goals: • Increase speed of service • Reduce contact center costs
13. Distinct product Types of chatbots Example: Capital One's Eno • Responds to natural language text messages • Heavily marketed • Self-contained (doesn't escalate to human support via chat)
14. Distinct product Types of chatbots Possible goals: • Portray Capital One as a technology leader • Increase number of interactions with customers
15. Routing and sales Types of chatbots Example: LiveEngage automated assistant • Gives the user pre-defined options to choose from • Goal is to connect you with a human • Offers multiple channels
16. Routing and sales Types of chatbots Possible goals: • Lead generation and retention • Revenue targets
17. Is it going to be useful? “Content strategy plans for the creation, publication, and governance of useful, usable content.” - Kristina Halvorson I wasn’t kidding about content strategy! Source: https://alistapart.com/article/thedisciplineofcontentstrategy/
18. How does this fit into other channels? Should this be handled on a website or mobile app? Should it be handled by humans via phone, email, or social media?
19. Pick an evaluation method. Focus on the user need. Source: https://www.gartner.com/en/documents/3978596/customer-service-issues-to-channel-mapping-tool
20. What could go wrong? Source: May-Li Khoe on Twitter - https://twitter.com/mayli/status/1305952098996097025?s=21
21. What could go wrong? Source: Ayodele Odubela on Twitter - https://twitter.com/DataSciBae/status/1308839841606533122
22. Ayodele Odubela’s upcoming book Book details: dazzlingdata.science/uncovering-bias-in-ml
23. Meet Jessica Zhang Koehler • Design leader • PhD in Cultural Theory of Technical Communication • Experience helping business navigate emerging technologies
24. So, if you're in the deciding stage, ask • What are our goals? • Is it going to be useful? • How does it fit into other channels? • What could go wrong?
25. "Here are some things I can help with." Figuring out what your chatbot should do.
26. What problem are we solving? Get involved in the problem space. Your perspective as a writer is critical.
27. Example: Customer service bot for an electric utility. X X Hi, I'm ElectricCo’s automated assistant. How can I help you today? Chat Support • Responds to natural language • Goal is cost savings and increased service speed • Available when customers are signed in to their accounts
28. What do people need? • Chat logs • Call recordings • Support emails and social media messages • Forums, reviews, and blog posts • Help content analytics How are they describing it? Protect users’ privacy when gathering data
29. “I need to reset my password.” “We just moved to a new house and I need to change the service address but I can’t log in.” “CAN YOU PLEASE RESET MY LOGIN?????” “can’t log on to my account” “I need a copy of my latest bill but I can’t figure out how to sign in.”
30. Change billing due date Route to correct department Get help logging in Move service to a new address
31. How will we make decisions? Create criteria for evaluating chatbot features as a team.
32. Alexa guidelines Feature evaluation Voice-specific, but contains principles that apply to chat. • Is specifically about Amazon’s technology • Checklist gives you a way to check in Source: https://developer.amazon.com/en-US/docs/alexa/alexa-design/ checklists.html
33. Google conversation design process Feature evaluation Designed to help teams understand whether conversation is appropriate. • Ties questions the team can discuss to benefits of conversational UI • Applies to chat and voice Source: https://designguidelines.withgoogle.com/conversation/conversation- design-process/is-conversation-the-right-fit.html#is-conversation-the-right- fit-take-the-quiz
34. Build your own evaluator Get this at go.mjmetts.com/chatbotevaluator
35. What’s the bot’s personality? Heavy personification No personification Capital One’s Eno Customer service for a utility
36. Where should we start? Get the Mural template: mural.co/templates/importance-difficulty-matrix
37. Impact Difficulty Change billing due date Route to correct department Get help logging in Move service to a new address
38. Impact Difficulty Change billing due date Route to correct department Get help logging in Move service to a new address
39. Strategic High value Short term Nice to have Impact Difficulty Change billing due date Route to correct department Get help logging in Move service to a new address 1 2 3
40. Meet Rebecca Evanhoe • Conversation designer • Has worked with conversational technology since 2011 • Co-authoring a book with Diana Diebel called ‘Conversations with Things’ coming in 2021 from Rosenfeld Media Book details: rosenfeldmedia.com/books/conversations-with-things/
41. So, if you're in the building stage, ask • What problem are we solving? • What do people need? • How will we make decisions? • What’s the bot’s personality? • Where should we start?
42. Thank you! • Check out ‘Writing is Designing’ by Andy Welfle and myself at writingisdesigning.com • See upcoming remote workshops at mjmetts.com