A precision workshop: parts lag reduced from three days to twenty minutes
A precision engineering workshop in the Midlands was procuring parts on a three-day lag because the shop-floor and the office did not share a stock view. The owner called it "the gap we keep paying fo
The reading
Twenty-two machinists, four CNC centres, a works manager who had come up through the shop floor. The workshop made small-batch precision parts for aerospace and medical clients. Parts were procured by the office against job cards passed up from the floor at the end of each day.
The gap was structural. A machinist noticed, at three in the afternoon, that a particular grade of steel was running low. The job card went up at six. The purchasing office saw it the next morning. The order went out midday. Delivery was the day after. The machine stood idle for two days or ran on substituted stock.
The owner had tried, twice, to fix this with a shared spreadsheet. Both attempts collapsed within six weeks.
The job card lag
The spreadsheet attempts failed for the same reason: machinists would not update stock on a shared document in the middle of a job. The friction was too high; the perceived benefit, to them, was too low. The works manager had accepted this as a human fact.
We thought the human fact was correct and the solution was wrong. Machinists would not update a spreadsheet. They would scan a barcode on a bin on the way back from the toilet.
- 22 machinists, 4 CNC centres, no shared stock view
- 3-day average lag from stock-low to delivery
- 2 previous spreadsheet-based fixes abandoned
- Approximately 14 machine-hours lost per week to stock-out
- Purchasing office working from day-old data
What we built
We fitted each stock bin on the floor with a small e-paper label showing a barcode and the current count. A pole-mounted scanner at the end of each aisle let a machinist decrement by one, five, or ten with a single scan. The purchasing system saw the decrement within twenty seconds.
The technical choice: the scanners were deliberately dumb. They did not have screens, did not have Wi-Fi settings, did not require login. A machinist with oily hands could use them without thinking. The intelligence lived in the backend, which was a small Go service with a Postgres database and a webhook into the existing purchasing software.
Automatic reorder thresholds were set per part, tunable by the works manager through a web page he could reach on his phone. The office saw reorder suggestions in real time instead of job cards the next morning.
The numbers at ninety days
| What | Before | After |
|---|---|---|
| Stock-low to delivery lag | 3 days | 20 min to order, 1 day to delivery |
| Machine-hours lost to stock-out | 14/week | 2/week |
| Job cards processed by purchasing | 180/week | 0/week |
| Stock count accuracy | ~80% | 99.1% |
| Emergency carriage charges | 1,400/month | 180/month |
The scanner, not the spreadsheet
The owner's two previous attempts had failed not because the shop floor was unwilling, but because the solution asked machinists to behave like office workers. The scanner asked them to behave like machinists.
That is the one-week training test in its purest form: a new machinist on day one scans the bin without being shown twice. The system fits the person, not the other way around.