- Task
- Prepare a customer-support pilot with expected answers and handoffs.
- Test date
- Not tested
- Tool version / plan
- Not tested; confirm the vendor plan before a pilot.
- Cost basis
- Not measured. Record subscription, credits, failed attempts and review time during a real test.
- Failures / limits
- No execution has taken place. Product distortion, unsupported claims or incorrect policy answers are cases to test.
- Evidence type
- Sample
Inputs
- Current approved shipping and returns policies
- Synthetic questions without customer records
Proposed steps
- Write an expected answer or escalation for each question.
- Test known, missing, ambiguous and conflicting policy cases.
- Review handoff context and action permissions before enabling live actions.
Sample output / expected behavior
- Sample test: “My order arrived late; can I get a refund?” → ask for permitted context and follow the approved policy.
- Sample missing-knowledge case → acknowledge uncertainty and hand off to a person. These are expected behaviors, not observed tool outputs.
Human review checkpoints
- Confirm policy citations, uncertainty and human escalation.
- Validate identity and permissions for every proposed store action.
Sources and vendor demonstrations
These links describe vendor capabilities or public demonstrations. They do not establish this plan’s results.
English video outline
Draft for a future evidence-backed demonstration. Replace conditional segments only after recording a real test.
- Introduce the policy set and synthetic questions.
- Compare Gorgias and Tidio roles; explain that Sidekick serves the merchant admin.
- After a real test, show a redacted answer and human handoff receipt.
- Disclose plan, test date, failures and cost before discussing results.
