OEE Explained for Indian Factories (with a Rupee Example)

OEE (Overall Equipment Effectiveness) scores how much good product a machine makes against its theoretical maximum: Availability × Performance × Quality. A line running 85% × 80% × 95% scores 65% — it loses roughly a third of its output. At an indicative ₹15/part margin, one machine's gap works out to about ₹1.4 lakh (~₹140,000) a month.

Most mid-size Indian plants — auto-components, textile, pharma, FMCG, fabricated metal — sense their machines are "running slow" but can't put a number on it. OEE is that number. Here it is, worked in rupees for one machine, with the part almost every explainer skips: why the figure you compute by hand is the best case, not the real one.

Key takeaways

• OEE = Availability × Performance × Quality. Three "decent" ratios (85% × 80% × 95%) compound to 65%.
• World-class OEE is about 85%; most discrete manufacturers run 60–75% (widely-cited TPM benchmarks, oee.com).
• The number you calculate by hand over-reports. Un-logged minor stops quietly inflate the Performance factor, so a hand-figured "65%" is the optimistic ceiling.
• Performance losses are the hidden bucket — minor stops and speed loss that no shift book records.
• Measure your bottleneck first. A high OEE on a non-constraint machine is a warning, not a win.
• In money: ~₹1.4 lakh/machine/month at indicative margins — plug in your own number.

The number you calculate by hand is the optimistic one

Here is the catch nobody puts on the first page. Availability and Quality leave paperwork — a breakdown gets logged, a reject gets binned. Performance losses don't. A two-minute jam every twenty minutes, a machine dialled down 10% to "play safe" — nobody writes those down, so when you hand-calculate OEE, the Performance factor is a guess, and the guess is always kind.

Vendors who compare hand-logged OEE against automated machine-state measurement (FlowFuse, Mingo Smart Factory) put the gap at roughly 8–15 percentage points: a plant that hand-figures "75%" is often really running low-to-mid 60s once every micro-stop is counted. So read the worked example below as a ceiling. The real job isn't computing OEE once — it's measuring the Performance bucket instead of estimating it.

The formula: Availability × Performance × Quality

OEE's inputs are specified in the international standard ISO 22400-2:2014 so plants can be compared on the same basis (iso.org). The three-factor form below is the classic Nakajima/TPM formulation; ISO frames the middle factor as effectiveness rather than performance and differs in some category treatment, but in practice the two agree. All three are ratios you multiply:

Availability dies to breakdowns, changeovers, material waits and (very real in India) supply dips before the DG picks up. Performance dies to minor stops and reduced speed — the hidden bucket. Quality dies to defects and rework. Because you multiply, one weak link drags the whole score: 85% × 80% × 95% = 65%.

These map to the Six Big Losses of Total Productive Maintenance, the taxonomy widely credited to Seiichi Nakajima (oee.com): breakdowns and setup (Availability); idling/minor stops and reduced speed (Performance); process defects and startup/yield loss (Quality).

How do you calculate OEE? (worked example, in rupees)

One machine, one 8-hour shift. Numbers are illustrative — plug in your own margin.

Availability. Shift = 480 min, minus 30 min planned breaks = 450 min planned. Lose 45 min to a changeover and 22 min to a supply dip = ~67 min down; Run Time = 383 min. 383 ÷ 450 = 85%

Performance. Ideal cycle = 0.5 min/part (2 parts/min). Over 383 run-minutes it should make 766 parts; it made 613. 613 ÷ 766 = 80%

Quality. Of 613 parts, 31 rejected/reworked; 582 good. 582 ÷ 613 = 95%

OEE = 0.85 × 0.80 × 0.95 = 65%. (Check: 0.646 × 900-part theoretical max ≈ 582 good parts — the two agree.)

Now the money. Theoretical max over 450 planned minutes is 900 parts; at 65% you ship 582 good parts/shift. A world-class 85% line would ship ~765 — about 183 parts short every shift.

One machine, one avoidable margin, ~₹1.4 lakh a month. Multiply across a ten-machine line and OEE stops being a maintenance metric and becomes a P&L line.

Two judgment calls that decide your number

Before you trust any OEE figure, settle two things — this is where numbers get quietly fudged:

• Ideal Cycle Time — use the best rate you have actually demonstrated on the floor, not the machine's nameplate. Nameplate flatters Performance and is the single most-gamed input in OEE.
• Changeover convention — this example counts the 45-min changeover as an Availability loss. Some plants exclude planned changeover, or push it to TEEP instead. Pick one and state it, so nobody thinks the math is wrong.

Measure the bottleneck first

Here is the contrarian rule most Indian factory-ops content skips: OEE only matters on your constraint. A 90% OEE on a feeder machine ahead of the bottleneck is not a trophy — it means you're overproducing into a pile of WIP the constraint can't clear. Chasing OEE everywhere just moves inventory around. Find the machine that gates the whole line's output, measure that one honestly, and improve it first (this is the Theory of Constraints logic behind Goldratt's work). High OEE off the bottleneck is a red flag.

You already paid for the capacity

The losses cost money even when they don't stop a sale — and in India the sharpest version of this is a fixed cost, not a variable one. Your electricity bill has two parts: a demand charge on your sanctioned kVA, billed every month whether the machine runs or idles, and a per-unit energy charge (around ₹6.4–7.1 per kVAh at 11–132 kV, ₹6.1 above 132 kV, for large industrial consumers in a state like Uttar Pradesh, FY 2026 — Mercom on the UPERC order). A slowed machine draws less energy, so the per-unit charge falls with it — but the sanctioned-kVA demand charge doesn't move. You bought that capacity. Every point of Availability and Performance loss is capacity you paid for and threw away, and an idling-but-energized machine still pulls power (motors, aux) while making nothing sellable.

Where the hidden losses hide — and what a camera can and can't do

The weak factor above is Performance (80%) — the minor stops and speed loss no shift book captures. Close that blind spot without wiring every machine into a PLC or MES: a camera already pointed at the cell can tell running from stopped and timestamp every transition, turning "the line felt slow today" into a ranked record — cell 4 stopped 38 times for 71 minutes this shift; cell 2 twice. You fix cell 4 first, not the machine that merely feels slow.

Be honest about the limits, because detection is now a commodity — plenty of vendors sell "CV-OEE from your CCTV." A camera reads machine state reliably only when there's a visible cue in its sightline: motion, a part flow, or a status/indicator lamp. Idle-but-powered can look like running; truly-stopped needs an unobstructed view of the signal. On one floor we keep cameras on, a single mains outage in a two-week window ran 110 minutes (11:08 to 12:58) mid-shift — one clean, timestamped Availability loss you could see start and end to the second, where a shift book would have rounded it to "power gone, afternoon." Continuous capture is what makes losses countable.

Which is why the hard problem isn't detection — it's placement. Which camera, at which angle, watching which machine, and whether a pillar is currently blocking the sightline to that machine's run-state lamp. Pointing a camera at a machine only helps if the sightline to its running signal is clear, and that's a survey problem the detection vendors skip. That's the gap Mama closes: record a two-minute phone walk of the floor, get back a plan showing which machine each camera should watch — and where a sightline is currently blocked.

And the same feed earns its keep twice more. It's your guarding record under the Factories Act, 1948 (the fencing and machinery-in-motion duties, §§21–22, you already answer for — India Code). And once people are in frame, worker video is personal data under the Digital Personal Data Protection Act, 2023 (India Code) — whose notice, purpose and retention obligations are phasing in through 2027, so build that discipline now: monitor machines with clear notice, a defined purpose and a retention limit, the same way you already handle biometric attendance. One feed, three jobs.

What "good" looks like by vertical

The global band — world-class ~85%, typical 60–75% — hides real differences, and there's no reliable India-specific published survey, so treat this as directional, not a benchmark to quote:

The point isn't the row you land on — it's that comparing yourself to a single global average tells you nothing. Baseline your own weakest factor and beat last month.

How a plant head should start

• Measure one machine, one shift, by hand. The arithmetic above takes 20 minutes and usually shocks people — just remember it's the optimistic figure.
• Find the bottleneck and measure that one first. OEE off the constraint can mislead.
• Attack Performance losses. Biggest, least measured, fastest rupee win.
• Instrument, don't estimate. Replace the guessed Performance number with a measured one — PLC or camera — before you spend on "fixes."
• Don't chase 85%. World-class 85% (≈90% × 95% × 99.9%) is the top few percent globally. Move your number.

FAQ

What is a good OEE score for an Indian factory? There's no magic target. World-class OEE is about 85%; most discrete manufacturers run 60–75% (widely-cited TPM benchmarks). For a mid-size plant, baseline your weakest of the three factors and improve on it — and measure your bottleneck machine before any other.

How do you calculate OEE simply? OEE = Availability × Performance × Quality. Availability = Run Time ÷ Planned Production Time; Performance = (Ideal Cycle Time × Total Count) ÷ Run Time; Quality = Good Count ÷ Total Count. Multiply the three ratios. The inputs are standardised in ISO 22400-2:2014.

Why is my OEE low even though the machines look busy? Performance losses — minor stops and reduced speed — which rarely get logged. A machine can look "running" all shift yet run at 80% speed with frequent short jams. And a hand-calculated OEE typically over-reports by 8–15 points because those un-logged stops silently inflate the Performance factor.

Is OEE the same as machine utilisation? No. Utilisation asks only whether the machine was on. OEE also asks whether it ran at full speed and made good parts first time. A machine can be 95% utilised and still score 60% OEE.

Can I measure OEE without an expensive MES or PLC? Partly. Availability and Quality can start on paper; the hard part — continuous Performance/downtime data — can come from a camera watching machine state, provided its sightline to a running-state cue is unobstructed. A practical start before a full MES project.