Plants that install sensors and dashboards frequently see no change in output, because measurement is not improvement. The gap is between having data and having a decision someone acts on during the shift it matters.
Downtime reasons beat downtime totals
Knowing a line lost 90 minutes is not actionable. Knowing 40 of them were one changeover, 30 a recurring jam at one station and 20 waiting for material gives three specific projects. Structured reason capture at the moment of the stop is worth more than any amount of retrospective analysis.
Total downtime is a number. Attributed downtime is a work plan.
Get OEE honest before optimising it
OEE calculated from optimistic assumptions and manual entries flatters the plant and misleads decisions. Availability from machine signals, performance against a genuine cycle time, quality from actual rejects — measured rather than estimated — often produces a lower and far more useful number.
Measure at the constraint
Improving a machine that is not the bottleneck changes nothing about output. Instrumenting the constraint first, and understanding what starves or blocks it, is the highest-return sequencing decision in most plants — and it is frequently not where the newest equipment is.
Predictive maintenance needs failure history
Prediction requires examples of failure to learn from. Where that history does not exist, condition-based rules — vibration above a threshold, temperature trending up — deliver value immediately while the history accumulates. Promising prediction on a plant with no labelled failures is how these projects lose credibility.
Put the data where the work happens
A dashboard in the office informs a weekly meeting. A screen at the line, showing current rate against target and the reason for the last stop, changes behaviour during the shift. The closer the feedback is to the operator, the more the measurement is worth.
Read-only, roll out by line
Anything touching plant control carries production risk. Keep acquisition read-only wherever possible, roll out line by line, and schedule anything invasive into planned maintenance windows with the plant engineers. Trust is lost permanently the first time a data project stops a line.
