A useful support reply needs more than fluent wording. The team must be able to check the source policy, identify what is still unknown, and give a person ownership when a decision needs judgment. Here is a reproducible example from the Anti Sandbox preview.
1. Start with the customer’s actual question
Hi! I ordered yesterday. When will my order arrive in Riyadh?
Open the demo and select Layan’s conversation in Inbox. The message asks about a specific order. A general shipping policy can explain the usual delivery window, but it cannot prove when that order was dispatched or where it is now.
2. Inspect the source before accepting the suggestion
Riyadh: 2–3 working days. Jeddah: 3–5 working days. Send tracking after dispatch.
Select Shipping policy beside the prepared suggestion to open Knowledge. Check that the destination and time unit match the question. These are fictional store rules for this example, not Anti Sandbox’s own shipping terms.

3. Review what the reply says—and what it cannot establish
Hi Layan! Standard delivery to Riyadh takes 2–3 working days. We’ll send your tracking link when the order ships.
The prepared reply follows the sample policy and makes no claim that this order has shipped. Select Use suggestion to copy it into the reply box. Review the wording before the demo’s Send reply action; nothing is sent outside the preview. In a connected workspace, verify live order facts before promising an arrival date.

4. Give a person the decision when the policy is insufficient
Try a separate handoff exercise: select Omar’s conversation, choose Internal note, record what needs checking, and assign the conversation to Noura. The sample returns policy requires a person to check stock and approve the next step. This demonstrates manual triage in the preview, not an automated escalation test.
Suggested internal note: Customer requests a different size. Check stock and confirm exchange eligibility before approving. No exchange or refund has been promised.
What to test with your own business knowledge
- Known answer: does the reply use the correct, current policy?
- Missing facts: does the agent avoid inventing a tracking link or delivery date?
- Human decision: does an exception reach the right person with useful context?
- Arabic and English: do both versions preserve numbers, conditions, and meaning?
- Connected channel: have provider setup and a real end-to-end test succeeded?
What is available, and what does this preview prove?
Anti Sandbox’s current product focuses on the helpdesk, knowledge-grounded replies, and human escalation. External channels need provider accounts, configuration, and testing. This preview illustrates the experience; it is not evidence of live channel connectivity, commerce actions, response accuracy, or customer outcomes. Workflow and App studio screens in the preview do not establish production availability.
Try the example, then assess your own setup
Use the demo to explore the interface. For an evaluation, start with a small set of approved policies and test questions covering a known answer, missing information, and a human decision. Record incorrect answers and handoff failures before expanding the scope.
Open the interactive preview