Introduction to AI Agent for Customer Support Case Processing recipe
Introduction to AI Agent for Customer
Support Case Processing recipe
The AI Agent for Customer Support Case Processing is an event-driven solution designed
to automate the analysis of incoming customer emails by identifying user intent and triggering
appropriate follow-up actions.
This versatile recipe is adaptable across various software environments and integrates
seamlessly with systems, such as Zendesk, Jira, and Pinecone. It leverages an internal
knowledge base powered by a configurable Large Language Model (LLM), with Gemini models
supported by default. Users can customize the recipe to work with other GenAI platforms by
establishing new connectors and replacing Gemini-specific calls as needed.
Designed for easy integration, the recipe enables seamless incorporation of customer-specific
software. It can be easily extended by adding new agents and services, making it flexible to
evolving business needs.
The agent processes emails based on three primary user intents, using distinct modes tailored
to address each effectively:
Query Mode:
Handles customer
inquiries seeking answers or problem resolution. The agent uses the LLM to generate
automated responses and creates or updates a Zendesk ticket to track the issue.
Feedback Mode:
Manages user feedback
on existing cases, such as acknowledgments, requests for additional support, or escalations.
In this mode, the system closes the corresponding Zendesk ticket, marking the agent’s task
as complete.
Feature Request Mode:
Addresses
requests for new features. The agent searches the linked Jira system for similar requests
stored in the Pinecone database. If a match is found, it references the existing Jira
ticket. Otherwise, it creates a new ticket for human follow-up. Newly created Jira tickets
are indexed in Pinecone for future reference.
The workflow begins when the EmailConnection app triggers the recipe through an Email
Listener Event. The agent extracts the email’s subject and body, passing this content to the
PlanningAgent process for intent analysis. Upon determining the user’s intent, Query,
Feedback, or Feature Request, the PlanningAgent generates a structured JSON object outlining
the necessary steps. It then sequentially invokes specialized subagents to execute these
steps.
Throughout the process, the system automates the following key actions:
Creating or updating Zendesk tickets with
the user’s input and LLM-generated responses, with ticket status updated according to
workflow progression.
Creating new Jira tickets for feature
requests or retrieving existing Jira tickets when similar issues are found in the Pinecone
index.
Sending timely email updates to users after
each interaction, keeping them informed throughout the support process.
This intelligent, event-driven automation enhances support efficiency and improves customer
satisfaction by combining AI-powered analysis with integration across popular support
platforms. Although originally implemented using Gemini, the solution’s flexible design allows
easy adaptation to other GenAI technologies through appropriate connectors.
ASK INFAPreview
