Generative AI Discussions in Scripts

Overview

Partners with access to this feature can add a special action in Scripts that puts contacts into a generative AI discussion. While contacts are in this state, all incoming messages are handled by a generative AI prompt (optionally using Knowledge Sources). This allows for fluid, unstructured conversations where learners are free to ask questions outside of the normal branching logic of a Script, but without deviating from the relevant topic.

Gen-AI Discussion Configuration

1. Click + Add after a simple message and select "Enter gen-AI discussion" from the Actions list.
   • 
2. Then select a built-in prompt, or select "Custom Prompt" (see below).
   • 
3. For a Custom Prompt:
   • Enter the AI instructions into the textbox. If you are unfamiliar with prompt engineering and not sure how to start, contact your Partner Success Manager or the Mainstay Support Team.
   • Select which contextual information to include when generating responses. 
   • 
4. If the built-in prompt uses scraped content, or, for a custom prompt, the "Scraped Content" box is checked, optionally limit the scope to specific Knowledge Sources:
   • Search for one or more Knowledge Sources by Title. Only these sources will be included when providing contextually relevant information to the AI.
   • 
5. The Gen-AI Discussion will continue until one of these conditions is met:
   • The learner replies "exit". (Remember, learners won't know this unless you inform them!)
   • The learner sends a message that indicates an intent to exit, such as "I'm done talking to you" or "I don't want to keep chatting".
   • The learner has sent in X messages (and received X responses).
   • The learner has been in this state for Y minutes.
   • The learner sends in a message that matches a Sensitive Topic and triggers an Escalation.
   • 
6. After the discussion ends, the bot can optionally look through the conversation and attempt to extract one or more data points. This information will be captured as Custom Contact Fields (or Default Fields: First Name, Last Name, Preferred Name, Email, and Phone).
   • 

Conversations

Messages that are delivered during the course of a Gen-AI Discussion will indicate both the Discussion topic and whatever Knowledge Sources were used to generate the replies.

Reporting

In the Campaign Insights tool, Scripts with a Gen-AI Discussion are considered Interactive. In addition to the Overview section showing the number of learners who were active, passive, etc., the Survey Results section will show reporting specifically for the Gen-AI Discussion.

• # Discussions / Sent to #: The report will indicate the Script message that led into the discussion, then the number of learners who received it (Sent to #) vs. the number who engaged (# Discussions). Below that is an indicator of the Discussion topic.
  • 
• Conversation Summaries: Below that is the name of the Gen-AI Discussion Topic. Click View summaries to open a drawer with a conversation summary for each learner who engaged.
  • 
  • 
• Engagement Histogram: Next is a chart indicating how much learners engaged. The y-axis indicates the number of learners in each bucket, while the x-axis indicates the number of messages those learners sent in. Discussions with more engagement show larger bars to the right.
  • 
• Exits: Finally, a donut chart indicates how learners exited the Discussion. (See above.)
  • 

Tips

1. If "Past Messages" is checked, this will include recent messages from before this Script. If unchecked, only messages within this Gen-AI Discussion are considered.
2. Make sure to provide the "exit" information so learners don't feel "stuck" in this state.
3. Instead of a long, continuous Gen-AI Discussion, consider exiting after a shorter timeframe and asking the learner if they'd like to continue. You can use the "Link to another message" feature to create loops.
   • 
4. For custom prompts, always provide explicit instructions for the AI. Best practice is to start with "You are a ___", describing the expected role and personality of the bot, as well as "You are talking to ___", describing the expected recipient in general terms. Be explicit and direct.