Somebody puts 71 percent on the board. The room accepts it. Nobody in the room could reproduce it from first principles, and everybody has enough experience to know it is roughly right, so the meeting moves on to the shift that came in at 64.

Here is the problem with roughly right. TeepTrak’s 2026 benchmark, built from direct sensor measurement across more than 450 plants in over 30 countries, found that manually logged OEE overstates the true figure by 8 to 15 points. The industry median measured properly is around 60 percent. The top quartile is around 75.

Eight to fifteen points is not a rounding error. It is the difference between a plant that thinks it is top-quartile and a plant that is median.

But the size of the error is not the interesting part. What is interesting is which losses go missing, and what happens to them.

The losses that hand-logging cannot see

Manual downtime logging captures events that are long enough and disruptive enough for a person to stop and write them down. A press goes down for forty minutes: that gets logged. A changeover runs long: that gets logged.

What does not get logged is the short stuff. Micro-stops under five minutes. Speed loss where the line runs but runs slow. A jam cleared in ninety seconds. A feed corrected by hand. A part reseated. None of it interrupts anyone’s day enough to become a data point, and across a shift it can quietly add up to eleven minutes in ninety-second pieces.

The same benchmark identifies micro-stops under five minutes as the largest hidden loss category across sectors, running 18 to 38 percent of total losses depending on the industry.

So the first answer to why is my OEE wrong is mechanical: your logging method cannot see the losses that occur faster than a person will reach for a pen.

The second thing your number is measuring

Now follow that one step further, because this is the part almost nobody says out loud.

Those short losses do not simply vanish into the arithmetic. Somebody absorbed them. A person stood there and cleared the jam, corrected the feed, reseated the part, nudged a setting, and kept the line running. That is why the loss never reached a threshold worth writing down.

Which means your OEE number is the output of two systems, not one.

The first is the machine and the process as designed. The second is a continuous, unwritten, unpaid layer of human correction sitting on top of it, holding the number up.

Your improvement program can see the first one. It has no instrument at all for the second.

This is why plants that install real-time measurement often get an unpleasant surprise, and then a more interesting one. The unpleasant surprise is that the true number is lower than the board said. The interesting one arrives a few weeks later, when the Pareto of micro-stops points at a step that is not in any procedure, being performed by somebody who never considered it worth mentioning because to them it is simply how the job is done.

Why this shows up as shift-to-shift spread

The same benchmark data reports that many US automotive operations run a 5 to 15 OEE point variation between morning, afternoon and night shifts on the same equipment.

Equipment does not change between four in the afternoon and midnight. Product mix mostly does not. What changes is who is absorbing the unmeasured losses, and how well.

Read that way, a shift spread stops being a discipline problem and becomes an information problem. The gap between your best shift and your worst is, in large part, a difference in undocumented compensation. One shift knows something, or does something, that the others do not, and no system in the plant is built to notice what it is.

Notably, the same benchmark names knowledge transfer, alongside standardized work and supervisor cadence, as one of the levers that closes cross-shift gaps. Not enforcement. Transfer.

Run the unaccounted-minutes reconciliation

This is a one-shift exercise, one line, and you need nothing you do not already have.

Step one. Take one line for one shift. Record the actual clock: shift start to shift end, minus scheduled breaks and planned changeover. That is your available time.

Step two. Take the good units produced and multiply by the ideal cycle time. That is your theoretical run time.

Step three. Subtract. The remainder is total loss.

Step four. Now open the shift’s downtime log and add up what was actually recorded.

Step five. Subtract the logged loss from the total loss. What remains is your unaccounted minutes: real time, really lost, with no entry anywhere.

Step six, and this is the whole point. Go to the person who ran that line and ask what they were doing during those minutes. Not accusingly. Ask what small things they handled that they would not normally mention.

Most plants doing this for the first time find between fifteen and forty unaccounted minutes on a single shift, and every one of them has a story attached. Those stories are the highest-value process documentation available to you, and they are free, and they expire at the end of the conversation if nobody writes them down.

What to do with what comes back

Sort the answers into two piles.

Pile one: the correction should be in the standard. The person has worked out a better sequence, a check that catches something early, an adjustment that prevents a stoppage. Capture it, validate it, write it in. You have just recovered an improvement your plant already paid for and never collected.

Pile two: the correction is masking a defect. The person is compensating for something that should not be happening at all, a fixture that drifts, a feed that binds, a spec that does not match the material. Capture it and hand it to engineering. The compensation was buying you time; it was not solving anything.

Both piles matter. What you must not do is treat either one as a compliance issue, because the moment the conversation becomes about following the procedure, the information stops arriving. People will happily tell you what they do to keep the line running. They will stop the second it starts to sound like an admission.

The direction the instruments point

Every measurement system in a plant is built to count failures. Downtime, scrap, defects, findings. That architecture is sound, and it has a structural blind spot: a failure that somebody prevented informally generates no record at all. There is no data class for things that would have gone wrong if this person had not been standing here.

So the honest answer to why is my OEE wrong has two parts. Your logging method cannot see short losses, which is fixable with sensors. And your organization has no mechanism for capturing the human corrections that absorbed them, which is not fixable with sensors, because the correction lives in a person and has to be asked for.

That second half is the work we do at SenseiLab: capturing how the job actually runs, from the people running it, and turning it into procedures that stay current. The reconciliation above is the one-shift version you can run yourself this week. It usually produces enough of an argument to be worth the hour.

Frequently asked questions

Why is my OEE wrong or higher than reality? Manually logged OEE typically overstates the true figure by 8 to 15 points, because hand-tracking cannot capture micro-stops under five minutes or speed losses where the line runs below rate. Those short losses are the largest hidden loss category in most sectors, running 18 to 38 percent of total losses.

Why do my shifts produce different OEE numbers on the same equipment? Many operations run a 5 to 15 point spread between shifts on identical equipment and identical procedures. Since the equipment does not change, most of the spread reflects differences in undocumented practice: what each shift knows, and what each shift absorbs by hand without recording it.

How do I find the losses my downtime log is missing? Calculate total loss for one shift from available time minus good units times ideal cycle time, then subtract the logged downtime. The remainder is unaccounted minutes. Ask the operator what they were handling during that time. The answers are your undocumented process steps.

Is buying real-time OEE monitoring enough to fix this? Sensors will correct the measurement and show you where the short losses cluster. They will not tell you what a person was doing to prevent those losses from becoming bigger ones. That information only comes from asking the operator, and it needs to end up in the procedure to be worth anything.

About the author: Diego Echenique is CEO and co-founder of SenseiLab, which does knowledge capture and Living SOPs for manufacturers in regulated, knowledge-intensive industries. He has spent over twenty years in manufacturing operations across the Americas, Europe, the Middle East and Asia, as a Lean and Six Sigma practitioner and in supplier development and plant leadership roles.