Customer Support

Every Relay.app workflow starts with a trigger. Here it's an event: when a new ticket lands in Intercom, the workflow fires in seconds. No polling, no daily batch — first response time drops because the draft is sitting there when the agent opens the conversation. Same shape works with any help-desk product.

Claude classifies the ticket into one of four buckets — bug, billing, feature request, how-to — and assigns urgency and sentiment alongside. The classification is the routing key for the path-group below. Swap Claude for GPT-5.5 or Gemini 3.1 Pro if those match your eval results better, or add a Knowledge Base lookup step before classification so the model has context to refer to.

Claude's classification kicks one of four paths. Bugs file a Linear issue and ping #engineering-bugs after the draft is ready. Billing pulls the customer's subscription from Stripe so the reply cites the right invoice. Feature requests append to the roadmap in Notion, then Claude drafts a "we're tracking it" note. How-tos search the Knowledge Base and ground the answer in the matching article. Click any tab above to see what runs in each branch.

Whichever path fired, the support agent gets the drafted reply pre-filled in Intercom with the classification labels attached. They can send as-is for the obvious ones, edit for nuance, or rewrite from scratch — the workflow doesn't get in their way. Once they approve, the reply ships through the same Intercom thread the customer started in.

Start from this template and adapt it to your stack and your team. Relay.app's copilot picks up from here — swap any app, change the AI model, tweak the trigger, whatever fits your use case.