AI Assistant & MCP

The Rule Recommender

Nightly AI analyst that watches your shipping patterns and surfaces automation rule suggestions you can accept in one click.

What the Rule Recommender does

Every night, an AI analyst pass runs over the last 90 days of your orders, shipments, and rate selections. It looks for repeating patterns — things you do manually over and over — and proposes automation rules that would do them for you. Example findings: "You always pick UPS Ground for orders over 30 lb to the West Coast" or "You add signature confirmation to every order over $400".

Reviewing suggestions

Open Admin → Automation → Suggestions to see the feed. Each card shows: the proposed rule (in plain English), the evidence (how many past orders match the pattern), the projected impact (e.g. "Would have auto-shipped 142 orders last month, saving ~3.5 hours"), and the exact conditions and actions. You can click into any suggestion to see the matching historical orders.

One-click accept

If the suggestion looks right, click Accept. ShipWave creates a real rule in Admin → Automation → Rules with the proposed conditions and actions, set to preview mode by default. In preview mode the rule runs on incoming orders but only logs what it would have done. After a few days of clean previews, flip it to active.

Dismissing with a reason

Not every suggestion is right. Click Dismiss and pick a reason: "Not actually a pattern", "Already handled elsewhere", "Edge case I review manually", or "Other" with a free-text note. The recommender uses these signals to tune the next cycle — it will stop surfacing similar patterns and learn what you consider out-of-scope.

Tuning the cycle

By default, the recommender runs nightly and surfaces up to 5 suggestions per workspace per day so the feed stays scannable. You can change the cadence (daily, weekly, monthly) and the max suggestions in Admin → Automation → Suggestions → Settings. Setting it to weekly is a good fit for mature ops teams whose rule library is already mostly built out.

FAQs

How long until I see my first suggestion?
The recommender needs at least 30 days of order history with at least ~50 orders to spot reliable patterns. New workspaces will see the feed populate after the first nightly run that crosses that threshold.
Does it ever create rules without my approval?
No. Every suggestion requires a human accept. There is no "auto-accept" toggle, by design.
Can I see why a rule was suggested?
Yes. Each card has an "Evidence" section showing the matching historical orders, the per-condition match rate, and the confidence score. Click any order ID to open it in a new tab.
What happens to dismissed suggestions?
They are stored so the recommender does not re-suggest the same pattern. You can review and un-dismiss them under Suggestions → Dismissed.
Does this work with marketplace channels?
Yes. Patterns across Amazon, eBay, Walmart, and other channels are all in scope. Channel-specific rules (e.g. "Always use FBA for orders under 2 lb on Amazon") are common suggestions.

More in AI Assistant & MCP