Yes, OEE works in a job shop if you measure it per machine: add up good parts × ideal cycle time across every job, then divide by scheduled time. Setups count as availability loss and short runs cap the score, so judge a shop against its own ceiling and trend, not a benchmark from long-run lines.
Why job shop people distrust OEE
The objection is fair. A shop running four jobs a shift on one machining centre looks nothing like a filler that runs one bottle for a week.
The formula still holds. What changes is the data, the size of the setup loss and how you read the result.
Measure the machine, not the job
In a job shop the machine stays and the jobs move through. So OEE belongs to the machine or work centre, summed over all the jobs it ran in a shift.
Start with the machine that has the longest queue in front of it. Bottleneck analysis helps you pick, because a point gained on a machine with spare time ships nothing extra.
Worked example: four jobs in one shift
Example: a machining centre is scheduled for 480 minutes and runs four jobs. Each has a proven ideal time per part.
| Job | Good parts | Ideal min per part | Good min | Setup min |
|---|---|---|---|---|
| 1 | 40 | 3 | 120 | 25 |
| 2 | 12 | 5 | 60 | 40 |
| 3 | 25 | 2 | 50 | 20 |
| 4 | 6 | 10 | 60 | 35 |
| Total | 290 | 120 |
OEE = 290 ÷ 480 = 60.4%. The other 190 minutes are the part to read.
By factor: availability is 330 ÷ 480 = 68.8%, performance is 305 ÷ 330 = 92.4% and quality is 290 ÷ 305 = 95.1%. Multiplied, that is 60.4%.
The scrap line is two parts that failed after setup ended: one from job 2 at 5 minutes and one from job 4 at 10. In short runs a single scrapped part hits quality hard, so read scrap job by job.
The setup ceiling
This idea changes how you read the score. Even if a machine runs perfectly between setups, the setup share caps OEE.
Nakajima's 85% figure, which the bar for batch and discrete work in Overall Equipment Effectiveness (R. C. Hansen, 2001) builds on, cannot be met by a shop whose average run is 60 minutes with a 30-minute setup. Read is 85% OEE realistic before anyone sets that target.
So work out your own ceiling from last month's setups and run lengths. A machine at 55% with a ceiling of 67% has a very different problem from one at 55% with a ceiling of 90%.
Parts with no proven cycle time
A high-mix shop keeps meeting parts it has never run. That is the real data problem, and it has a practical answer.
- Repeat parts: use the fastest cycle proven in production, as OEE for CNC machines describes.
- New parts in a known family: use the family rate, grouped by material and operation.
- True one-offs: use the programme or estimated cycle time and flag it as provisional. Quoted times usually include allowances, so they make performance look better than it was.
Our advice is to print one extra number: the share of good minutes from parts with a proven rate. "OEE 60%, with 90% of minutes on proven rates" tells a reader how far to trust it.
Once a one-off comes back as a repeat, replace the estimate with the proven time and keep that rate fixed. Ideal cycle time gives the full rules.
When the machine has no job
Job shops often have gaps, and each gap needs the same test. If no job was planned for that machine, the time is excluded and stays out of OEE.
If a job was planned but its material, drawing or programme was not ready, the wait is a stop. OEE with no orders has the full rule.
One operator on two machines is a common case too. Time a machine waits for that person counts as a stop, and its own code shows whether the staffing is the real limit.
What to record on the floor
You need few fields, and they all fit on a job traveller or a simple scan screen.
Job number and time. The clock on setup begins with the last good part of the job before.
This ends the setup. First-off trials and adjustments stay inside it.
Waiting for tools, material, programme or inspection, each with a reason.
Good count and scrap count, for parts made after setup ended.
Measure setup from last good part to first good part, as in how to measure changeover time. Waiting for a programme or a gauge inside that window is a real and common job shop loss, so give it its own code.
Practice varies on parts scrapped before the first good one. We keep them inside setup and out of the scrap count, so the same minutes are not charged twice.
The trap: chasing the score with bigger batches
Longer runs raise OEE, as the ceiling chart shows. That does not make them good for the business.
Running 200 parts when the customer ordered 50 lifts the score, ties up material and uses machine time for stock nobody asked for. In a job shop, pair OEE with on-time delivery and plan attainment, so the number cannot be gamed that way.
The honest lever is the setup itself. SMED raises the ceiling without changing a single batch size, and changeover frequency shows the other side of that trade.
How to read the result
- Trend, per machine, over weeks. A single shift depends too much on which jobs landed there.
- Loss split before the score. Setups, waiting, speed and scrap each have a different owner.
- Same rules for every machine, or the comparison means nothing. Check them against common OEE calculation mistakes.
- Job mix alongside. OEE by product shows which part families cost the most time.
Short runs mean many small events a shift, too many to write down by hand. Fabrico takes stops, speed and counts live from PLCs or external IoT sensors on older machines, and operators tag the cause at the line. See how Fabrico connects machines.
What people ask next
Is OEE useful for low-volume, high-mix manufacturing?
Yes, as long as you add ideal time across jobs and do not compare the score with long-run lines. The loss split tells you more than the percentage. OEE for batch processes covers a close relative of this case.
Should setup time count in job shop OEE?
Yes. Setup is a planned stop within normal running, so it lowers availability. Taking it out would hide the largest lever in many shops, as how to count changeovers explains.
What is a good OEE for a job shop?
We know of no neutral benchmark for job shops, so we will not quote one. Work out your setup ceiling, then set a staged target with how to set OEE targets.
How do I set an ideal time for a part I have never made?
Use the family rate or an estimated cycle time, flag it as provisional, and replace it after the first proven run. Keep estimates and proven rates apart in your report.
Which machines should I track first?
One machine, the one with the longest queue. Add more only when the first one produces decisions, as in OEE for one machine.
Fabrico is a light MES for OEE. It takes stops, speed and counts from PLCs or IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and operators tag the cause at the line.
Related guides: OEE for Printing Presses: Makeready, Waste, Speed · Manufacturing Lead Time: Formula and How to Cut It · OEE Calculation for a Machine Shop, by Example