Can I get useful warehouse analytics if my data is a mess?

Useful warehouse analytics when your data is a mess

Yes. Most warehouse data is messy, because every system keeps it its own way. Useful analytics for warehouse operations doesn't need it perfect first: it needs the few measures that matter on the floor, checked against the floor itself. The checking also shows where the records are wrong, which is often the first finding.

What a mess usually looks like

By warehouse analytics we mean the operations kind: reading what the WMS, the clocking system and the rest record about the work, rather than analytics on a data warehouse. The mess tends to be some mix of these:

The same product under two codes, one in the WMS and one in the ERP.
Locations renamed after a re-rack, with the old names still in the system.
Tasks confirmed in a batch at the end of the shift rather than when they were done.
Clocking records that don't match the rota, with agency staff on a separate timesheet.
A spreadsheet as the only record of the work the WMS doesn't hold.

None of it is unusual. None of it means the data can't be used, either: it means the data has to be read by somebody who knows which parts matter and how to check them.

Why it doesn't have to be cleaned up first

A project that sets out to clean every record before anybody reads a number tends to run long and reach the floor late. The quicker route starts from the questions the warehouse gets run on and puts right what each one needs, in the order you need the answers.

Knowing what matters is most of the work. For a question about your people, that's the paid hours, the tasks and the off-system work. For a question about stock, it's where each location sits, how often each line is picked and how full each face is. The rest can wait.

Nor does every minute have to be accounted for. Getting most of the day attributed to the right job and the right client is enough to manage well, since the last few minutes cost more to capture than they tell you.

The mess is often the finding

In two warehouses we've worked in, labour utilisation read around 42 per cent on their own systems. The records weren't wrong so much as incomplete: the off-system work never reached them. Counted in, both read 65 per cent.

That's a difference of 22 and 23 points. On a 7.5-hour shift, 22 points is 99 minutes, so each operator had more than an hour and a half of work a shift that the records never showed.

Two warehouses we've worked in, measured on their own systems and then with the off-system work counted.

The same goes for stock. Reading how often each line is picked against where it sits turned up eight sets of brake pads in a full pallet location, where they'd been for 223 days without anybody noticing.

AN ILLUSTRATION

Where the stock sits against how often it's picked

An illustration of a pick-face map. The flagged location is a real finding from a parts warehouse we've worked in.

When two sets of numbers disagree

In one distribution business we've worked with, the depots kept saying they couldn't get stock, while the distribution centre's fulfilment report read close to perfect every day. Both were right. At the end of each night the system asked whether to carry over the orders that couldn't be picked. Deleting them took one button, after which the report ran on what was left.

Nobody had done anything the system didn't allow. The two numbers measured different things. It took somebody reading both, against what the depots actually received, to see it. A number that needs a caveat before anybody acts on it gets checked by hand, which is where a lot of management time goes.

Where AI comes in

AI is only as good as the data going into it. That's why nothing in Fettle reads the raw feed from each system, not even Albie, the assistant that answers questions about your operation: the data is standardised and checked first, against the rules you give us in the first week on site. The checks are part of the work rather than a project before it, which is why a messy system isn't a reason to wait.

Where to start

Pick one question the warehouse gets run on rather than the whole dataset.
List the systems that hold its answer: the WMS, the clocking system, the agency's timesheets and whichever spreadsheet fills the gaps.
Check each one against the floor, by walking the locations and holding a shift's records up against the clock.
Put right what that question needs, then move on to the next.

Fettle starts the same way, with two or more days on site in the first week to map the layout, walk a day in the life of the work and write down the service rules before anything reads the data.