When goods are received, a delivery is checked against the packing slip and the order, and the receipt is processed in ERP or WMS. The question is whether AI can take over that process. The answer is partly, and the part that remains is not coincidentally the part where someone needs to physically stand next to the goods.
Booking-in itself is a structured task: there is an order, there is a packing slip, and the system expects a match between what was ordered and what comes in. That comparison is exactly the kind of work software systems are good at. The structuredness axis therefore scores high, and recognising data on a packing slip or invoice via image is a task AI can already handle today. Volume also plays a role here: a warehouse processing a hundred lines a day has more to gain from automatic recognition than a location that receives two deliveries a week.
Compliance also works in favour of automation: if a delivery must be recorded according to a fixed protocol, with timestamp, quantity and reference number, that is something a system does more consistently than manual entry at the end of a long shift. Anyone who already has the order itself placed automatically will see that the link between purchase orders that are automatically placed with suppliers and the receipt of that same order is a logical next step: the system already knows what should be arriving, and can match the incoming data against that directly.
The three axes that are decisive here are physical, error costs and structuredness, and the first two hold back what the third makes possible. Physical scores low because someone is needed to open boxes, count quantities, spot damage and smell whether something has spoiled. A pallet with eighty boxes of which three have a torn wrapping is not noticed by a camera unless that camera has been specifically set up and trained for it, and even then the judgement of what counts as acceptable damage often remains human work.
Error costs score low for a different reason: a wrongly booked-in receipt leads to stock that does not add up, which weeks later can feed through into a wrong delivery promise to a customer or a production stoppage. Those costs are not always immediately visible, which makes the risk greater rather than smaller. That is why a judgement margin of 3 remains here: a system can flag a discrepancy, but whether that discrepancy is acceptable, what should be done about it with the supplier, and whether the goods are accepted after all, remains a decision with a reason attached.
A distribution centre receives a shipment of electronics. The packing slip states 200 units, the order also 200. A system can compare those two documents and automatically book in the receipt as soon as the quantities match. But the box that opens to reveal a broken screen calls for a physical check and a decision: accept, return, or set aside for assessment. That decision falls under oversight that approves or rejects with a reason, not under full takeover.
At a company that receives loose parts in small quantities, with a high risk of damage or spoilage, the physical share remains larger and there is little to gain from automating the booking-in alone. At a company that receives large, uniform batches from fixed suppliers, with stable packaging and few discrepancies, it is different: there, booking-in is often already largely system work, and the remaining task shifts towards spot checks rather than full inspection. The difference lies not in the sector, but in how predictable the shipment is and how high the costs of an error turn out to be.
What already works today is image recognition: a photo or scan of the packing slip, linked to the system, that automatically creates a receipt line. What is needed for that is scanning equipment on the floor and a link between that scanning moment and the ERP or WMS. Without that link, entry remains a manual step, even if the checking itself has already become smarter. The physical check on damage and correctness remains, in most cases, with a human, just as quality control on products largely remains human work as long as the assessment requires sensory perception.
The shift taking place here does not lie in replacing the warehouse worker, but in shifting their time: less time on keying in quantities, more time on assessing discrepancies. That pattern is broader than goods receipt alone. With access management of site and premises too, you see that registration increasingly runs automatically, while the assessment of exceptions remains with a human. If this shift affects staffing levels on the floor, its own legal requirements apply to that; that is not a conclusion that follows from a task analysis.
Whether the receiving process in your own company is for the largest part system work or for the largest part physical checking depends on the type of goods, the suppliers and the error-sensitivity of your stock. The free quickscan of twelve questions, without an account, gives an indication of what share of the hours in this kind of work profile can be taken over by AI today. The full work scan, which maps the work of an entire company task by task, is still under construction.
Vraag maar. Ik ken de kennisbank van deze site; wat ik niet weet, zeg ik erbij.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.