How AI Chatbot Human Handoff Should Work

Step 1: define the escalation boundary before launch

Do not wait until a customer gets stuck to decide what needs a human.

Create categories that always or usually escalate:

• refunds and replacements
• policy exceptions
• complaints
• missing deliveries
• order changes
• custom quotes
• account-specific actions
• questions where source information is missing

Then create categories the AI can usually answer:

• FAQs
• published policies
• product or service information
• routine order status
• navigation questions

The exact boundary depends on your business, but it should exist before the chatbot goes live.

Step 2: tell the customer what is happening

Do not hide escalation behind vague wording.

Good handoff language is simple:

“I’ll connect you with our support team. Please share your email so they can follow up with the conversation context.”

Or, when direct messaging is available:

“I’ll connect you with our support team. Tap below to continue.”

The customer should understand whether the next step is email follow-up, live messaging, or another channel.

Step 3: preserve context

The customer should not need to repeat the whole story.

Save the transcript or pass a useful summary to the human workflow. At minimum, the person following up should know:

• what the customer asked
• what the AI already answered
• where the customer became stuck
• any relevant reference captured by the system

Farnsla preserves chat history so merchants can review the previous conversation.

Step 4: stop the bot from continuing after escalation

A common failure mode is a chatbot that says it will escalate and then keeps generating more automated answers.

Once the customer needs a person, make the transition clear. The AI can collect the information needed for follow-up, but it should not keep pretending it can solve a problem that has already crossed the human boundary.

Step 5: match the handoff method to the urgency

Email follow-up is suitable when the issue can wait and a person needs time to investigate.

Direct messaging is useful when the merchant already supports customers in a messaging app and wants the customer to continue there.

For Shopify, Farnsla Customer Messaging currently supports Direct Messaging and AI handoff to configured supported apps including WhatsApp, LINE, Messenger, Instagram, and Telegram. In Direct Messaging mode, Farnsla does not call the AI.

For Wix, the current public product documentation focuses on email capture and saved transcript for human follow-up. Do not claim direct-messaging handoff there unless that capability is live.

Step 6: review escalated conversations

Escalations show you where automation stops working.

Some should always stay human. Others reveal a knowledge gap you can fix.

For example:

Repeated escalation: “Do you ship to military addresses?” If the business has a clear policy, add it to the knowledge base.

Repeated escalation: “Can I get a refund outside the return window?” Keep it human because it is an exception decision.

Human handoff analytics are therefore useful for deciding what to automate next — and what never to automate.

A simple Farnsla handoff model

Routine question → AI answers from approved knowledge.

Secure order lookup → customer verifies email → current status returned.

Missing information or business decision → collect follow-up / offer supported human channel.

Human reviews transcript → continues with context.

That structure gives customers faster answers without making automation a barrier between them and your team.