How should an ecommerce AI assistant hand off unanswered questions to a human?

An ecommerce AI assistant should hand off as soon as it cannot answer confidently, the shopper asks for a person, or the issue needs human judgment. The handoff should explain what happens next, preserve the full conversation and relevant product or order details, and avoid making the shopper repeat the question. If no human is available, collect only the information needed to follow up and give a clear expectation rather than pretending the transfer is live.

Decide when the AI assistant must stop

An AI assistant should escalate before it guesses, not after it gives a misleading product or policy answer. A useful rule is: let the assistant clarify an ambiguous question once, but route the conversation when the answer still depends on information it cannot verify.

Use explicit escalation rules for these situations:

Trigger	Why it should go to a human
The shopper asks for a person	The shopper has already stated that automation is no longer their preferred path.
The assistant cannot find a reliable answer	A confident-sounding guess can create the wrong purchase expectation.
The question concerns an exception	A human may need to interpret a warranty, return, discount, delivery, or product-use exception.
The shopper reports a problem with an existing order	The case may require account, order, payment, or fulfillment information that the assistant cannot safely infer.
The shopper shows frustration or repeats the question	Another bot response is likely to feel like resistance rather than help.
The request involves risk or sensitive information	Route decisions involving safety, privacy, payments, or unusual customer circumstances to the appropriate team.

Do not escalate every unfamiliar phrase. First try a short clarification that gives the shopper useful choices: “Are you asking about fit, material, or delivery?” If the shopper answers and the assistant can respond from approved information, continue. If not, make the boundary visible and hand off.

A strong fallback names the limitation without blaming the shopper: “I don’t have enough verified information to answer that accurately. I can connect you with our team and pass along what you’ve already told me.”

Transfer the context the human needs

A handoff is only successful if the human can begin with the shopper’s situation, not with “How can I help?” Send the conversation history, the assistant’s reason for escalating, and the entities that matter to the decision.

The handoff record should include:

• The shopper’s original question and any clarification they supplied
• The products, variants, sizes, colors, or categories discussed
• What the assistant answered, including any uncertainty
• The exact escalation trigger
• Relevant cart, order, account, or location details that the shopper has consented to share
• The shopper’s requested outcome, such as choosing a size or checking an exception
• A concise summary for the human agent

Show the shopper that this context moved with them. For example: “I’ve passed your question about the Trail Shell’s sleeve length and your preferred size to our product team.” That sentence reassures them that the transfer is specific, not a reset.

Do not ask for information the business already has. Ask only for a missing detail that is necessary to route or resolve the case. If the shopper must provide an email address for an offline reply, say why it is needed and do not imply that a live agent has joined.

Guidance on AI-to-human context transfer from Typewise similarly emphasizes transferring the transcript, extracted entities, and a short case summary instead of making the receiving agent reconstruct the interaction.

Make the next step honest and specific

The assistant should tell the shopper whether the handoff is live, queued, or asynchronous. A vague “someone will get back to you” creates uncertainty; a precise next step lets the shopper decide whether to wait, continue shopping, or use another contact route.

Use different messages for different states:

A human is available now:

I’m connecting you with a product specialist now. I’ve shared your question about fit and the two items you’re comparing.

The team is online but not immediately available:

Our team is online, but there may be a wait. I’ve sent them your question about compatibility so you won’t need to repeat it here.

The team is offline:

Our product team is offline. Leave your email and question, and we’ll follow up through the support channel. You can also review the compatibility guide while you wait.

Only state a response time if the business can support it consistently. Otherwise, provide operating hours or a clear queue state without inventing a promise. Give the shopper a transcript or reference number if that helps them return to the conversation.

The Gorgias handoff documentation provides a useful operational distinction between online and offline handoffs, including communicating availability and, where configured, expected wait time. The underlying principle applies regardless of the tool: the message must match the actual path the conversation will take.

Keep the shopper moving without trapping them

A handoff should offer a choice, not hold the shopper hostage in a dead-end queue. After collecting the minimum context, provide a relevant next action that does not undermine the human request.

Good options include:

• Continue browsing the products already discussed
• Save the conversation or receive a follow-up by email
• Open the relevant size, compatibility, shipping, or returns information
• Choose another support channel when the issue is urgent
• Cancel the handoff and ask a different product question

Do not repeatedly present self-service articles after the shopper explicitly asks for a human. One relevant option can help; a carousel of deflection attempts signals that the business is avoiding the request.

The assistant should also prevent duplicate work. If the shopper starts a second conversation before the first is answered, connect the new message to the existing case where possible, or tell the human agent that the shopper has returned. Repeatedly restarting the flow is one of the fastest ways to make automation feel broken.

Turn unanswered questions into improvements

An unanswered question is both a service case and evidence of a gap in the buying journey. Review escalated conversations for the reason they stopped, not just the number of transfers.

Track categories such as:

• Missing product specifications
• Confusing or incomplete policy language
• Questions about a product combination or use case
• Inventory, delivery, or location uncertainty
• Requests that require account access
• Assistant misunderstandings or incorrect source data

Then assign each category an owner. Product or merchandising teams may improve product-page content; customer experience teams may clarify policies and routing; ecommerce teams may fix catalog or availability data. Keep questions that require judgment in the human workflow rather than forcing them into an automated answer.

Anagram’s guidance for its Site Agent recommends reviewing the question log after launch and flagging questions that need human escalation. That makes the handoff a feedback loop: the assistant supports shoppers now, while their questions show the brand where product information and purchase guidance need work.

Measure handoff quality, not just automation

A lower handoff rate is not automatically a better customer experience. Measure whether the shopper received the right help and whether the human could pick up the case efficiently.

Useful measures include:

• Escalation accuracy: how often a human was genuinely needed
• Context completeness: whether the receiving agent had the transcript, reason, and relevant entities
• Repeat-question rate: how often shoppers had to restate the issue
• Time to human response: measured separately for live and offline routes
• Resolution after handoff: whether the case ended in a useful answer or next step
• Shopper outcome: assisted purchase, continued browsing, abandonment, or follow-up completion
• Top unanswered topics: the content and product-data gaps worth fixing first

Test the entire journey with real scenarios before launch: an unknown product question, an explicit request for a person, an angry shopper, an offline handoff, and a case involving an existing order. Inspect exactly what the human sees and exactly what the shopper is told.

The best ecommerce AI assistant does not try to win every conversation. It answers well within a defined boundary, recognizes when that boundary has been reached, and hands over with enough clarity and context that the shopper feels progress rather than abandonment.