Not sure which one you're looking at?
That's the conversation we have for free. Tell us what your line does today and what you need it to do — we'll tell you honestly what it takes, even when the honest answer is the smaller project.
Insights · Scale · September 14, 2026 · 9 min read
Why Your Capacity Model Says 95% OEE and Your Line Says 58%
The gap between the spreadsheet and the floor is where capital gets wasted. How to measure the real number, decompose it, and decide what's actually worth fixing.
A refrigerated foods manufacturer asked us to help them hit 280,000 pounds a day for the coming season. Their capacity model said the line could do it with one more filler. The model assumed 95% OEE.
When we measured the line station by station, true OEE was about 58%. The filler wasn't the constraint. The chill loop was — running in batch mode at 96 cups per minute while the 120 cpm filler sat idle waiting on it. And the "one more filler" they were about to buy was already sitting in their own building, uninstalled.
That gap — 95% assumed, 58% actual — is the most common thing we find in a plant, and it's the most expensive. Here's why it happens and how to close it.
What OEE actually is
Overall Equipment Effectiveness is three ratios multiplied together:
- Availability — of the time you planned to run, how much did you actually run? Downtime, changeovers, and waiting all subtract.
- Performance — while running, how close to the line's real rate were you? Slow cycles and minor stops subtract.
- Quality — of what you made, how much was sellable the first time? Scrap and rework subtract.
Multiply them. A line at 75% availability × 82% performance × 95% quality is at 58% OEE — which is roughly where that client sat, and roughly where most small and mid-size food plants sit when someone finally measures. World-class discrete manufacturing is 85%. Ninety-five percent is not a plan; it's a typo.
Why the model says 95%
Because nobody built it from measured data. The number comes from one of three places: the equipment nameplate (which assumes perfect conditions and infinite operators), a "best shift ever" that got remembered as normal, or a consultant's template with a default in the cell. All three describe a line that doesn't exist.
The model then compounds the error. If every stage is modeled at 95%, the line looks balanced and the fix looks like "add capacity at the slowest nameplate." But real stages don't lose efficiency evenly. One stage loses 40% to changeovers, another loses 25% to a jam that happens every eleven minutes, another runs fine. The real bottleneck is wherever the losses stack up — and that's invisible in a model that assumes they don't.
How to measure the real number in a week
You don't need software. You need a clipboard and honesty.
- Pick a normal week. Not the week after the big fix, not the week the new operator started.
- At each stage, log every stop longer than about a minute, with a reason. Changeover, jam, waiting upstream, waiting downstream, cleaning, break, no operator, no material.
- Count output at each stage per hour against the rate the machine can actually hold when it runs clean.
- Count scrap and rework by stage — where was it made, not where was it caught.
At the end of the week you'll have availability, performance, and quality per stage. Multiply. Rank. The lowest number is your constraint, and the reasons column tells you what it's made of.
Two warnings. First, "waiting upstream" and "waiting downstream" are not this stage's problem — they point at the neighbor. Assign them correctly or you'll fix the wrong machine. Second, the crew will run better while being watched. Budget for a 5–10 point Hawthorne effect and treat the measured week as optimistic.
What to do with the decomposition
This is where most OEE efforts stall — a dashboard gets built and nothing changes. The point of decomposing is that each component has a different fix and a different price:
| Loss type | Typical cause | Typical fix | Cost order |
|---|---|---|---|
| Availability — changeover | Tooling, fixturing, cleaning sequence | Changeover engineering, quick-release tooling, SMED | $10K–$60K |
| Availability — breakdowns | No PM program, worn components | Preventive maintenance built from failure history | $5K–$30K program |
| Availability — waiting | Upstream/downstream imbalance, batch vs. continuous | Buffering, conversion to continuous flow, resequencing | $20K–$200K |
| Performance — minor stops | Product handling, jams, sensor faults | Guide/transfer redesign, controls tuning | $5K–$50K |
| Performance — slow cycles | Running below rate to avoid faults | Fix the fault, then restore rate | Often free after the above |
| Quality | Process drift, wrong equipment for the product | Process control, inspection, sometimes redesign | Wide range |
Notice that most of the table costs less than a filler. In that client's case, moving the chill loop from batch to continuous flow was the lever — and the sensitivity math made the priorities obvious: every OEE point recovered was worth about 3,600 pounds a day; every added production hour was worth 13,500.
The number to put in the model
When you rebuild your capacity model, use the measured OEE for the current state and a defensible target for the future state — 70–75% is aggressive but achievable for a small food plant with a real PM program and engineered changeovers. Then ask the model what capital, if any, you need to hit your volume at that OEE. The answer is frequently "less than we thought," and occasionally "none — install the thing in the warehouse."
How we do this. Our capacity assessment is a fixed-price engagement — two to three weeks, station-by-station measurement, true OEE decomposed, the constraint named with numbers, and a ranked list of levers with cost and payback. It ends with a plan you can fund, not a dashboard. The last one saved a client a filler they already owned.