900 of nearly 1,000 materials were wrong, and the system wasn't what caught it
Nearly 1,000 materials. More than 900 of them didn’t match.
When we found out, go-live was 7 days away. The system was in trial operation, and the first batch of orders had already gone out.
What this cost in the end wasn’t the few days spent rebuilding the data. It was a batch of orders.
I was the project manager on that project. This post covers why the sheet hid the errors from the system, and the check I added before go-live afterwards.
1. The opening data was wrong across the board, and the system wasn’t what caught it
In that trial run, the WMS opening data and the PFEP material delivery sheet had both been imported. The system was running, orders were being released, and everything looked fine.
The system can’t see this kind of problem. It doesn’t know whether the numbers in the sheet are right. It just follows the sheet to stage and deliver.
It was the people on the floor who found it. They took their picking tasks to the real parts and compared: the quantity was wrong, the workstation was wrong, and the storage location was wrong too.
If only a few lines had been wrong, we could have fixed them and kept going. But as we dug in, it was the attributes of a large batch of materials that were wrong: how many, and where they should go, scrambled across whole sections. Of nearly 1,000 materials, more than 900 had problems.
That sheet was basically unusable.
It turned out the rows had shifted: one row’s numbers had landed on another row’s material. Each number looked plausible on its own, so nobody could tell from the sheet.
And before go-live, nobody had checked it against the real parts.
2. We stopped 7 days before go-live, and that batch of orders was lost
Go-live was 7 days away. But if we kept running on that sheet, the system would send the wrong quantities to the wrong workstations.
We told the business owner: the initial data is wrong, we can’t go live this time, and we’re falling back to the pre-go-live way of working.
I made that call, and it wasn’t a hard one to make.
The hard part is whether the business trusts you. Being able to stop at a moment like that comes down to the business’s trust in you as the project manager, and the track record you’ve built.
Rebuilding the data took a few days. We went through it one material at a time: which storage area each material belonged to, how it was palletized, how many per pallet, which delivery point it went to. Then we had the frontline warehouse staff check it again. The data got sorted out, but the production plan doesn’t wait. The batch of orders that was supposed to run in the new system had already missed its window and couldn’t be used. The go-live date had to be reset as well.
The date was fixed, the data couldn’t keep up, and in the end a batch of orders paid for the gap.
3. Before go-live, the opening data should go past the people who handle the parts
Looking back, the first people to find the problem were the ones who handle the materials every day.
The sheet was made by the planners. The planners didn’t know how the materials were actually grouped on the floor or which delivery point they went to. How many of a part fit on a pallet, which workstation it should go to: on the sheet, that’s just a number.
The people on the floor are different. They handle these parts every day. When a number on the sheet is off, one look at the real parts gives it away.
That time, though, by the time they saw the sheet, the orders had already gone out.
So afterwards we added a double check before go-live: once planning finalizes the data, the site logistics supervisor goes through it again. Mainly three things: material quantity, workstation, and warehouse.
For raw data, apart from having the floor go through it, I still don’t have another safeguard.
Put simply: put the people who find the problems in front of the problems.
If I did it again, I’d put that check before the first order of the trial run goes out. If the data hasn’t been past the floor, don’t release the orders.
This post adds to the PFEP post from the previous series: that one covered which columns this sheet should keep; this one is about who should check each number in it against the real parts.
Before your go-live, who signed off the opening data: the planners, or the floor?
Related notes
Published when written, never an ad: subscribe via RSS