Most support questions are the same questions, asked over and over. Where's my order? How do I return this? What's included? Answering by hand is fine at low volume and drains your team at scale. This playbook sets up your AI Agent to handle the routine ones reliably, so your team only steps in for the cases that genuinely need a person.

An AI Agent that answers common support questions accurately and in your voice, escalates anything it can't resolve, and flags unresolved contacts to your team so nothing slips through.

For support, this is the most important step. In AI Agent → Knowledge access and behaviour, add entries for:

Write each entry the way you'd explain it to a customer: short, direct, honest. Then set the access mode to Limit to instructions so the agent sticks to what you've taught it and won't guess at a policy. Full setup in Knowledge Base: Teaching Your AI Agent What to Know.

In AI Agent → Voice and personality, pick the Support agent role. Add message examples from real support moments you handled well: a calm reply to a late delivery, a clear answer on a refund, a response to a damaged-item report. Tell it to acknowledge the issue before moving to a solution and never to promise things it isn't sure about.

In AI Agent → Goals and instructions, set Goals and capabilities to: resolve common support questions accurately and escalate anything that needs a human, quickly. This stops the agent from trying to sell mid-support.

Use Other instructions for your most important rules: never promise a specific resolution timeline, never confirm a refund is approved, acknowledge how an upset customer feels before solving, and always offer to escalate if they're not satisfied. See Goals and Instructions: Telling Your AI Agent What to Do.

In AI Agent → Human intervention, turn it on and pick a folder like "Needs team" so you have a clean list of escalations. Then notify your team when someone lands there: a scenario with the A contact is added to a folder trigger that emails your team works well. See Human Intervention: Handing Off to a Human.

In AI Agent → Activation rules, All pending messages is usually right for support. Questions don't always match a tidy intent, so you want the agent to pick up anything that doesn't already have a scenario running.

A customer DMs: "My order was supposed to arrive 3 days ago and nothing's here."

The agent acknowledges the issue, shares the relevant shipping timeline from your knowledge base, and if it looks like a genuine delay, asks for the order number, saves it to the profile, files the contact in a "Delayed order" folder, and notifies your team. If a question falls outside what it knows, it says so plainly and flags the contact for follow-up.

Capture context automatically. Add an "Order number" or "Issue type" property to User Data Collection: Saving Contact Info Automatically so your team has what it needs before they open the conversation.

Answer order status with live data. If you use Shopify, connect it so the agent can reference real order data instead of guessing. See Shopify: Connect Your Store and Send Product Links.

A support agent fielding routine questions around the clock cuts your team's volume sharply, especially after the first couple of weeks of closing gaps. Escalations should drop steadily as you turn each "needs team" case into a knowledge entry.

Get the handoff right so escalations reach a person cleanly. See Human Intervention: Handing Off to a Human to configure exactly when and how the agent steps aside.