"Happy" and "Sad" Conversational Flows serve as a point of reference to check the quality of chatbot answers. By predefining the expected chatbot behavior for understanding (and not understanding) end user intents, you can focus on a clear goal for the quality of chatbot responses.
As a meta-guideline, always prioritize creating value for chatbot visitors.
This article covers the following:
- Happy Conversational Flows
- Sad Conversational Flows
- Testing Happy and Sad Conversational Flows
- Maintenance of Happy and Sad Conversational Flows
- Practical advice
Happy Conversational Flows
Happy Conversational Flows outline a conversation on a specific topic during which the bot understands the user's intent.
Create a Happy Conversational Flow for any topic the chatbot has an answer for. You expect Happy Conversational Flows to help, inspire, or create value for the chatbot visitor.
In other words, if the chatbot visitors are likely to answer "Yes" to the bot's question of if the answer it provided was helpful, you call the conversation a Happy Conversational Flow.
Here, you can see an example:
Illustration of a Happy Conversational Flow
|Visitor||"I'd like to know how you handle refunds..."|
|Chatbot||"Happy to help with your inquiry about our refund policy! You have 30 days from the date of delivery to return the item(s)."||Understands AI Intent Refund|
|Visitor||"Thank you! How much does it cost?"|
|Chatbot||"Are you referring to the cost of sending the items back or to another price?"||Understands AI Intent Price|
|Visitor||"The refund, of course 😃"|
|Chatbot||"Right. We have a free return policy 🙂"||Understands AI Intent Yes|
|Chatbot||"Glad I could help!"||Understands AI Intent "Thank you", i.e., answer was helpful|
Sad Conversational Flows
Sad Conversational Flows are those events when it is anticipated that the chatbot does not understand the end user's intent. In those scenarios, the expected chatbot behavior is to fail to answer accurately at the time of its first response.
It is essential to understand that "sad" chatbot responses must be planned and accounted for. The chatbot must always still have a proper response when it does not know the answer straight away.
You can find four examples of Sad Conversational Flows below:
Illustration of a Sad Conversational Flow - Fallback
|Visito||"I'd like to who is handling your warehouse in Moskau."|
|Chatbot||"Sorry, I'm not sure I've understood you correctly. Can you please rephrase?"||Fallback Module 1 of 2|
|Visitor||"Warehouse operator Moskau."|
|Chatbot||"Unfortunately, I'm still having trouble. Would you like to leave us a message, and we will come back to you as soon as possible?"||Fallback Module 2 of 2|
|Chatbot||"Thank you for your understanding! Can I send this message on your behalf? 👇
Message: Warehouse operator Moskau"
|Chatbot has saved end user's message as a Custom Variable|
|Visitor||"Yes, that's fine."|
|Chatbot||"Great. I've sent your inquiry. Is there anything else I can help you with?"||Understands AI Intent Yes|
Illustration of a Sad Conversational Flow - Non-Supported Language
|Chatbot||"Hi, there! It's nice that you're here. How can I help you?"|
|Visitor||"Ciao. Mi piace ordinare Pizza Magaritha. "|
|Chatbot||"I'm sorry, but I don't know Italian yet. Would you like to leave a message for my human colleagues, or can we proceed in English?"||Automatic Language Detection (Italian)|
|Visitor||"Ah, sorry. Yeah. Let's write in English."|
|Chatbot||"Thanks for your understanding. Can you please ask me again in English?"||Understands AI Intent Confirmation|
Illustration of a Sad Conversational Flow - Too Long End User Message
|Visitor||"I'd like to order Pizza Margeritha. But please, don't add any garlic. That's not following the recipe as I know it from my Neapolitan grandma 😄 I'd also like to have it cut. Make sure the pizza is still warm when I pick it up. Oh, and I'd like to have two pairs of cutlery. Do you have silver cutlery? Thank you 😇"|
|Chatbot||"I'm sorry, but I cannot understand very long messages yet. Would you like to order Pizza?"||Automatic Word Count (66 words); Key word identification ("Pizza")|
|Chatbot||"Super. Which kind of pizza would you like?"||Understands AI Intent Yes|
Illustration of a Sad Conversational Flow - Technical Issue
|Chatbot||"Unfortunately, there was a technical issue. I apologize for the inconvenience! Can you please email email@example.com? Once my human colleagues have received your message, they'll reach out to you as soon as possible."||See that the status of a Webhook Response is not beginning with "2," indicating a Webhook failure|
|Visitor||"Sure, will do."|
|Chatbot||"Thank you for your understanding!"||Understands AI Intent Confirmation|
Testing Happy and Sad Conversational Flows
Outlining Happy and Sad Conversational Flows helps with testing chatbots, as you can compare the chatbot's responses against the Happy and Sad Conversational Flow maps.
Furthermore, those conversational maps can be used for automized testing. If you intend to check the chatbot's technical setup regularly, automized testing allows you to compare a conversational flow's outline with the present chatbot's answers.
Maintenance of Happy and Sad Conversational Flows
It should be clear by now that Happy Conversational Flows provide the opportunity to map value-creating chatbot answers. Organizational value proposition changes over time, so Happy Conversational Flows must be maintained flexibly, too.
Note the importance of constantly increasing expectations of what can be defined as Happy Conversational Flows.
The longer a chatbot is live, the more data is available, and the more opportunities arise to make the chatbot even smarter and more human-like.
Ensure that the expected bot behavior for Sad Conversational Flows changes by raising the expectations toward Happy Conversational Flows. Some Sad Conversational Flows might become redundant.
We want to underline the importance of designing both Happy and Sad Conversational Flows. It can be helpful to start with the Happy Conversational Flows and only continue with the "Sad" part once you are satisfied with the "Happy" ones.
It is also beneficial to determine a ratio of how often you expect a Sad Conversational Flow to be triggered compared to Happy Flows during testing or when the bot is live. If Sad Flows outweigh the Happy Flows, improve the chatbot's configuration.
Most frequently, the following three aspects do most of the improvement work:
- Improve the chatbot's AI.
- Check if the Happy Flows contain Modules with incorrectly set up Module Connections.
- Set up new integrations to ease the chatbot's work.
Try the following to evaluate the quality of your chatbot's responses.
Ask, "Was the answer to this question helpful?" at the end of Happy Conversational Flows. If users reply "no", consider redesigning the flow.