When an AI handling inquiries hits a special price request, contract terms or a complaint, can it hand off to a human automatically with the chat history?
Yes, and it should be a hard requirement when you choose a vendor. A proper AI Sales Agent detects these situations using rules you define, routes the conversation to the right person, and passes along the full transcript, buyer profile, summary and trigger reason. The human picks up the same conversation; the buyer never repeats themselves.
Why treat handoff as a must-have?
Because buyers treat what the AI says as what you said. In Moffatt v. Air Canada (February 2024), British Columbia’s Civil Resolution Tribunal held the airline liable for fare-policy information its chatbot got wrong and rejected the argument that the chatbot was a separate entity (BC Civil Resolution Tribunal, 2024).
China’s recommended national standard GB/T 47746–2026, on coordination between human and AI customer service, was published on May 25, 2026 and took effect on September 1, 2026. It calls for clear human entry points, priority handoff for complex issues, synchronized identity and history at handoff so customers need not repeat themselves, and company responsibility for AI replies (Xinhua, 2026). Buyers want it too: Gartner’s May 2026 survey of 645 B2B buyers found 69% would check AI-provided conclusions with a sales rep (Gartner, 2026).
Which situations must go to a human?
- Special pricing below your floor, extra discounts or rebates.
- Payment terms: open account, L/C changes, installments.
- Contract terms and exclusive distribution.
- Complaints and quality disputes.
- Legal and compliance questions.
- The buyer explicitly asks for a person.
- The AI fails to answer twice in a row, or the buyer is frustrated.
- Orders above a size threshold.
The principle: hand off rather than improvise. See how guardrails work.
How should trigger rules be configured?
| What it is | Example | |
|---|---|---|
| Intent | What the buyer is trying to do | Negotiating, complaining, asking about credit terms |
| Tag | Labels on the inquiry or account | High intent, existing customer, key distributor |
| Conversation signals | How the chat is going | Two failed answers, negative sentiment |
| Message type | Form of the content | Contract file, voice note, defect photo |
| Time | When it happens | Business hours: immediate; night: inform and prioritize |
Start strict and loosen over time: in the first weeks it is cheaper to over-escalate than to let the AI improvise on price or terms. Review the handoff log weekly and turn recurring safe questions into data the AI can answer.
Combine conditions with AND/OR and configure them per channel, market and customer tier. Keyword-only triggers misfire often and miss requests phrased differently.
What should be passed to the human?
- The full transcript, including images.
- Buyer profile: company, country, buyer type, history.
- A short summary of what the buyer wants.
- The sales stage.
- The trigger reason.
- A suggested reply draft.
Example: a German distributor asks for exclusivity
After confirming voltage and CE for a 100-ton hydraulic press, the buyer asks: “If we order 20 units a year, can we be your exclusive distributor in Germany, and can you take 15% off?” That hits both the exclusivity and special-pricing rules. The AI does not answer either point; it tells the buyer the sales manager will confirm and when, while the manager receives a card with profile, summary, stage, reason and a draft reply.
This is what the standard’s “synchronize history, don’t make the customer repeat themselves” looks like in a sales context. A handoff that arrives as a bare notification (“new chat needs attention”) forces the rep to scroll through the whole conversation and often to re-ask questions the buyer already answered, which is exactly the experience buyers abandon.
What should the AI tell the buyer during handoff?
Three things: who will take over, when, and what the buyer can do in the meantime. Never pretend to still be working on it, and never go silent.
- Business hours: “Our sales manager needs to confirm this and is reviewing your message now.”
- After hours: “Our sales manager will reply first thing after 9 a.m. Beijing time. Could you leave an email so we can send materials too?”
- Complaint: “Sorry about this. I’ve passed your photos and order details to our after-sales lead, who will contact you directly.”
What if no one is online?
Tell the buyer honestly when a person will reply, collect contact details, keep answering non-committal questions from your data, and alert staff before the WhatsApp 24-hour window closes—outside it only template messages are allowed (Meta developer docs). See capturing night-time inquiries.
How do you test handoff before signing?
- Ask for 30% off: does the AI decline to agree and route it?
- Ask for exclusivity: does it reach the right person?
- Send a damage photo with a complaint: routed to after-sales?
- Say “I want to talk to a human”: immediate response?
- Ask twice for a spec that is not in the data: handoff instead of a guess?
- Inspect what the human receives: transcript, profile, summary, stage, reason, draft?
- Does the buyer stay in the same thread after the human takes over?
- How fast does a rule change take effect, and who can change it?
How does DeepSales handle handoff?
DeepSales, Singoo Cloud’s done-for-you AI Sales Agent, turns your top reps’ judgment into handoff conditions combining intent, tags, conversation signals, message type and time (AND/OR), per channel, market and customer tier. When triggered, a card with buyer profile, summary, funnel stage, trigger reason and suggested reply is pushed to Feishu (DingTalk and WeCom are planned). The human takes over the same conversation in the Singoo Cloud platform with full history. It also alerts on WhatsApp windows about to close. Negotiation and final decisions always stay with people; quote drafts on higher tiers require human approval. If your team does not use Feishu, factor that in. Book a demo to see a special-pricing handoff on your own catalog.