Where they started
A third party logistics operator running two warehouses in Melbourne’s west for about forty customers, mostly consumer goods and building products. Orders arrived in every format those customers used. Emailed PDFs, spreadsheets, EDI from the two biggest, and exports from three different online store platforms.
Four people spent most of their day keying orders into the WMS. Errors made at entry showed up as pick errors, credit claims and angry phone calls.
What we found in discovery
Sitting with the order entry team for two weeks showed that the formats were messy but stable. Each customer sent orders the same way every time. The team had informal rules for each one written on sticky notes and in their heads.
About one order in ten needed a judgement call. A product code that had changed, a delivery address that did not match the account, a quantity that looked wrong. Those were the orders worth a human’s time, and the other nine were typing.
What we built
An intake workflow that reads each inbound order, identifies the customer and format, extracts the lines and creates the order in the WMS. Anything below a confidence threshold, or matching one of the team’s exception rules, goes to a single review queue with the reason attached.
Three things mattered.
- Thresholds were set against their own history. We ran the workflow over six months of past orders before anything went live and set the auto-create threshold where its error rate matched the team’s.
- The sticky notes became rules. The informal per customer knowledge was written down and encoded, and the team can change it without a developer.
- Exceptions went to one queue instead of an inbox. Exceptions surface in one place, in priority order, with what needs checking already highlighted.
Where it landed
About ninety one percent of orders now reach the WMS without a person typing them. The four person entry team became a customer service team of three plus one exception handler. Pick errors traced to order entry fell by roughly seventy percent.
What we would do differently
We left EDI for the two biggest customers out of scope because it already worked. It did, but it broke in a different way to everything else and the team still had to watch it separately. The cleaner design was to bring it into the same queue in phase one.