Ken is the plant manager at a job shop with 70 employees, 14 CNC machines, and a backlog that swings between 3 weeks and 11 weeks. He called me because his promise dates were fiction. He would quote 6 weeks, deliver in 9, and apologize in every email.
His team had already built a beautiful bill of materials for all 340 active parts. That was the frustrating part. The data looked good. The schedule still lied.
Assumption one: if the bill of materials is right, the schedule will be right
Accurate parts lists tell you what to buy. They tell you nothing about how long a job actually sits on a machine. Ken’s routing said 4.2 hours for a part that consistently took 7 hours to run, because setup time had been averaged across a batch of 50 and he mostly ran batches of 8.
We pulled 90 days of job history and compared quoted time to actual. On the 12 parts that made up 61% of revenue, the gap was between 40% and 90%. The parts list was fine. The time model was fantasy.
Assumption two: every job deserves a full work order
Ken’s shop was creating full work orders for repair work, one-off fixtures, and prototype jobs worth $300. Each work order took a planner 25 minutes to create and 40 minutes of operator time to report against. On a $300 job, that overhead is the profit.
We split work into three lanes instead:
- Repeat production jobs over $2,000 got full work orders with routing, material issue, and time tracking.
- Repair and rework under $500 got a single line item against the customer order, no routing, no material issue.
- Prototypes got a job cost code so engineering could see the spend without touching the production schedule.
Planner time dropped by roughly 30 hours a month, and the schedule stopped being polluted by jobs that never belonged on it. Adding detail to every job made the important jobs invisible.
Assumption three: operators will update the system if you ask them nicely
They will not. Not because they are difficult, but because they are standing at a machine with greasy hands and a chip conveyor that needs clearing. Any step that requires walking to a terminal will be skipped by Thursday.
Working with ERPNext Consulting, we put a tablet on a swing arm at each of the six busiest machines and cut the job reporting screen down to three buttons: start, pause, complete. Nothing else on the screen. Job reporting accuracy went from roughly 60% to 94% in seven weeks.
That accuracy is what made promise dates believable. Ken now quotes 5 weeks and delivers in 5 weeks, and his on-time delivery sits at 91%, up from 63%.
The lesson generalizes. Before you buy scheduling software, measure your actual run times against your quoted times on your top 10 revenue parts. If the gap is over 30%, no system will fix your dates. Fix the number first, then automate it.