Why couldn't we do this before?
The data was there for years, in the WMS, the clocking system and the rest. Reaching it was the hard part, because each system stores it differently and joining them meant software few warehouses could justify. AI changed both halves: the software is quicker to build and the answer quicker to find, provided the data going in is checked.
What this is
By this we mean seeing what needs attention in your warehouse today, across the systems you already run, with the off-system work counted and your own rules applied. Nothing gets replaced to do it.
The questions haven't changed. Will we deliver today's commitments, are we using our people effectively, is the warehouse set up to work efficiently and are we making money on the work we do: managers have asked those for as long as there have been warehouses. What's changed is how much it takes to answer them from the data.
The data was always there
The WMS has timestamped every task it handed out since the day it went live. The clocking system has recorded who came in and when. The ERP holds the orders and the contracts, while the supervisors' spreadsheets hold most of the rest. In most warehouses that's years of it, sitting in systems that each do their own job well.
Take the time between clocking on and starting the first task. The clock-on time sat in one system and the first task's timestamp in another, both recorded every shift for years. Putting one against the other for every operator in one warehouse we've worked in showed 572 hours in a single week. That's the same as the whole paid week of more than fourteen people on 40-hour contracts, gone before the first task of each shift. No report had shown it.
ONE WEEK, ONE WAREHOUSE
Time between clocking on and the first task
572 hours the same as the whole paid week of more than fourteen people on 40-hour contracts
572 hours in one week between operators clocking on and starting their first task, drawn as fourteen full 40-hour weeks and part of a fifteenth.
One week in a warehouse we've worked in.
Why reaching it was hard
- Every system keeps its data its own way. Different product codes, different clocks and different ideas of what counts as a task. Joining them meant somebody writing the links for that warehouse, then keeping them working every time a system changed.
- That software cost more than one warehouse could justify. So the choice was the vendor's standard reports, a wait for the next release or a project with a data team. Most warehouses sensibly stayed with the reports.
- Somebody had to know which questions to ask. The measures that matter on a warehouse floor, and how to make them fair, come from time spent on one rather than from a data model.
While he was an IT Director, our founder was building his own views of the warehouse's problems in Tableau from 2010, to answer the questions the standard reports left open. It worked, because the data was there. It also took an IT Director who knew the warehouse, the systems and the finance side, with the time to build each view by hand, which is why most warehouses never had anything like it.
What AI changed
It changed both halves of the problem.
Reaching the data is the first. The software that joins one system to another, standardises what comes across and checks it can now be written in a fraction of the time it used to take, so a small team that knows warehouses can build what one warehouse needs in weeks rather than running it as a project.
Reading it is the second. Once the data is joined and checked, you can ask a question of it in plain English and get the number back with where it came from. Ask Albie, the assistant inside Fettle, how many hours went on replenishment last week and the answer comes back in one sentence, without anybody building a report first.
AI is only as good as the data going into it. That's why nothing in Fettle reads the raw feed from each system: the data is standardised and checked against your own rules first. Only then does anything read it, Albie included.
What AI didn't change
Knowing which questions matter and how to measure people fairly still comes from the warehouse side rather than from the software. The numbers still only pay if somebody acts on them, which takes a manager's judgement and a team that trusts what it's being shown. What AI has changed is how quickly that knowledge can be built into something a warehouse uses every day.