Use risk to order the review queue

Question for this chapter

Which of the 12 orders should a person review first?

Try it

Open the risk_assessment pipeline under refund_ops and select Run. The score combines past refund count, order amount, and customer tier on a 0.0–1.0 scale.

• Below 0.3: low
• 0.3 through 0.6: medium
• Above 0.6: high

Success looks like this

risk_assessed_orders contains 12 rows.

Inspect the five highest scores.

The maximum 0.600 is still not high; the code requires a value above 0.6.

Trace one order relationship

Run ontology_materialization. It produces 10 customer rows, 12 order rows, 6 refund-request rows connected to current orders, and 18 relation rows. Run this query in Graph Explorer:

MATCH path=(r:refund_request)-[:requested_for]->(o:sales_order)-[:placed_by]->(c:customer)
WHERE o.order_id = 'ORD-2025-0001'
RETURN path

The result should show RF-2025-005 → ORD-2025-0001 → CUS-001, placing risk and customer context around the same order.

Next decision

You have five orders to review first. Next, retrieve policy for a refund reason instead of deciding from the risk score alone.