A reported OEE of 78% and an observed 55%

Almost every plant reports an OEE. Very few can explain how losses were classified, who decided a stop counted as planned, or why short stops never appear in the data. Without an MES this is fixable. With an MES it is often worse, because the number looks authoritative.

Written for operations leaders running or evaluating manufacturing in Brazil.

Spend two hours on a line with a stopwatch and compare the result against the reported number. In mid-size plants the gap is routinely twenty points or more. Nobody falsified anything. The measurement system simply was not designed to be true.

This matters well beyond reporting. Capital decisions are built on these numbers. Plants buy a fourth machine while three run at half their real capability, because the reported figure said capacity was exhausted.

Where the number goes wrong

Micro-stops disappear

Stops under a few minutes rarely get recorded — an operator clears a jam and carries on. Individually trivial, collectively they are frequently the largest single loss in the plant.

Changeover is classified as planned

Calling setup “planned downtime” and excluding it from the calculation makes the number look better and makes the largest recoverable block of time invisible. In a high-mix plant this single classification choice can move reported OEE by twenty points.

Rated speed is fiction

The design speed each asset is measured against was often set years ago, sometimes from a sales brochure, sometimes optimistically. If the machine has never achieved it, performance is calculated against a number that does not exist.

Quality loss lives in another system

Scrap and rework are usually tracked in the quality system, separately from downtime. Nobody ever sees the combined picture of what a shift actually cost.

Building manual measurement that holds

Manual collection is not a compromise. Done with discipline it is accurate, and it forces definitional clarity that automated systems let you skip.

1. Write the loss taxonomy first. Every stop category defined unambiguously, including the boundary cases. If two people can classify the same stop differently, the taxonomy is not finished.
2. Validate rated speeds. Establish, per asset, the speed it demonstrably sustains in good conditions. Use that.
3. Record micro-stops. A simple tally sheet at the line. Operators will do this if the reason is explained and if the data visibly gets used.
4. Bring quality loss into the same calculation. Scrap and rework as OEE losses, not as a separate report.
5. Decide who records and when. Ambiguous ownership produces gaps that get filled with estimates.
6. Have someone independently observe a shift in the first weeks and compare against what was recorded. This calibrates the system and it is the step most often skipped.

On targets

There is no universal OEE target, and anyone quoting one without seeing your process is selling something. Realistic performance depends on process type, product mix, asset age and changeover frequency. The useful objective is an honest baseline and a Pareto of where loss actually sits — not a number to compare against another plant running a different product mix.

The organisational part

Honest measurement usually reveals that reported performance was overstated, and somebody has been presenting it upward. How this is handled determines whether the new system survives. Framed as discovering the real opportunity, it works. Framed as catching someone, the measurement quietly reverts within a quarter and you have made things worse than before.