AI-assisted WMS: what really changes in the warehouse?

Records first, then AI

In a warehouse the raw material of AI is the record: which goods are in which aisle, level and bay, when they came in, when they left and who scanned them. If this data is not kept by scanning and on time, no forecast can be trusted. So the first job of a good WMS is to make the location hierarchy (aisle, level, bay) and scan-based receipt and issue solid.

Where does AI create value in a warehouse?

• Document reading: drafting a receipt record from a delivery note or advance notice.
• Location suggestion: proposing a likely shelf position from the goods’ characteristics and how often they leave.
• Dock planning: seeing dock occupancy ahead of time from the arrival times of incoming vehicles.
• Waiting analysis: measuring waiting at the dock and making the bottleneck visible.
• Value-added services (VAS): drafting work orders for tasks such as labelling and packing.

Every suggestion goes to a human approval; the warehouse worker or shift lead decides whether it is applied.

Bonded and customs-controlled stock

Goods held in a bonded warehouse or customs-controlled area follow different rules from ordinary stock: each item has an MRN, a status and a deadline counter. When these fields live inside the WMS record, overdue deadlines become visible early. For high-impact steps such as the release of goods under customs control, approval by two people (the four-eyes rule) reduces errors. For the customs side see the guide to AI in customs.

Evidence and archive

Disputes in a warehouse are usually settled with evidence: who received which goods, when and in what condition. An evidence file and a read-only archive are a base layer that is independent of AI but also makes AI outputs auditable. Recording the AI suggestion together with the person’s final decision lets you understand the reason afterwards.

What to ask when choosing

• Are records kept by scanning, or typed in afterwards?
• Is there mobile terminal support? Do the handheld and the phone write to the same record?
• Do dock and vehicle arrival data talk to the transport system?
• Is an AI suggestion applied without approval?
• Is the interface in the worker’s language?

Common mistakes

• Entering records by hand afterwards: without scanning, stock information drifts from reality and forecasts become meaningless.
• Not measuring dock waiting time: a bottleneck that cannot be seen cannot be improved.
• Managing bonded and ordinary stock by the same rule: deadline counters and release approval are separate for goods under customs control.
• Applying an AI location suggestion unchecked; a shift lead’s approval catches mistakes early.