What it reads
The signals it works from. Typically connects to MES or historian, Quality system, ERP and Cost data.
Reads production data against spec targets and proposes the setting changes that hold quality while costing less.
gross margin
to the Scan's evidence — before any scope
to a system in production, not a pilot
modeled targets · not client results · timings are CRFT's standard engagement
Targets were set once, by someone who has usually moved on, with a safety margin nobody has revisited. The line runs to them because that is what the sheet says, and the cost of that margin is invisible because it has always been there.
Where the Scan startsWith production data, current spec targets and who sets them today.
Four lines, the same four on every system we build. Where they fall is what makes this one different from the last one.
The signals it works from. Typically connects to MES or historian, Quality system, ERP and Cost data.
The calls it makes on its own, unattended, every time it runs.
Into the tools the work already lives in.
Process engineering approves every change and quality holds a veto. This is the case where the human line is least negotiable — the system proposes and predicts, it never touches a setting.
The same four legs as every CRFT engagement — written for this system rather than in general.
We test whether your data can support the question at all. Sampling rate and sensor coverage decide this, and the honest answer is sometimes that the instrumentation comes first.
One line, one product family, one set of targets. Yield work generalizes badly across lines and should not be scoped as though it does.
Prediction is validated against held-back history before a single change is proposed. Nothing reaches the floor on the strength of a backtest alone.
Every trial, successful or not, is a labeled experiment. The model gets specific to your line, which is exactly what a generic optimizer cannot be.
Margin points against the pre-change baseline, with quality holds flat or better — both conditions, or it has not worked.
Five inputs, a published price, a report in 48 hours. Decide with numbers — then own what gets built.