I’ve had great success over the last three weeks feeding Fin Conversation Transcripts that have a negative CX score, to our other internal AI chat tools, and working to improve Fin AI Agent immensely. (Our Fin Ai Agent Resolution rate has increase by 6% up to 54%+ over the past four weeks. 💪🏼
The setup:
- Build a custom report that pulls Fin AI Agent conversations where the CX score was 1, 2 or 3.
- Pull transcripts for the conversations each week
- Train our internal AI Agent on all of the info it needs to guide me best
- What tools I have to train this Fin AI Agent (All different Guidance, categories Communication settings, etc)
- Trained on all of our products using both the Help Center that Fin uses, and our full internal knowledge base with even deeper product info and knowledge.
The Prompting:
Next, I iterated on a good prompt that works smooth, and scales. I started out with a simple one conversation prompt.
This is a support conversation that was auto-categorized as low CX score for the reasons
CX reason 1(high user effort for example), andCX reason(negative product feedback). Using the tools I have available to train this AI Support agent, and without changing the product, how could fin be trained better to communicate differently and improve the customer experience with this engagement?
The cleanup and tip
I’ve added probably 20+ guidance entries over the last three weeks, and the OPTIMIZE button is game changer. I now never skip it. It will not only improve the logic of guidance I’m inputting, but it will also find conflicts and tell you where they are. (Always name your guidance, as it’s helpful to find them when it starts growing).