Auto-Execution Rules: when the AI can act on its own

Auto-execution rules decide which actions the AI is allowed to take on its own and which need a human in the loop. The system is intentionally conservative by default — anything that moves money or changes the order state requires explicit configuration.

What can be automated

OrderWise can execute four classes of action against your Shopify store:

• Read-only lookups — order status, tracking, customer history. Always automatic; no rule needed.
• Refunds — partial or full. Gated by refund cap and confidence threshold.
• Cancellations — cancel an unfulfilled order. Gated by cancel cap and fulfillment status.
• Returns — initiate a return request. Gated by return window and item condition (if a condition field exists in your return policy).

Safe defaults to start with

When you first turn on auto-execution, start with these limits:

• Refund cap: 20 EUR or 5% of the order total, whichever is lower.
• Cancel cap: orders placed in the last 12 hours that haven't been fulfilled yet.
• Return window: 14 days from delivery (the EU statutory minimum).
• Confidence threshold: 0.8. Below 0.8, the AI hands off to a human regardless of the rule.

These caps mean the AI can handle the long tail of small, obvious cases without you ever seeing them, while escalating anything that could materially affect your bottom line.

The escalation path

When a rule blocks the AI from acting, the conversation goes into the human-handoff queue inside the embedded admin. The merchant sees:

• The customer message.
• The action the AI wanted to take.
• The rule that blocked it (cap exceeded, low confidence, etc.).
• A one-click approve/decline button that runs the action with the AI's pre-filled inputs.

Approval triggers the same Shopify mutation the AI would have called, plus a system message in the conversation logging the human decision.

When to raise the caps

Watch the human-handoff queue for two weeks. If the same kind of action keeps showing up and you keep approving it without changes, raise the relevant cap. If you keep declining a certain action type, the rule is working — keep the cap low.