Customer Support Ticket Routing

Configure Intercom to capture customer inquiries from live chat, email, and in-app messaging. Use Intercom's custom attributes and conversation data to collect context including the customer's plan, account age, recent activity, and previous support history. Set up an automation that forwards new conversations to your AI processing layer via webhook.

Send the conversation text and customer metadata to Claude with a prompt that classifies the issue by category, identifies the product area involved, assesses urgency based on business impact signals, detects customer sentiment, and suggests a resolution path. Output a structured classification object that downstream systems can consume.

For common issue types with known resolutions, use Claude to generate a contextual auto-response that acknowledges the customer's specific problem, provides relevant help documentation or troubleshooting steps, and sets expectations for response time. This reduces first-response time for straightforward issues while the ticket is simultaneously routed to a human agent.

Use the Zendesk API to create a ticket with the AI-generated classification, priority, and suggested resolution pre-populated in custom fields. Apply routing rules that assign tickets to specialized agent groups based on the issue category and urgency. Attach the full Intercom conversation history and Claude's analysis as an internal note so the assigned agent has complete context.

For tickets classified as critical or involving VIP customers, trigger an immediate Slack alert to the support leadership channel with the ticket summary, customer impact details, and a direct link to the Zendesk ticket. This ensures high-priority issues receive immediate human attention rather than waiting in the standard queue.

After ticket resolution, use a Zapier automation to log the issue category, resolution steps, and customer feedback into a Notion knowledge base. Track recurring issue patterns and resolution effectiveness over time. This growing database feeds back into Claude's classification prompts to improve accuracy and enables the support team to identify systemic product issues that need engineering attention.