Your OEE report does not tell you what it counted. Overall equipment effectiveness is availability times performance times quality, and availability is the term plants define most differently while believing they measure the same thing. Whether a changeover, a material wait, a tool change or a four-minute jam sits inside planned production time is a local decision, usually undocumented, usually made years ago by somebody who has since left. Until those boundaries are written down and shared, two sites’ numbers cannot be compared, and any decision that ranks one site against the other is ranking their record-keeping, not their performance. I have sat in that meeting more times than I want to count. Three plants on a Tuesday call, availability within four points of each other, and the maintenance planner at the worst-performing site staring at the screen because he knew his line had run better that month than the site above him. He was right. His plant counted every changeover as downtime. The site above him had decided years earlier that a scheduled changeover was planned work and came out of the denominator. Nobody was cheating. Nobody could explain the gap either, so the call moved on to the next slide, and the ranking stood. Why three plants report three different overall equipment effectiveness numbers Take the same eight hours and three rule sets. Plant A excludes scheduled changeovers, planned maintenance and the fifteen minutes of startup from planned production time. Availability comes out in the low nineties. Plant B counts everything inside the shift except breaks. Changeover is downtime. Startup is downtime. Availability lands in the high seventies. Plant C sits between the two, and adds one more rule: run time starts at the first good part, not the first motion. Every ramp-up minute lands as downtime with no code attached to it. Same eight hours, same product, same crew skill. The spread between those three availability figures is not a difference in performance. It is a difference in what each plant counts. And because the OEE calculation multiplies, a fifteen-point difference in availability moves the headline number more than any real improvement any of those plants made that quarter. The six choices behind every availability number Availability is run time divided by planned production time, so the number depends entirely on what counts as planned and what counts as a stop. It looks like one calculation. It is actually six separate choices. Every plant has made all six. Most have never written any of them down. What is inside planned production time. Unstaffed hours, weekend shifts with no orders, holidays. In or out. How scheduled changeovers are treated. Downtime inside the denominator, or excluded as planned work. When run time starts. First motion, first part, or first good part. The minimum recordable stop. If the system ignores anything under two minutes, every micro-stop on the line goes unrecorded, and the lost time shows up later as a performance problem instead. Which causes belong to the line. No orders, no material, no operator, supplier hold. Each of those can sit inside availability or outside it. Who enters the record, and how long after the stop. A code typed at the end of the shift from memory is less accurate than one entered at the machine while the line is still down. Any answer to any of these can be defended. The damage comes from three sites answering them differently and then putting the results in the same table. How to standardise availability definitions across plants in about two weeks This is cheap. It needs no software purchase, no outside help, and no change to how anybody runs a line. Week one, measure your own spread. Pick one week and one shift pattern. Have each site export stop-level records, not the summary: start time, end time, duration, reason code, who entered it and when. Then take one site’s raw week and recalculate it by hand under the other sites’ rules. Two hours in a spreadsheet gets you a number nobody in the business has ever seen: how far apart your sites are for identical performance. Publish it. That number ends the argument about whether this matters. Week two, write one page. Six lines, one for each choice above, in plain language a new supervisor can follow. Put names on it, a date, and a version number. Then pick the rule that records the most lost time, because the point of the exercise is to see the hours you are losing, not to protect a score. Then re-baseline and say so out loud. Every site whose definition was loose will see its availability drop. That drop is arithmetic, not a decline in performance, and telling people that before the first reconciled report goes out saves a month of defensiveness. One more habit worth building: the definition changes only with a version bump and a written note. When someone asks in eight months why the second quarter looks strange, the answer takes a minute instead of a week. What a standardised availability number is for Ranking plants is the least valuable thing a standardised record gives you. The valuable thing is that you can tell whether a fix worked at a second plant. When one site cuts twenty minutes a shift off changeover and every other site is keeping the record the same way, you can check whether the same change did the same thing somewhere else. Under three definitions you cannot. You get a slide showing that everyone improved, and a year later nobody can say which change caused what. That is the difference between a report and a decision. A comparable availability number lets somebody move a person, a spare, or a schedule and then find out if it worked. An incomparable one only gives the meeting something to look at. The maintenance planner on that Tuesday call already knew his line was running well. What he did not have was a way to prove it in the numbers the meeting used. That happens in a lot of plants, and the cost is not the argument. It is that the best operator in the group stops bringing evidence, and the money keeps going to whichever plant has the loosest definition. This week: Take one shift from each plant, run it through the other plants' availability rules by hand, and see how far apart the same eight hours come out.