A Shopify AI agent handoff is the moment the AI passes a conversation to a person, with the order details and a summary attached, while it keeps helping the customer until someone replies. It should fire on clear triggers (the customer asks, is frustrated, the AI keeps failing, a VIP, a sensitive topic) and never make the customer repeat themselves.
This post is for Shopify store owners and support leads setting up AI to human handoff rules. It goes deeper than the short handoff section in our Shopify AI customer service guide. Fyncall makes an AI helpdesk for Shopify, so we have an interest. The rules below work with any tool, and each Fyncall detail links to our docs or product pages.
What is a Shopify AI agent handoff?
A Shopify AI agent handoff is a live agent transfer: the AI flags a conversation, a person picks it up, and the person decides whether the AI keeps answering simple follow-ups or steps back completely. Good handoff design covers three things: when it happens, what the person receives, and what the customer hears while they wait.
In Fyncall, an escalated conversation lands in the Needs Agent queue, described in the inbox overview as “conversations waiting for a human agent (AI escalated)”. AI handovers also raise a conversation’s score in the Needs Attention queue, so the urgent ones rise to the top.
When should the AI hand off to a person?
The AI should hand off when continuing would cost more trust than it saves time. Fyncall’s AI Escalation settings offer five triggers, and each can be switched on or off. These settings are on every plan, including Free:
- Explicit Human Request. The customer asks for a person. This should always work, with no attempt to deflect.
- Customer Frustration Detection. Sentiment analysis counts negative indicators. You set the threshold at 2, 3, 4 or 5.
- Query Complexity. The AI has failed to resolve the question after a set number of attempts, from 1 to 5.
- VIP Customer Prioritization. High-value customers are prioritised when another trigger fires.
- Sensitive Topic Detection. Legal mentions and complaints go to a person.
Start strict. A frustration threshold of 2 and a failed attempts limit of 2 will send more conversations to your team at first, which is fine while you learn what the AI handles well. Loosen one setting at a time once you have read a week of escalations. For the wider question of which work suits AI at all, see AI vs human customer support for ecommerce.
Which Shopify order situations should trigger a handoff?
Order situations that involve money, a shipped parcel or a dispute should reach a person quickly. Fyncall’s AI can look up orders, share tracking, and change an address or cancel an unfulfilled order only after the customer confirms. It checks refund eligibility and refund status, but it does not issue refunds. The table maps common situations to the trigger most likely to catch them.
| Order situation | What the AI can do first | What usually triggers the handoff | What the person needs |
|---|---|---|---|
| Address change after fulfilment | Look up the order and see it has shipped | Failed attempts, or a rule telling the AI to escalate | Tracking number, carrier, new address |
| Refund request | Check refund eligibility and status | Explicit request, or a complaint | Order, eligibility result, your policy |
| Damaged item | Look up the order and items | Complaint (sensitive topic) or frustration | Order, items, anything the customer sent |
| Chargeback threat | Nothing beyond the lookup | Sensitive topic (legal or complaint) | Full history and payment details |
| VIP customer | Answer as normal | VIP prioritisation once another trigger fires | Total spend and order history |
| Wrong item sent | Look up what was ordered | Complaint or frustration | Order line items and fulfilment record |
Two of these need care. An address change after fulfilment is not one of the five triggers, so write it into the agent’s instructions in Agent Config, or a Must Follow rule in Guardrails (on every plan, including Free), for example “If the order has shipped, do not promise an address change; escalate to the team.” A chargeback threat usually reads as a legal mention or a complaint, but test a few real wordings in the Test sandbox before you rely on it.
What should the person receive at handoff?
The person should receive enough to reply without asking a single question the customer has already answered. On Fyncall, when the AI hands over, your team gets the conversation with a short summary, as set out on our AI support agent page. The AI keeps helping: it stays with the customer while they wait, and on Pro, AI Copilot can draft replies, summarise and look up orders for the person who picks it up. The customer panel beside it shows Shopify order history with status, items, tracking, payment and address, plus total spend.
When the AI opens a ticket, the tickets docs list what it carries in an AI Context card:
- Intent: what the customer wants to achieve.
- Sentiment: satisfied, neutral, frustrated or upset.
- Actions taken: what the AI already did or tried.
- Suggested resolution: the AI’s recommended next step.
- Trigger type: which escalation rule fired.
Ticketing with SLAs and escalations is listed under Fyncall Pro on the pricing page.
How do you stop customers repeating themselves?
Customers repeat themselves when the handover summary is missing or the person does not read it. Three habits fix most of it.
- Read the summary and the AI Context card before typing anything.
- Open with what you already know: “I can see order 1042 was delivered on Tuesday with the wrong size.”
- Use the Actions taken field to avoid repeating a step the AI already tried, such as resending a tracking link.
Avoid generic escalation wording too. In the AI Escalation settings you can write a custom Escalation message in place of the default. Something like “I’ve passed this to the team with your order details, so you won’t need to explain it again” sets the right expectation.
What happens to a handoff outside working hours?
Outside working hours, a handoff needs a plan for the customer who waits. Fyncall’s working hours setting uses your company time zone and offers three options for outside hours behaviour:
- Normal (same as working hours): escalations behave as they do in the day.
- Suggest creating a ticket: the customer is offered a ticket.
- Automatically create ticket: a ticket is opened for them.
You can add an Outside hours message that says when the team is back. Pair it with timeout settings for after-hours support: an Escalation timeout of 2, 3, 5, 10, 15 or 30 minutes, Send follow-up on timeout with up to 5 follow-ups at 5 to 30 minute intervals, and Suggest ticket creation on timeout if nobody answers. The follow-up and ticket suggestion messages can be customised as well.
Alerts arrive in the app and by email. Fyncall does not have an agent mobile app yet, so plan coverage around those.
How does handoff work on WhatsApp and Instagram?
On WhatsApp and Instagram, the person should reply on the same channel the customer used, inside that channel’s reply window. The window is the main risk with an overnight escalation.
| Channel | Reply window | After it closes |
|---|---|---|
| 24 hours from the customer’s last message, reset on each reply | Approved templates only | |
| 7 days from the customer’s last message (in Fyncall) | Wait for the customer to write again | |
| None | Reply any time |
Meta’s Cloud API documentation states that when the customer service window closes, “you can only send pre-approved template messages.” The channels docs show a “Reply window expired” banner in Fyncall once the WhatsApp window closes, and free-form replies are blocked from then on. Instagram has no template fallback, and Meta has its own rules on what can be automated there, covered in our post on Shopify Instagram DM automation. So treat WhatsApp escalations raised late in the evening as first in the morning queue. For more on running this channel, see our guide to WhatsApp support for a Shopify store.
How do you hand the conversation back to the AI?
Fyncall has three controls in the conversation header, described in the managing conversations docs:
| Action | Who is assigned | Does the AI reply? |
|---|---|---|
| Claim | You | Yes, the AI keeps responding |
| Take Over | You | No, until released |
| Release to AI | Nobody | Yes, it resumes |
Use Claim when you want ownership but the AI can keep answering simple follow-ups. Use Take Over for human takeover of the AI chat: the AI stops replying, and on web chat the customer sees that a person has joined. When the issue is settled, Release to AI so the next “where is my parcel?” message gets an instant answer; our guide to Shopify order status automation sets out a WISMO flow the AI can follow. Before releasing, Fyncall warns you about open tickets or pending follow-ups, and only the assigned agent or an admin can release.
How do you reduce unnecessary handoffs?
Unnecessary handoffs usually come from a gap in the knowledge base or a trigger set too tightly. Read each week’s escalations and sort them into three groups: the AI lacked an answer, the AI had the answer but a trigger fired early, and the handoff was right.
- Missing answers. Add the policy to the knowledge base, then check it with Test Knowledge Base, which shows each source and its match percentage.
- Early triggers. Raise the frustration threshold or failed attempts limit by one step and watch the next week.
- Right handoffs. Leave them alone. These are the conversations your team should own.
Save hard cases as scenarios in the Test sandbox and rerun them after each change. If you are earlier in the process, how to automate a Shopify helpdesk covers which work to automate first.
What should your handoff checklist include?
Copy this into your setup notes and tick each item before turning on Auto-Reply. If you are still deciding whether to switch the AI on at all, start with when to use AI support on Shopify.
- Explicit Human Request is on.
- Frustration threshold and failed attempts limit are set (start at 2).
- Sensitive Topic Detection is on and tested with your own wording for chargebacks and legal threats.
- A rule covers address changes on orders that have already shipped.
- Custom escalation, follow-up and ticket suggestion messages are written.
- Escalation timeout and follow-ups are set.
- Working hours, outside hours behaviour and the outside hours message are set.
- Someone owns the Needs Agent queue each day, with WhatsApp handoffs first.
- The team knows when to Claim, when to Take Over and when to Release to AI.
- Each week, escalations are reviewed and the knowledge base updated.
Common questions
What does the customer hear while they wait for a person?
In Fyncall, the customer first gets your Escalation message, and the AI keeps helping with anything it can answer while they wait. If nobody replies within the escalation timeout, they can get up to 5 follow-up messages and an offer to open a ticket, all written by you or left on the defaults. Outside working hours, they see your Outside hours message, which should say when the team is back.
Can the AI issue a refund before handing off?
No. Fyncall’s AI checks refund eligibility and refund status, then your team issues the refund in Shopify. That makes refund requests a natural handoff point.
How does Fyncall decide who counts as a VIP?
The AI Escalation settings page does not define VIP criteria. VIP Customer Prioritization only moves high-value customers up when another trigger has fired, so check the setting in your Fyncall account.
Does Fyncall route escalations to specific agents?
The escalation settings cover triggers, timeouts, messages and working hours. Escalated conversations go to the shared Needs Agent queue, where a teammate claims or takes them over. For how one AI agent per channel fits with a small team, see ecommerce AI agent support.
