YundaDeskGood support scripts are not memorized. They are consolidated.
Many cross-border e-commerce teams treat scripts as onboarding material: a shared document, a few approved replies, and a lot of copy-paste. That helps, but it misses the bigger win. A useful script should also become a knowledge base entry that AI support can cite, test, and improve under human control.
The same template should serve agents and AI: agents answer faster, AI catches routine questions first, and corrections become owner-approved learning suggestions.
A good script does not freeze every word. It helps the team answer accurately under pressure, keep the tone steady, and know when to hand off.
– YundaDesk Support Team
How to use these templates: capture, teach, improve
Do not keep scripts only in a training deck. Put them into the knowledge base by scenario, then record when to use each answer, what to collect, and when to hand off.
Low-risk questions can be answered directly by AI support. Refunds, compensation, price changes, complaints, and review threats should only receive de-escalation, information collection, and human handoff. The team should maintain one approved source of truth, not one answer for agents and another for AI.
Opening and welcome
A good opening says: we saw you, we can start checking, and a human will step in when needed.
Hi, thanks for reaching out. I will take a look first, and if this needs a human agent, I will hand it over to our team.
Hi, I am here. You can send your order number, email, or a short description, and I will check the current status first.
Got it. I will check the available order and policy information first. Anything involving refund, compensation, or approval will be reviewed by a human.
Thanks for waiting. I will try to give you a clear answer in one go. If anything is missing, I will tell you what we need.
Store these under “opening and information collection”. YundaDesk AI support can follow the customer’s language, but the handoff rule still needs to be explicit.
Shipping and orders
Shipping questions are frequent and repetitive, so AI can answer first. The danger is overpromising. Cross-border shipping includes customs, carrier transfers, and local delivery handoffs, so scripts should explain the flow without guaranteeing outcomes.
| Scenario | Script |
|---|---|
| Shipping time | Your order will be processed according to the timeline shown at checkout. Once it ships, we will send tracking information. You can send your order number and I will check the status first. |
| Shipping time | If this is a pre-order, customized item, or sale-period order, processing may take longer than usual. Please send your order number and I will check the next step. |
| Address change | If the order has not entered warehouse processing or carrier pickup, we will try to help. Please send the order number, new address, and recipient details. I will hand it to a human agent for confirmation. |
| Address change | If the package has shipped, address changes depend on the carrier and destination rules. We can record and check the request, but cannot guarantee success. |
| Shipping nudge | I understand you want the package soon. I will first check whether the order is in the processing queue. If there is an exception or delay, I will hand it to the team. |
| Shipping nudge | We have received your shipping nudge. The team will review order status, inventory, and fulfillment batch. We will not create a duplicate order or skip required steps. |
A useful rule: AI can explain and collect information, but it should not say “it will ship today” or “the address will definitely be changed.”
Discounts and pricing
Coupon issues often involve campaign rules, excluded products, stacking limits, regions, or expiry. Explain likely causes first, then ask for evidence.
Coupon did not apply
I understand you want to confirm whether the discount was applied correctly. Common reasons include minimum spend, ineligible items, expired codes, or non-stackable discounts. Please send the coupon code, cart screenshot, and account email so I can check first.
Checkout amount looks wrong
Please send the product link, cart screenshot, coupon code, and checkout amount. If this looks like a display or system issue, I will hand it over to a human agent.
Customer price comparison
Prices may differ by channel, campaign timing, and region. We can verify the current official-store price and coupon rules, but cannot guarantee matching every third-party channel price.
Do not turn this into a price-match promise. If you have a price protection policy, write the conditions clearly. If not, verify official-store rules and hand off special cases.
After-sales and emotion
After-sales conversations need more than facts. Customers need to know they have not been ignored. Acknowledge the emotion, collect facts, then state the next step.
Delay apology
Sorry for the wait. I understand the delay can affect your plan. Please send your order number and I will check where the process is blocked. If needed, I will pass the full context to the team.
Tracking has not moved
I understand why this status feels worrying. For cross-border packages, tracking may pause during customs, transfer, or local delivery handoff. Please send your order number and I will check the latest record.
Complaint de-escalation
I am sorry this experience did not meet your expectations. We will review the order, conversation history, and relevant policy. To avoid an inaccurate promise, I will hand this to a human agent first.
High-risk and escalation
High-risk conversations are where AI needs to know exactly where to stop. Refund requests, compensation claims, review threats, platform complaints, lawyer mentions, and regulatory language should trigger handoff.
Risk tiers before scripts enter the knowledge base
Separate "can answer directly" from "must hand off"
| Scenario | Script |
|---|---|
| Refund request | I understand you want this resolved quickly. Refunds need human review based on order status, item condition, and after-sales policy. I will record your request and hand the context to the team. |
| Compensation request | I am sorry for the inconvenience. Compensation or amount adjustments require human approval, so I cannot promise a specific amount. Please provide the order number and issue details, and we will send this to the team. |
| Review threat | I understand you are very unhappy, and thank you for telling us directly. We will prioritize reviewing this case. To avoid an incorrect promise, I will hand it to a human agent immediately. |
| Platform complaint | We will take your feedback seriously. Please provide the order number, issue details, and screenshots. I will send the full information to a human teammate for review. |
Do not include lines like “we will definitely refund you”, “do not leave a bad review and we will compensate you”, “I will change the price now”, or “the system has approved this automatically.” Use: received, understood, needs approval, handing off with context.
Appendix: turn one script into a knowledge base entry
Example: coupon did not apply
- Title: How to respond when a coupon code does not apply
- Customer phrasing: discount code not working / coupon cannot be used / checkout did not reduce the price
- Standard answer: Explain minimum spend, eligible products, expiry, and stacking rules. Ask for coupon code, cart screenshot, and account email.
- AI can answer directly: Explain common reasons and collect information.
- Must hand off: Customer asks for price adjustment, compensation, says they were misled, or threatens a complaint.
- Learning rule: New campaign restrictions from agents become learning suggestions and take effect only after owner approval.
- Rollback point: After the campaign ends, review whether this entry should be retired or changed into a historical rule.
Scripts should not turn support into copy-paste work. They make each reply traceable, bounded, and reviewable. Let AI catch routine questions first, route high-risk cases to humans, and fold human experience back into the knowledge base. That is how a temporary support document becomes a real operating asset.