Resolve support tickets with guarded AI orchestration
Classify new tickets with AI, route them by category, and reply or escalate only when confidence is high enough.
Workflow preview:
Overview
Support volume grows faster than headcount. Ungoverned AI makes that worse.
When ticket volume climbs, the first instinct is to throw a chatbot at every conversation. That works until the bot invents a refund policy, replies to a VIP in the wrong tone, or sends a customer down a path your team never approved. Free-range AI is fast. It is also hard to defend in a CX org that cares about brand, compliance, and containment.
The other extreme is pure human triage. Agents read every ticket, classify it, dig through help center articles, and draft replies by hand. That keeps quality high and burns hours on work that follows the same patterns every day.
This template sits between those extremes. It is an orchestrator: new support conversations get classified, scrubbed of obvious PII, and routed by category. AI drafts replies only inside guidance you define in Google Sheets. Confidence gates decide whether to send, tag, close, or hand off to a human with an internal note. You predefine the path. AI decides inside the rails.
This version ships with Dixa as the ticketing system. The same pattern works with other helpdesks — swap the ticket trigger and actions for Zendesk, Freshdesk, Gorgias, or similar, and keep the Sheets + AI + confidence-gate middle.
Why build a guarded AI support orchestrator?
- You keep humans on the tickets that need judgment— Routine FAQ and predefined flows can reply or close; low-confidence or do-not-automate paths escalate with context already written for the agent.
- You control prompts, categories, and routing in Sheets— CX operators can update the prompt library, category definitions, routing rules, and style guide without rebuilding the Zap.
- You can swap the helpdesk without losing the pattern— Dixa is the default in this template. Sheets hold the CMS. Sub-Zaps are stubs for your own category-specific flows and FAQ responder. Replace the edges to match your stack.
What key features come with this template?
- Classifies tickets with AI against your category sheet— the model picks category, subcategory, user role, and a confidence score using definitions you maintain, not a free-form guess.
- Scrubs PII before AI sees the message— a Code step strips HTML and masks emails and phone numbers so classification and drafting run on cleaner text.
- Routes by category into three orchestrator paths— predefined flow Sub-Zap, FAQ responder Sub-Zap, or do-not-automate handoff with an agent-ready summary.
- Gates outbound replies on confidence— AI-drafted replies only send when confidence clears your threshold; otherwise the draft becomes an internal note and the ticket moves to a human queue.
- Writes outcomes back to your helpdesk— categories, summaries, confidence, tags, queue transfers, and closes stay on the ticket so reporting and agents stay in sync.
How does this template work?
This ready-made template builds you a support orchestration Zap centered on your helpdesk (Dixa in this version) and Google Sheets. Here's how it works:
- A new support conversation is created in your ticketing system and the Zap starts.
- The Zap pulls conversation messages and ticket details, then skips channels or queues you mark as off-limits (for example phone or spam).
- It loads prompts, category definitions, and routing rules from Google Sheets.
- A Code step cleans the message text and masks emails and phone numbers.
- AI by Zapier categorizes and summarizes the ticket, then writes those fields back to the ticket.
- Routing decides: run a predefined-flow Sub-Zap, run an FAQ Sub-Zap, or escalate without automating a customer reply.
- On reply paths, AI drafts from flow guidance plus your style guide. High confidence sends and tags the ticket; low confidence escalates with an internal note.
- Backend-only paths can close, transfer to a specialist queue, or reopen after a short delay.
You'll find detailed setup instructions inside the template. Scroll down for a preview.
Who should use this template?
- CX systems managers— who want AI in the support path without giving it free rein on every ticket.
- Support operations leads— who already keep policies and macros in spreadsheets and want those sheets to drive automation.
- IT / automation builders— pairing a helpdesk with Zapier AI steps and Sub-Zaps for category-specific work.
- Enterprise support teams— that need confidence gates, queue filters, and audit-friendly tags on automated replies.
- Teams inspired by high-volume marketplace or ecommerce support— who need orchestration first, not a single chatbot step.
Start routing support with guarded AI.
Connect your helpdesk (Dixa in this version) and Google Sheets, map your attribute and queue IDs, attach your flow and FAQ Sub-Zaps, and turn the orchestrator on. Keep the rails in Sheets. Let AI decide only where you said it could.
Frequently asked questions
What apps do I need to use this template?
Out of the box: Dixa (ticketing), Google Sheets (prompts, categories, routing, style guide), AI by Zapier, and Code by Zapier. Sub-Zap steps are stubs you point at your own Zaps. If you use another helpdesk, swap the Dixa steps for that app's equivalents.
Is this Mercari's exact production workflow?
No. This is an orchestrator-pattern template inspired by Mercari's approach. Sensitive CX sub-workflows are not included. You bring your own Sub-Zaps, Sheets, and helpdesk field IDs.
Can I swap Dixa for another helpdesk?
es. Replace the Dixa trigger and actions with your ticketing app's equivalents (for example Zendesk, Freshdesk, or Gorgias) and remap fields. Keep the Sheets + AI + confidence-gate pattern.
What do I put in the Google Sheets?
At minimum: a prompts sheet, a category definitions sheet, a routing rules sheet (category/subcategory → flow type), and a style guide sheet. Column layouts should match what the Zap lookups expect after you map them.
Why are the Sub-Zap steps required?
The orchestrator decides which path to take. Category-specific flows and FAQ answering live in separate Zaps so you can change them without exposing or rewriting the whole orchestrator. Replace the Sub-Zap ID placeholders with your Zaps, or remove paths you are not using yet.
How does the confidence gate work?
AI returns a 1-10 confidence score on classification and on drafted replies. Filter steps stop automated sends when confidence is below your threshold and escalate to a human queue with an internal note instead.
Does this template scrub all PII?
No. The included Code step masks common email and phone patterns and strips HTML. Extend it for your compliance needs (order IDs, addresses, etc.) before running AI on sensitive queues.
Can I turn off automated customer replies and only use classification?
es. Use the do-not-automate path, raise confidence thresholds so sends never clear, or remove the send-message steps and keep tagging, notes, and queue transfers.
Will this guarantee the same results Mercari published?
No. Outcomes depend on your volume, category design, Sub-Zaps, and confidence thresholds. Use the Mercari customer story for proof of the pattern, not as a promise for your account.