Why your OEE number is probably wrong, and why it matters more when you’re capacity-constrained

Your plant reports 72% OEE. Two things are almost certainly true about that number.

The first is that it is not 72. On most lines we have measured, the reported OEE is 10 to 15 points above what machine data shows for the same week, and almost every error that produces the gap pushes in the same direction: upward.

The second is that if your plant is capacity-constrained, the gap between 72 and the real number is the whole capex decision. At 72% a plant believes it is close to the ceiling and that the only way to make more is to buy a line. At 58% the same plant has 14 points of headroom it has not seen, worth more than the line it is about to buy.

OEE is the most quoted and least audited number in manufacturing. This article explains how it drifts, how to audit it in three questions, and why the honest number is the beginning of the capacity case rather than an embarrassment.

Seven ways the number drifts

None of these is fraud. Each is a reasonable-looking choice made years ago by someone who needed a number by Friday. The problem is that nearly all of them flatter the result, so nobody has a reason to revisit them.

1. The time basis is unstated, or quietly narrowed. OEE should be calculated on planned production time, also called scheduled or loading time: the hours the line was meant to run. Some exclusions from it are legitimate and standard: shifts not scheduled, unmanned breaks, no-demand periods and maintenance planned well ahead. The drift is in excluding losses that happen inside planned time: changeovers booked as “planned”, start-ups and cleans that overrun, short stops. Every such exclusion raises availability. A line that runs at 58% of planned production time can report 72% on “net operating time” with no change on the floor.

2. The target speed comes from habit, not the nameplate. The ideal cycle time in the system is the speed the line ran “reliably” in 2022, after someone turned it down to stop a jam. Performance reads 95% against a target that is 10 to 20% below rated. The speed loss is still there; it has been promoted into the baseline.

3. Downtime is self-reported. Stops entered at the end of a shift from memory undercount the real figure, often by 20 to 50% when checked against a sensor, and the undercount is largest on the categories operators would rather not write down. A line that logs 15 hours of stops can easily have had 20.

4. Micro-stops vanish. Anything shorter than the logging threshold, usually two to five minutes, is either not recorded at all or gets absorbed into the performance factor, where it is indistinguishable from speed loss. On packing lines micro-stops are often one of the largest loss categories, and in most OEE reports they do not exist.

5. Changeovers are excluded as “planned”. The largest controllable loss in FMCG, moved out of the denominator by definition. A changeover that takes 41 minutes against a 25-minute standard costs the plant 16 minutes of capacity, and in a report that excludes changeovers, it costs nothing.

6. Quality is counted in the wrong place. Good count taken from the QC sheet or the ERP confirmation rather than from the machine, so rejects that never reached QC are counted as good, and rework that went through the line twice is counted once. This one can push the number either way, which is the only reason anyone notices it.

7. Everything is averaged. A plant OEE of 72 hides a 45% line and an 88% line. A line OEE of 65 hides a Saturday at 50 and a Tuesday at 74. A monthly figure hides the SKU that ran at 40 for three days. The average is the number in the board pack, and the losses are in the variance it removed.

The three-question audit

You do not need a project to find out whether an OEE number is honest. Ask the person presenting it three questions.

1. What is the time basis, and what is excluded from it? If the answer is “net time”, a list of exclusions that includes changeovers, or a pause, the number is soft.
2. Where does the ideal cycle time come from? If the answer is “the standard” or “what the line normally runs at”, performance is overstated.
3. Who recorded the stops? If the answer is “the operators”, availability is overstated.

Three honest answers (planned production time, the rated speed per SKU, a sensor or the PLC) and the number can be trusted. Anything else and it is soft, almost always upward, and in our experience the gap is usually 10 to 15 points.

Worked example: 72 becomes 58

Same packing line as in the earlier articles, same week, same physical output: 835,200 good units. Relief crews cover breaks on this line, so breaks are in neither column. The only thing that changes between the two columns is how the number is built.

Three things are worth noticing. Runtime fell from 93 to 88.8 hours, because the shift log missed 4.2 hours of stops. The quality factor went up when recomputed, because the QC log was double-counting rework; the reported number was wrong in both directions, which is normal. And the output did not change. Nobody made fewer units. The plant simply discovered that it had 42 points of loss where it had been reporting 28, and that 14 of those points had been hidden by definitions.

A quick check confirms the honest figure: good output divided by what the line could make at rated speed in all 120 planned hours is 835,200 ÷ 1,440,000 = 58%. The gap here is 14 points. We often see 10 to 15 on the first machine-verified week; occasionally more where changeovers are long and excluded.

Why it matters more when you’re capacity-constrained

At 85% OEE a 10-point reporting error is an argument between the CI team and the data team. At 58% reported as 72%, it is the capex decision.

A plant that believes it runs at 72% believes it has 28 points of theoretical headroom, most of which it assumes is structural, and that it is close enough to the ceiling that the only serious way to add output is a new line. The business case goes to the board with demonstrated capacity as the baseline and a seven-figure quote as the answer.

The same plant at an honest 58% has 42 points of loss, and when those are sorted the way the hidden capacity method sorts them, 12 of them are recoverable without capital. That capacity is the output of most of a new line. It does not exist in the 72% version of the world, which is why nobody looked for it.

There is a second trap, and it is the benchmark. The 85% “world class” figure comes from Seiichi Nakajima’s work on Total Productive Maintenance (TPM) in the 1980s: 90% availability, 95% performance and 99% quality. It is a fair target for a stable, dedicated process. Measured honestly on planned production time, few high-mix FMCG packing lines reach it, and in our experience most run between 45 and 65%. A plant that compares a soft 72 to an aspirational 85 concludes it is doing reasonably well and is capacity-constrained. The same plant comparing an honest 58 to a realistic 70 concludes it has a recovery program to run before it has a capex case to make. Those are different years.

Don’t fix OEE by redefining it

Once a plant has an OEE target, the number starts to attract games. Changeovers move to “planned” and availability rises. The standard speed is lowered after a bad quarter and performance recovers. The Saturday shift is excluded because “it isn’t representative”. A bad SKU is reclassified. Each change makes the number better and the plant worse, and each is defensible on its own.

The rule is simple. Pick a basis, state it on every report, and never change it to hit a target. If the basis has to change for a good reason (a new shift pattern, a new line), restate history on the new basis so the trend survives.

For capacity decisions, the basis is planned production time, with changeovers and every stop inside it counted as loss. That is the basis the new-line business case is implicitly using when it says “we need more hours”, so it is the only basis on which recovered capacity and a new line can be compared. When the question is “could we run more hours than we schedule”, use TEEP on calendar time alongside it, because that is the only number that shows how much of the week the asset is idle by choice.

How to get an honest number in four weeks

1. Take stops and counts from the machine. A sensor on the line or a signal from the PLC, timed automatically. Operators add the reason code; they do not estimate the duration.
2. Set the ideal speed per SKU from the rated speed, or from the best demonstrated sustained run if the OEM figure is unrealistic for the current format. Write down which, and why.
3. Put all planned production time in the denominator, changeovers included, and list every exclusion explicitly. If a reader cannot see what was excluded, the number is not auditable.
4. Report by line, shift and SKU, never only the average. The average is for the board; the variance is where the capacity is.
5. Keep the reason-code list short and coded, with “other” below 10% of lost time as a first step and close to zero as the goal. Above 10%, the Pareto is fiction.

The first honest number will be lower than the reported one. It always is, and it causes a difficult week. The way to present it is not as a correction but as a finding: the plant has more recoverable capacity than it believed, and here is the Pareto that says where. Measured this way for four weeks, the line in the example above went from “close to the ceiling at 72” to “12 points recoverable from 58”, and that reframing is worth more than any OEE improvement program that starts from a soft baseline.

Frequently asked questions

Is 85% OEE realistic? For a dedicated line with stable conditions and few changeovers, it is a demanding but real target. For a multi-machine FMCG packing line with frequent changeovers, measured honestly on planned production time, rarely. In our experience most such lines run between 45 and 65%; 70 to 75% sustained is a strong result.

What is a good OEE for FMCG? On planned production time with changeovers included, 60 to 70% is good for a high-mix packing line and 70 to 80% for a low-mix one. Any benchmark comparison is meaningless unless both numbers use the same time basis and the same speed reference.

What is the difference between OEE and TEEP? OEE measures performance against planned production time. TEEP (Total Effective Equipment Performance) measures it against calendar time, 24 hours a day, 7 days a week. TEEP is lower and answers a different question: how much of the asset’s total potential is being used, including the hours you choose not to run.

Why does my OEE differ from my equipment vendor’s figure? Vendors quote OEE for a single machine at rated speed, often excluding changeovers, on a short acceptance run. Your figure covers the whole line, your product mix, your crews and a full week. Both can be correct; they are not comparable.

Should changeovers be included in OEE? Yes, for any capacity decision. A changeover consumes production time the plant has planned and paid for. Excluding it hides the largest controllable loss in most FMCG plants. The exception is a short, fixed changeover that genuinely cannot be reduced, and even then it should be shown as a separate planned loss rather than removed from the denominator.

Can OEE be too high? If a line reports above 85% sustained, check the time basis and the ideal speed before celebrating. A number that high on an honest basis is unusual; a number that high on a soft basis is common.

Audit the number before you build on it

If a capex request, a capacity plan or an overtime roster rests on an OEE figure, run the three-question audit first. If any answer is soft, get a machine-verified number for one line before the decision is made. The honest number is almost always lower, and almost always the beginning of a better plan.

The fastest route is a fixed-scope pilot: one representative line, six weeks, an industrial sensor and hub installed with your team, and an executive readout with the OEE on a stated basis, the loss Pareto covering at least 80% of lost time, and the recoverable capacity in dollars. The fee is fixed and credited in full against a first-year subscription if you roll out. The pilot runs on Fabrico’s manufacturing performance platform (MES, OEE, CMMS & AI), which connects machine data, OEE and loss analysis, production scheduling, SKU-level output value and maintenance in one system, so the honest number stays honest after the pilot ends.

Request a demo or check your own basis with the OEE calculator.