Availability in OEE is run time divided by planned production time. It is the share of the time you planned to run that the machine was actually producing, so it captures every unplanned stop and changeover. The number is only as honest as its two denominators: what you count as run time and what you count as planned time.
Most gamed OEE numbers are gamed here, at the availability factor, and almost never through arithmetic. The division is trivial. The judgment calls are in the definitions: which minutes belong in planned production time, which stops count as downtime, and whether the same rules hold from one shift to the next. This guide fixes those definitions, then walks a repeatable way to measure availability that a supervisor cannot quietly inflate. If you want the arithmetic done for you once the inputs are clean, the free OEE calculator runs the same method.
What is the availability factor in OEE?
The availability factor is run time ÷ planned production time, expressed as a percentage between 0 and 100. It is the first of the three OEE factors, and it isolates time loss the stops that ate into time you intended to produce. It says nothing about speed or quality; a line can be 100% available and still make junk slowly. Availability only answers one question: of the time we planned to run, how much did the machine actually run?
Two numbers decide it. Run time is planned production time minus every unplanned stop and every changeover. Planned production time is the time you scheduled the line to make product, after you subtract things you deliberately chose not to run for, breaks, planned maintenance, meetings. Get those two straight and availability follows. Get either one loose and the factor drifts, usually upward. For the full three-factor picture around it, see the OEE calculation walkthrough.
How do you calculate availability step by step?
Collect two time totals for the period, subtract, then divide. The discipline is in writing the rules down once and never bending them shift to shift. Here is the procedure:
- Fix the period. Pick a shift, a day, or a run, a fixed window with a clear start and end. Availability is only comparable across periods that use the same rules.
- Start from scheduled time and subtract planned non-production. Take the time the line was staffed and scheduled, then remove breaks, planned maintenance windows, scheduled meetings, and any time you chose not to produce. What is left is planned production time. Write this exclusion list down and post it.
- Log every unplanned stop and changeover. Breakdowns, jams, material-outs, quality holds, and setups all count. A stop is a stop whether it lasted 40 minutes or 90 seconds. Total the minutes.
- Subtract to get run time. Run time = planned production time − total stop and changeover minutes.
- Divide. Availability = run time ÷ planned production time. Multiply by 100 for the percentage.
- Reconcile against the clock. Run time plus all stop minutes plus all planned exclusions should equal the raw shift length. If it doesn't, minutes are missing, usually short stops nobody logged.
What belongs in planned production time?
Planned production time is scheduled time minus the events you decided in advance not to produce during. That is the whole rule, and the fights are all about where "in advance" ends. The safe test: if you could have shipped product during those minutes and chose not to, it is a planned exclusion; if you meant to be producing and couldn't, it is downtime.
Standard exclusions are contractual breaks, planned preventive maintenance, scheduled cleaning or sanitation, and staffing gaps you scheduled on purpose. The trap is treating recurring problems as planned exclusions because they happen every day. A daily 20-minute "warm-up" that is really the line struggling to stabilize is not a break, it is a slow, recurring stop, and burying it in the exclusion list is the most common way availability gets quietly inflated. When people shrink planned production time to make availability look better, output falls while the percentage rises. If availability climbs and units shipped doesn't, the denominator was moved.
| Minutes in the shift | Classification | Affects availability? |
|---|---|---|
| Contractual lunch and breaks | Planned exclusion | No, out of the denominator |
| Scheduled PM window | Planned exclusion | No, out of the denominator |
| Unplanned breakdown | Downtime | Yes, lowers run time |
| Changeover / setup | Downtime | Yes, lowers run time |
| Material-out / starved line | Downtime | Yes, lowers run time |
| Daily "warm-up" that's really instability | Downtime (often mislabeled) | Should, often hidden |
What counts as downtime against availability?
Any minute you planned to produce and didn't counts as downtime. That includes the stops people forget: micro-stops under a couple of minutes, waiting on materials, and changeovers. Changeovers are the one operators most often argue about, so settle it plainly, a changeover is time the line was scheduled to make product and made none, so it counts against availability. Excluding setups hides one of the six big losses and usually the most improvable one, because setup-reduction methods can cut it hard without capital spend.
Starved and blocked time is the subtle case. If the machine is idle because the line upstream is down or the buffer downstream is full, it is still not producing during planned time, so it still counts against that machine's availability. That is why measuring availability on every machine of a line produces a wall of low numbers, upstream and downstream assets look "unavailable" simply because the constraint governs the pace. Measure availability rigorously at the constraint, and keep simple machine downtime tracking everywhere else. Whatever you decide, decide it once: the definition, not the arithmetic, is what makes availability comparable over time.
How do you measure run time without gaming it?
Measure it at the machine, at the highest frequency you can afford, and record stop reasons at the moment of the stop. End-of-shift recollection is where run time goes wrong. A supervisor writing the log at 6 a.m. remembers the one long breakdown and forgets the dozen 90-second jams, so those minutes silently migrate out of downtime and inflate availability. The fix is not a better memory; it is capturing stops as events.
Three practices keep run time honest. First, log stops as they happen, the operator flags the stop at the station, or the machine signals it automatically and the operator adds a reason code. Second, count short stops. A two-minute threshold that drops everything shorter is a decision to make a real loss invisible; the smaller stops belong in downtime, not written off. Third, reconcile to the clock every period so the minutes add up. This is the practical argument for wiring availability to the source rather than a clipboard: a live number the crew can act on this hour beats an autopsy written next month. It is also how Harmony computes availability, from machine signals and reason codes at the line, not shift-end estimates (see the platform).
Why do two plants get different availability from the same line?
Because they defined the denominators differently, not because one runs better. Plant A excludes changeovers, counts a two-minute stop threshold, and buries a daily warm-up in planned exclusions. Plant B counts changeovers, logs every stop, and keeps planned time tight. Run the identical machine and Plant A reports 94% while Plant B reports 84%, and Plant B has the more useful number, because its availability moves when the floor actually improves.
This is why comparing availability across sites, or across lines, mostly measures how honest each site's definitions are. The only reliable benchmark is a line against its own history under stable rules. Set your target off that baseline and the loss structure underneath it, the way our guide to setting OEE targets lays out, rather than chasing a borrowed percentage. And remember availability is one of three factors, a high availability paired with a soft ideal cycle time or weak quality still leaves OEE mediocre. Availability tells you about stops; it takes all three, tracked together in your manufacturing KPIs to tell you about the line.
What is a good availability number to aim for?
The often-quoted world-class availability figure is 90%, but it is a reference point, not a law, and it means little without your own loss structure behind it.
| Reference | Availability figure | Provenance |
|---|---|---|
| Nakajima "world-class" availability | 90% | Commonly cited; from Seiichi Nakajima's TPM work, Introduction to TPM (Productivity Press, 1988) |
| Your line | Its own trend | The only benchmark that reflects your process and rules |
Two facts are worth keeping in view. The 90% world-class availability component of the 85% OEE benchmark traces to Seiichi Nakajima's original Total Productive Maintenance work, popularized in Introduction to TPM (Productivity Press, 1988); it is a widely cited target, not an audited industry statistic, and no standards body certifies it. For macro context, the Federal Reserve's G.17 release tracks U.S. manufacturing capacity utilization, a different measure than availability, but a standing reminder that real plants run well below theoretical maximums by any method. Use both as orientation, then judge your line by whether its availability is trending up under rules you didn't change.
Availability is the easiest OEE factor to compute and the easiest to fool. Nail the two denominators, log stops as events, keep the rules fixed, and it becomes a number the crew can trust and act on. Once your inputs are clean, drop them into the OEE calculator watch how availability compounds with the other two factors, and see how a plant that fixed its downtime capture did it in the CLS case study.