How we built ICON Amsterdam an AI command center for delivery and return exceptions
A global fashion order does not end at checkout. It moves through a warehouse, one or more carriers, a delivery attempt, and sometimes a return portal, with each handoff creating a new place where the customer experience can break.
Most support systems wait for the customer to notice first. ICON Amsterdam needed the opposite: a layer that continuously watches the order journey and starts recovery while there is still time to protect the experience.
We built an Order Recovery Command Center that turns fragmented post-purchase events into one operational decision stream.
The system notices the exception before the customer does
The agent follows each order from Shopify through warehouse status, carrier scans, delivery attempts, non-collection, and the returns portal. Missing movement, inconsistent scans, overdue delivery windows, and stalled return states become operational exceptions instead of invisible data gaps.
Each exception receives context: order value, customer history, the promised delivery window, carrier behaviour on that lane, destination country, and available inventory. That context decides whether the situation needs a proactive message or an immediate operational fix.
The same logic covers returns. A parcel that arrives back at the warehouse but never moves to refund or exchange is treated as an exception with a clock on it, so the customer is not left to discover the delay alone.
Recovery is automated, but policy remains in control
A policy engine selects the next best action: send a proactive update, request a redelivery, offer store credit, dispatch a replacement, or escalate to a person when order value, fraud risk, or customer history crosses a defined threshold.
Support works from one risk-ranked queue instead of a wall of where-is-my-order tickets. Every exception arrives with the evidence already gathered, so the agent or the person resolving it does not need to reopen four systems first.
Leadership sees the operational picture: SLA performance by market, recovery rate, cost per exception, and which carriers or routes repeatedly create avoidable work. That view is what turns individual delivery failures into network decisions.
Orders, warehouse, carriers, returns
Delivery exception prediction
Policy-controlled next action
Exception queue and root-cause view
All data shown is adapted for use in this case study.
Result: a faster, clearer operation
5+
delivery and return exception types monitored
1
risk-ranked queue for the support team
24/7
proactive order monitoring
ICON Amsterdam moved post-purchase support from reactive ticket handling to proactive operations. Customers receive an answer or a recovery path earlier, while the team spends its time on the exceptions that genuinely require judgment.
Routine cases follow the same policy every time, so quality no longer depends on who opened the inbox that morning, and expensive edge cases reach a person with the context to resolve them in one pass.
The command center also turns individual delivery failures into network intelligence, showing which carrier, route, market, or policy creates the most cost and customer friction.