Agent handoff

What agent handoff actually means in Shopify and WooCommerce ops

In a Shopify or WooCommerce support stack, agent handoff is the moment an AI agent stops answering and pushes the conversation. Plus everything it already knows. Into a human agent's queue. It is not a simple "talk to a human" button. A real handoff carries the transcript, customer identity, order ID, shipping status, refund eligibility, the AI's failed intent, and the customer's sentiment.

The goal is continuity: the human picks up where the machine left off. Without that payload, the customer has to re-explain the problem, and the agent has to hunt through the order admin. That friction is expensive. In practice, most handoff failures I see are not AI comprehension failures; they are integration failures. The bot understood the issue, but the helpdesk ticket was created with none of the order context.

Shopify Inbox automations and Gorgias AI Agent both emphasize passing variables into the ticket, yet many merchants still launch with a bare transcript.

Handoff quality matters more than containment rate

Vendors love to sell containment rate. The share of chats the AI "handles" without a human. A high number looks great on a slide. It is also easy to game. An AI can containment a conversation by giving a vague answer, exhausting the customer, or routing them to a form that never gets answered. That is not success; it is deflection dressed up as automation.

I would rather see a lower containment rate with clean handoffs than a 90% containment rate that creates silent rage tickets three days later. The metric that matters is resolution quality: did the customer get the right outcome in one touch, and did the human agent receive enough context to finish it fast?

McKinsey's work on generative AI in customer care notes that value comes from workflow redesign and good routing, not from removing humans for the sake of a dashboard number. McKinsey generative AI customer operations

The context payload: what must travel with the ticket

Think of the handoff as a package, not a ping. At minimum the package should include the original customer message, the full AI transcript, the customer's email or phone, the order or subscription ID, the last known order status, any policies the AI already checked, the attempted answer, the detected intent, sentiment score, and the escalation reason.

If the issue is a WISMO ("Where is my order?"), the payload should already contain the tracking number and carrier. If it is a refund, it should contain return window status and previous refund history. The receiving agent should not need to open Shopify, WooCommerce, Recharge, or Loop Returns to reconstruct the case.

Some platforms call this a "context card" or "conversation summary." Zendesk AI agent handoff and Gorgias macro variables let you inject this data, but you have to map the fields yourself. Most merchants skip the mapping and then wonder why agents still ask for order numbers.

When to trigger a handoff

Handoff rules should be explicit before launch. I recommend four triggers: high-risk intent (refund, chargeback, complaint, legal threat), low-confidence classification (the AI is below 0.8 certainty), repeated customer frustration signals (two "that did not help" responses), and access to actions the AI is not allowed to take (issuing refunds, editing subscriptions, canceling orders). Do not let the AI guess its way through a refund over $50 just to protect a containment metric.

Also set business-hours logic: if the issue needs a warehouse or supplier check and humans are offline, a handoff should include a clear "we will reply by 9am" promise rather than a silent ticket. Ada handoff best practices and Gorgias AI Agent rules both suggest intent-based escalation, though in my judgment most out-of-the-box rule sets are too aggressive about keeping conversations in the bot.

How to measure handoff success

Stop reporting containment rate as a standalone KPI. Pair it with handoff quality score, first-contact resolution after handoff, agent handle time on handed-off tickets, customer repeat contacts within 72 hours, and CSAT for tickets that were handed off. A good handoff should reduce. Not increase. Agent handle time compared with a ticket created from scratch.

If your human agents spend longer fixing AI-handled chats than they would on normal tickets, your AI is adding work. I also track "context gap" tickets: cases where the agent had to ask the customer for information the bot already saw. That is a handoff design failure, not an agent failure. Zendesk benchmark metrics and Gorgias analytics give you the raw numbers, but you have to define the quality thresholds yourself.

Implementation checklist and common mistakes

Before flipping the AI live, map every intent to a handoff decision: resolve, collect, or escalate. Connect your helpdesk to Shopify or WooCommerce so order data flows into the ticket. Build a context card template and require it on every handoff. Train agents to read the card before greeting the customer; a human who says "I see your order #1234 is stuck in Memphis" saves 30 seconds and earns trust.

Test handoffs with real orders, not demo accounts. The most common mistake is launching with a generic "I'll connect you" message and no payload. The second most common mistake is routing all AI escalations to a single catch-all queue instead of the right team. Refunds to finance, WISMO to logistics, technical issues to product support. Shopify Inbox routing and Gorgias rules can automate that routing, but only if you configure it.