Where the margin actually goes
None of this is new to you. What is new is seeing it while the job is still running, instead of at month end when the costing sheet closes.
You quoted 38 seconds. The machine runs 44.
A six-second gap on a 200,000-piece contract is a year of margin. Without machine-level actuals, nobody finds out until the job is long closed.
Setup is the real constraint, and nobody measures it
In high-mix work, changeover eats more capacity than cutting does. Setup start and end almost never get logged honestly, so the scheduler plans against a number that isn't true.
OEE is believed, not measured
Most plants quote an OEE they arrived at from manual logs. When the machine itself is asked, the number typically lands 15 to 20 points lower.
Scrap is found at final inspection
Gauge and CMM results sit in one system, the machine in another. By the time a dimension drifts, a whole bin has been made.
Tools are changed by the calendar
Either an insert fails mid-cut and takes the part with it, or a third of usable tool life gets thrown in the bin to be safe. Both are paid for.
Traceability gets assembled by hand
When a customer raises a complaint, someone spends two days rebuilding which machine, operator, program revision and material lot made that part.
What we connect to in your plant
We connect at the controller, not through a middleware layer you have to buy and maintain. Where a machine has no port, we retrofit it.
Business systems
The signals we read, and what each one tells you
A tag on its own is noise. What matters is the question each one answers for the person standing at the machine.
| Signal | What it tells you |
|---|---|
| Spindle load and axis load curve | Tool wear, wrong feed and speed, and material hardness variation — visible in the shape of the cut, not after the part fails. |
| Program number and block | Which program revision actually made this part, which is the question every customer complaint eventually comes down to. |
| Cycle start and stop timestamps | True cycle time against the routing standard, and a clean split between setup time and run time. |
| Alarm and stop codes | Downtime cause taken from the controller instead of reconstructed from an operator's memory at end of shift. |
| Part and reject counters | Live yield on the job in progress, rather than a tally reconciled the next morning. |
| Operator login or badge | Who ran what, and where the skill gap between your best and average operator is actually costing money. |
| Weld current and voltage, press tonnage curve, mould cushion | The in-process quality signature — a bad part usually announces itself in the curve before the gauge sees it. |
| Gauge and CMM results | SPC tied back to the machine, the tool and the program that produced the deviation. |
| Compressed air and kWh per cell | Energy per part by part number, which is the line item most quotes ignore entirely. |
The numbers your plant is judged on, on one screen
Same definitions your team already uses. The difference is that they update through the shift rather than being compiled after it.
True OEE by cell
availability × performance × quality
Actual vs quoted cycle
seconds, by part number
Setup time per changeover
minutes, by machine
First pass yield
%, by job
Scrap cost by part number
currency per month
Spindle utilisation
% of attended hours
Tool life realised vs plan
% of expected life
MTBF and MTTR
hours, by asset
Energy per part
kWh per piece
On-time delivery
%, by customer
Downtime by cause
hours, ranked
Cost per part, actual
currency, against standard
AI trained on your process. Not a generic factory.
These are not generic anomaly detectors. Each one is trained on the way a machining and assembly plant loses money.
Quote-to-actual margin watch
Compares every completed job against its routing standard and ranks part numbers by margin leak, splitting the loss into setup, run time, scrap and downtime. The output is a list of the part numbers you should re-quote or re-engineer.
Setup-aware job sequencing
Groups jobs by fixture, tool family and material so the schedule minimises changeovers across the shift, then re-plans live when a machine goes down or a hot order lands.
Tool wear prediction from load signature
Learns the spindle-load fingerprint of a healthy cut for each tool and operation, warns before the edge fails, and stops good inserts being scrapped early on a calendar rule.
Scrap root-cause tracing
Correlates rejects against machine, operator, shift, program revision, tool and material lot, so the driver gets named by the data instead of by whoever argues hardest in the morning meeting.
Your first thirty days
We connect one cell first. If the numbers don't hold up, you have lost a month and no capital.
Connect one cell
Three to six machines, tapped at the existing controller. No PLC rewrites, no line stoppage, no new network for your IT team to approve.
Â
Â
2
Establish the real baseline
True OEE, the honest setup-versus-run split, and downtime reasons taken from the machine. Expect the first week to be uncomfortable.
3
Close the loop with ERP
Work orders flow in, actuals flow back. Costing starts using machine time instead of standard time.
4
Prove one number
Pick one part family and show the movement in scrap, changeover or margin in currency. That number decides whether you scale.
What plants aim for in the first quarter
8–15 pts
Three to six machines, tapped at the existing controller. No PLC rewrites, no line stoppage, no new network for your IT team to approve.
Â
Â
20–35%
reduction in changeover time from setup-aware sequencing
15–25%
less unplanned downtime once tool and stoppage patterns are learned
3–6%
scrap reduction on high-volume part numbers
Target ranges for scoping a pilot. Replace with your own validated customer results before this page goes live.
