Where OEE projects usually go wrong — and the fix for each.
OEE measurement fails in a small number of recognisable ways, and almost none of them are about the sensor. They are about what gets counted, where it gets counted, and whether anyone can act on the result. Each of these is worth checking before you trust a number enough to change a shift pattern on it.
| What goes wrong | How you notice | The fix |
|---|---|---|
| Counting off the wrong signal | The system’s count drifts away from the operator’s tally over a shift | Count as late in the line as you can — counting at the infeed records what you fed, not what you made |
| Micro-stops under the threshold | Availability looks healthy while the shift felt like a fight | Lower the stoppage threshold until the short stops appear. You cannot manage what the system rounds away |
| Changeover boundaries undefined | Two shifts running the same job report different OEE | Write the rule down — last good part of A to first good part of B — and apply it on every line |
| Rework counted as produced | Quality sits near 100% while the scrap bin fills | Take the quality count at the point of inspection, not the point of production |
| Sensor drift or fouling | A slow, one-directional trend with no process change behind it | Check the sensor before you investigate the machine; a slipping proximity mount looks exactly like a slowly failing line |
| Reason codes nobody can use | Most stops land on “other” | If the majority of stops are uncoded, the list is wrong — not the operators. Build it from what they actually say |
| No baseline before the fix | The improvement is argued rather than shown | Take two weeks of untouched data before you change anything. Without it, every gain is a matter of opinion |
| Dashboards nobody opens | The screen is on the wall and the meeting still runs off a spreadsheet | Publish one number per shift that someone is accountable for. Reports that inform nobody get switched off eventually |
Two of these decide whether the project survives its first month. Reason codes are the one operators feel: a list written in an office produces “other” for half the shift, and once that habit sets in the data never recovers. Build the list from the words people already use on the floor, keep it short enough to pick from with gloves on, and revise it after two weeks rather than defending it. The missing baseline is the one management feels: without a clean before, every improvement becomes a debate about whether the line was always like that, and the fastest way to lose a measurement programme is to win an argument nobody can verify.
