The short answer
To improve OEE and machine utilisation you do two things the plant usually skips: measure the right number, then attack the specific loss it exposes. OEE — Overall Equipment Effectiveness — is Availability × Performance × Quality, and because the three factors are multiplied, the losses compound. A machine that scores 90% on each factor is not at 90% OEE; it is at roughly 73%. That multiplication is the whole insight: you cannot buy your way out of one weak factor by perfecting another, so improvement means lifting all three — and first knowing which is bleeding.
The practical route is to break every stoppage and reject into the six big losses, map each to the OEE factor it damages, and counter it — cut changeover time, kill minor stops, run at standard speed, design out the rejects. Utilisation is raised separately by load levelling — spreading work so no machine drowns while another idles — and by focusing on the bottleneck, since whole-plant output is set by the constraining machine. Underneath all of it sits one enabler: booking actual times and quantities at source so OEE is measured, not estimated. This guide, part of the production planning software library, works through each step.
OEE = Availability × Performance × Quality
Each factor is a ratio between 0 and 1, and each answers one honest question about the machine. Keep them separate and the number becomes diagnostic instead of decorative.
| Factor | Formula | What pulls it down |
|---|---|---|
| Availability | Running time ÷ planned time | Breakdowns and setup / changeover — every hour the machine was scheduled but not running |
| Performance | Actual output speed ÷ standard cycle time | Minor stops, idling and slow running — the machine is on, but below the standard cycle |
| Quality | Good pieces ÷ total pieces produced | Startup scrap after a changeover and rejects during steady running |
| OEE | Availability × Performance × Quality | All three compounded into one headline figure for the machine or work centre |
Availability is time the machine actually ran divided by the time it was planned to run; two things eat it — unplanned breakdowns and avoidable changeovers. Performance compares the speed achieved against the standard cycle time on the routing — a job that should take 40 seconds a piece but averages 50 is running at 80% performance, and nobody notices because the machine looks busy. Quality is good pieces over total: rework earns no standard time and scrap consumes capacity for nothing. Multiply the three and you have OEE — a single figure management can trend, provided every input traces to something the shop floor actually recorded.
Utilisation vs efficiency vs OEE
Three metrics get used interchangeably, and confusing them sends improvement effort to the wrong machine. They ask different questions about the same equipment.
The rule that makes this useful: a machine that is highly utilised but inefficient is running all shift and still falling behind standard — chase setup time, tooling and speed. A machine that is efficient but under-utilised is fast when it runs but idle too often — chase material feeding, sequencing and breakdowns. OEE catches both, but you need utilisation and efficiency separately to know which lever to pull. The companion guide on planning reports and KPIs puts these alongside plan attainment and schedule adherence, and the full derivation lives on the Plan vs Actual & OEE feature page.
The six big losses
OEE improves fastest when you stop treating it as one mysterious number and split it into the six standard losses. Two hit each factor, and naming the loss tells you which factor you are fighting.
| Loss | OEE factor | First counter-measure |
|---|---|---|
| Breakdowns | Availability | Preventive / planned maintenance and fast fault response |
| Setup / changeover | Availability | Group similar jobs; apply SMED to shrink changeover time |
| Minor stops / idling | Performance | Better sequencing and material feeding so the machine never waits |
| Speed loss | Performance | Restore the standard cycle time; check tooling and settings |
| Startup rejects | Quality | Stabilise the process faster after changeover; first-off checks |
| Production rejects | Quality | Design out the defect; catch it at source, not at despatch |
How to raise each factor
Lift Availability — cut changeover and breakdown
Changeover is attacked by grouping — sequencing similar jobs so the setup between them is small — and by SMED, converting internal setup steps that stop the machine into external ones done while it still runs. Breakdowns yield to planned maintenance and fast response when something does stop. A finite machine-loading view showing daily load and projected availability lets a planner group changeover-friendly work rather than jumping between dissimilar jobs and paying the setup penalty every time.
Lift Performance — kill idling, hold standard speed
Minor stops and speed loss are the quietest thieves because the machine looks occupied. Reduce idling with better sequencing and priority and reliable material feeding, so a machine is never waiting for the previous operation or for parts. Hold standard speed by keeping the routing's standard cycle times honest — if the floor consistently runs slower, either the tooling and settings need attention or the standard itself is wrong.
Lift Quality — stop the rejects earning nothing
Startup scrap shrinks when the process stabilises quickly after a changeover, with a first-off check before a full run. Steady-state rejects need the defect designed out and caught at the operation that makes it, not at despatch — where all the value added afterwards is scrapped with it. Every reject is capacity you paid for and cannot sell.
Guessing at OEE from a monthly report?
We can show you Availability, Performance and Quality computed live from your own work orders and shop-floor bookings — the six losses broken out by machine — in 30 minutes, on your own data.
Load levelling and the bottleneck
OEE improves the effectiveness of a machine while it runs; utilisation is a plant-level problem, and it is dragged down far more by unevenness than by any single machine being slow. If one work centre sits at 130% loading and another at 40%, the plant is not well utilised — the overloaded machine simply builds a backlog and misses dates while the idle one earns nothing. Load levelling is the fix: spread and re-sequence work so no resource is drowning while another sits empty, flattening the peaks a finite loading view makes visible.
The second lever is the bottleneck. In any routing, whole-plant output is set by the constraining operation — the slowest, most-loaded machine in the chain. Improving OEE on a machine that already has slack changes nothing downstream; improving or protecting the bottleneck raises the throughput of the entire plant. So the sequence is: find the constraint, level the load around it, and spend your OEE effort there first. That is where an hour of Availability recovered turns straight into shippable output. The guide on work order prioritisation covers how to sequence around a constraint, and common planning mistakes covers the trap of optimising a non-bottleneck.
Measure, don't estimate: booking at source
None of this works if OEE is reconstructed from memory at the end of the shift. Minor stops go uncounted, speed loss is invisible, and rejects get rounded — the number that results is a comfortable fiction. The enabler behind a real OEE programme is booking actuals at source: capturing what happened as it happens, at the machine.
- Scan the context — operators scan shift, machine and operator barcodes so every record is attributed without typing.
- Log real times — actual start and end, setting time and cycle time per operation, so Availability and Performance are computed from events, not estimated.
- Record stoppages — each halt captured against a reason, so the six losses separate themselves instead of hiding in one "downtime" bucket.
- Split OK vs not-OK — good and rejected quantity per operation, so Quality is real and rework is visible where it happens.
- Capture from the machine where possible — IoT / Industry 4.0 machine-data capture reads counts and states directly, removing even the scan.
The payoff is not just an accurate OEE today; it is the feedback loop. When real cycle times replace optimistic standards in the next planning run, plan attainment climbs — not because the floor tried harder, but because the plan finally asked for something achievable. A number you cannot trace to a booked event is an opinion; measured OEE traces to an actual start, end and quantity someone captured.
What measuring the six losses changes
A precision-component shop reported a single monthly OEE figure and nothing beneath it. When bookings moved to source — operators scanning shift, machine and operator, logging cycle times, stoppage reasons and OK-versus-not-OK quantity — the one number split into six. The picture surprised everyone: Availability was fine, but Performance was low because a family of jobs ran below standard cycle, and startup rejects spiked after every changeover. Effort that had been aimed vaguely at "uptime" moved to holding standard speed on that job family and stabilising the first-off after changeover. The monthly OEE number still mattered for trend, but now it was made of losses someone could actually attack — the profile behind real automotive and precision deployments.
Cycle-time capacity and India benchmarks
For Indian automotive and precision-component makers, capacity is cycle-time driven: plant output is a direct function of standard cycle times against machine availability, so OEE and efficiency are not vanity metrics — they are the capacity model. If the standard cycle on the routing is wrong, every downstream number is wrong: machine load is understated, the plan over-commits, attainment collapses. Booking real cycle times back against the standard is what keeps that capacity calculation honest, run after run.
Machine-shop and job-work units benchmark utilisation to decide whether the answer to more demand is another shift or another machine — a decision that only holds when the utilisation number is measured, not guessed. Indicative INR pricing for the planning module, sized for small-to-mid Indian manufacturing units, is on the pricing page; treat those figures as a starting point and confirm the commercials that apply to you with your CA rather than reading any planning report as tax advice.
How Fast Planning improves OEE
Fast Planning Software — built by Improsys in Pune under the Fast Technology brand, cloud or on-premise — turns OEE from a monthly guess into a measured, improvable number, because the plan and the actuals live in the same system.
Frequently asked questions
What is OEE and how is it calculated?
OEE — Overall Equipment Effectiveness — is Availability multiplied by Performance multiplied by Quality. Availability is running time divided by planned time, so breakdowns and changeovers pull it down. Performance is actual speed against the standard cycle time, so minor stops and slow running pull it down. Quality is good pieces divided by total pieces produced, so scrap and rework pull it down. Because the three are multiplied, a machine at 90% on each factor scores only about 73% OEE — the losses compound, which is why chasing all three matters more than perfecting one.
What is the difference between utilisation, efficiency and OEE?
Utilisation asks whether a machine's available time was used — running time divided by available time — so idle time, waiting for material and breakdowns pull it down. Efficiency asks how well the running time was used — standard time earned by good output divided by the actual time taken — so slow cycles and rework pull it down. OEE combines both, plus quality, into one number through Availability × Performance × Quality. A machine can be highly utilised but inefficient (running all shift yet below standard speed) or efficient but under-utilised (fast when it runs but idle half the day); reading the three together tells a planner which lever to pull.
What are the six big losses?
The six big losses are the standard categories that erode OEE. Two hit Availability: breakdowns (unplanned stoppages) and setup or changeover time. Two hit Performance: minor stops or idling (short, uncounted stoppages) and speed loss (running below the standard cycle time). Two hit Quality: startup rejects (scrap while a process stabilises after a changeover) and production rejects (defects during steady running). Mapping every stoppage and reject to one of the six tells you exactly which OEE factor to attack, rather than treating OEE as a single mysterious number.
How does load levelling improve machine utilisation?
Load levelling spreads work across machines and time so no single resource is drowning in a queue while another sits idle. Plant-wide utilisation is dragged down by unevenness far more than by any one machine being slow: a work centre at 130% loading cannot make up for one at 40%, because the overloaded machine simply builds a backlog and misses dates. A finite machine-loading view shows daily load and percentage loading per machine, so a planner can move or re-sequence work orders to flatten the peaks — and because output is set by the bottleneck, improving or protecting the constraining machine raises the whole plant's throughput.
Why must OEE be measured at source rather than estimated?
An OEE number reconstructed from memory at the end of a shift is a guess — minor stops go uncounted, speed loss is invisible and rejects are rounded. Booking actuals at source means operators scan shift, machine and operator barcodes and log actual start and end times, setting and cycle times, stoppages and OK versus not-OK quantity, optionally captured straight from the machine through IoT. That gives a measured OEE every factor traces to a real event, and it closes the feedback loop: real cycle times replace optimistic standards in the next plan, so the plan finally asks for something the floor can hit.