l’m working on a workflow involving AI agent conversations and I’m encountering some challenges. I’m hoping to get some advice on improving our current setup or finding a better approach altogether.
Current Setup
We’re using AI agents for customer conversations and want to implement a system to:
Schedule follow-ups when there’s no response from the customer
Currently, we’re using schedule triggers for follow-ups, follow up is some sales problems our software tried to solve to get engagement.
The Challenge
Our main issues are:
Maintaining Context: Make sure we don’t repeat questions or followup.
Scheduling Follow-ups: While we’re using schedule triggers.
Questions
Is there a more efficient way to handle scheduled follow-ups while maintaining the conversation history?
How can we structure our workflow to ensure that follow-up messages include the necessary context from previous interactions?
Are there any best practices or patterns in n8n for building conversational AI workflows with persistent memory?
Any insights, suggestions, or examples would be greatly appreciated. Thank you in advance for your help!
Have you tried just bringing in an external database of some sort to manage customer session and chat history separately?
Here’s a quick mockup of how I could imagine this workflow with something like Airtable.
I would probably ditch the Window buffer chat memory. It’ll clear if your instance is restarted and you can’t easily review the messages anyway? It’ll be more work for sure but try saving the conversation elsewhere.
For the follow-up workflow, if you use an external database you may not need to use a conversational agent and reply on the chat memory. A basic LLM chain will probably be simpler if you just need to summarise the previous conversation to send the follow-up message.
