GUIDE · AI READINESS

Prepare the knowledge layer before the agent speaks for your store.

Shopify Inbox can use products, policies, knowledge-base material and published storefront content. That makes source quality and exception handling the real launch work—not generating more unreviewed copy.

1. Inventory the sources

List the current source for shipping, returns, product care, sizing, payment, privacy and support hours. Give every source an owner and review date; conflicting pages should be resolved before automation.

Source register: topic → canonical URL or document → owner → last reviewed → next review.

2. Define answer boundaries

Separate facts the agent may answer from decisions that require a person. Fraud, chargebacks, privacy, missing delivered parcels, legal exceptions and financial commitments need an explicit handoff.

Allowed: quote the published return window. Escalate: approve an exception, refund an order or accept liability.

3. Run a test set

Test normal questions, missing-information cases and adversarial wording against the expected source. Record unsupported claims, wrong policy selection and failures to escalate instead of judging only tone.

Minimum test: 10 common questions + 5 edge cases + 5 questions the agent must refuse or hand off.

4. Measure the launch

Capture a baseline before enabling automation. Compare unresolved conversations, reopenings, escalation quality and recurring topics as well as response speed; a higher automation rate alone is not proof of a better service.

Two-week gate: continue only if answer accuracy holds and unresolved or reopened cases do not worsen.

Turn this guide into a system tailored to your store.

The free audit generates a score, priorities and a sample macro without asking for access or an email address.

Generate my audit →