Here’s how I would build a simple ticket sentiment workflow.

1. 𝗦𝘁𝗮𝗿𝘁 𝘄𝗶𝘁𝗵 𝗮 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻 𝘁𝗿𝗶𝗴𝗴𝗲𝗿:

Here I have manual enrollment set up, but you could automatically enroll new conversations or updated ones.

2. 𝗚𝗿𝗮𝗯 𝗮𝗹𝗹 𝘁𝗵𝗲 𝗺𝗲𝘀𝘀𝗮𝗴𝗲𝘀 𝗼𝗳 𝘁𝗵𝗲 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻:

There’s no easy way to do this in workflows, but a custom code action suffices nicely.

3. 𝗕𝗿𝗲𝗲𝘇𝗲 𝗔𝗜 𝗧𝗶𝗺𝗲!

I use a simple message ‘Given the following text, output a categorisation from the following “Positive”, “Negative”, “Neutral” or “Other” ’.

The “Other” is our catch all, which will improve the accuracy of the other labels.

4. 𝗔𝗻 𝗶𝗻-𝗮𝗽𝗽 𝗻𝗼𝘁𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝘁𝗼 𝗺𝘆𝘀𝗲𝗹𝗳 - always helpful for debugging.

5. 𝗖𝗼𝗲𝗿𝗰𝗲 𝘁𝗵𝗲 𝗕𝗿𝗲𝗲𝘇𝗲 𝗔𝗜 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗲 𝗶𝗻𝘁𝗼 𝘁𝗵𝗲 𝗰𝗼𝗿𝗿𝗲𝗰𝘁 𝗱𝗿𝗼𝗽𝗱𝗼𝘄𝗻 𝗼𝗽𝘁𝗶𝗼𝗻𝘀:

We all know free-form text is the devil (doubly so with AI), so formatting it correctly into a dropdown is a must.

6. 𝗨𝗽𝗱𝗮𝘁𝗲 𝗺𝘆 ”𝗔𝗜 𝗖𝗮𝘁𝗲𝗴𝗼𝗿𝘆” 𝗽𝗿𝗼𝗽𝗲𝗿𝘁𝘆 𝗼𝗻 𝘁𝗵𝗲 𝗮𝘀𝘀𝗼𝗰𝗶𝗮𝘁𝗲𝗱 𝘁𝗶𝗰𝗸𝗲𝘁.

Easy peasy!

A decent ticket sentiment system that can be put together quickly. I’m looking forward to exploring more use cases for Breeze AI.