Output 20% Below Plan While Every Machine Is Running: The Five Hidden Causes
Why is output 20% below plan when every machine is running? Because "running" counts minutes, not good pieces. As of 17 Oct 2026, the gap usually sits in five places: slow cycles, micro-stops, operators waiting, rejects counted late and machines running empty. At ASI 2023-24 averages (MoSPI), each lost point costs a 300-person plant about ₹19.6 lakh of sales a month.
By the Mama Editorial Team. We scope camera and floor-visibility projects for Indian factories. Last checked: 17 Oct 2026. All ₹ figures in the worked example are stated assumptions, not benchmarks.
Key takeaways
- "All machines are running" is a statement about time. Output depends on time × speed × good pieces, and the shift register records only the first.
- Five causes explain most of the gap: speed loss, micro-stops, operator waiting, rejects and rework counted late, and running empty (machine on, nothing loaded).
- Each cause shows itself on a 15-minute walk if you know where to stand and what to count.
- Put ₹ on lost output at contribution, not at sales value. In the worked example below, a 20% gap at an assumed ₹15 crore a month plan costs about ₹60 lakh a month in contribution.
- Cameras see stops, waiting and empty running. They don't see the speed setpoint or the quality inside a part. Those need the controller and the QC lab.
How this differs from the register-versus-month gap
If your problem is that the shift register says 85% and the month's dispatches say 60%, that is a bookkeeping reconciliation. Our article on the register-versus-month efficiency gap covers it.
This page is about a different complaint: every machine is switched on and humming, yet the day's pieces are a fifth short of plan. The time is being spent, just at the wrong speed, in tiny stops, in waiting, on parts that will be scrapped, or on machines turning with nothing in them.
In OEE language, almost all of this sits in the Performance and Quality factors, not Availability. If you need the formula first, read OEE explained for Indian factories, or put your own figures into the OEE calculator.
How big is one point of output? The ASI anchor
MoSPI's Annual Survey of Industries for 2023-24, released on 27 August 2025, puts gross output per person engaged in registered manufacturing at ₹78,24,322 a year (PIB release on ASI 2023-24, Table 2).
The same table gives ₹7,196 lakh (about ₹72 crore) of output and 92 persons engaged per factory in operation for 2023-24.
The newer ASI 2024-25 press note of 30 September 2026 lists output of ₹1,652,384,144 lakh (about ₹165 lakh crore) and 20,996,913 persons engaged, which works out to about ₹78.7 lakh per person (MoSPI press note, ASI 2024-25, Table 3).
Industry matters. Dividing each NIC 2-digit industry's share of all-India output by its share of persons engaged gives output per person by industry (our arithmetic on the PIB ASI 2023-24 Table 3):
| NIC 2-digit industry (ASI 2023-24) | Share of output (%) | Share of persons engaged (%) | Output per person engaged (₹ lakh/year, derived) | 1 point of output, 300-person plant (₹ lakh/month) |
|---|---|---|---|---|
| All manufacturing | 100 | 100 | 78.2 | 19.6 |
| Food products (NIC 10) | 13.23 | 11.06 | 93.6 | 23.4 |
| Motor vehicles, trailers (NIC 29) | 8.46 | 7.01 | 94.4 | 23.6 |
| Basic metals (NIC 24) | 14.49 | 7.77 | 145.9 | 36.5 |
The last column is 300 × output per person ÷ 12 months × 1%. Textiles (NIC 13) and apparel (NIC 14) appear in the release only by share of employment (8.75% and 6.83%), so they are left out rather than guessed. Labour-intensive plants usually sit below the all-manufacturing average, which petroleum refining and basic metals pull up.
The five causes, and how to see each on a 15-minute walk
1. Speed loss: the machine runs, but slower than rated
What it is. The machine cycles below its ideal speed. OEE.com, a reference site run by the monitoring-equipment maker Vorne, defines reduced speed as time "where equipment runs slower than the Ideal Cycle Time", with causes such as worn equipment and poor lubrication (OEE.com, six big losses). Also ask whether anyone turned the speed down after a bad batch and never turned it back.
On the walk. Stand at one machine with your phone stopwatch and time ten cycles. Compare it with the cycle time on the process sheet or the speed on the nameplate. A 10% gap never appears in the register, because the machine was "running" all shift.
₹ method. Lost pieces = run hours × (rated rate − actual rate). Value them at contribution per piece.
2. Micro-stops: the two-minute stops nobody writes down
What it is. Short stops for jams, misfeeds, a stuck carton or a sensor that needs a tap. OEE.com describes them as stops "well under five minutes" that do not need maintenance staff, and says "most companies do not accurately track" them (OEE.com).
On the walk. Watch one machine for ten minutes and count the stops, however short. Three stops of 40 seconds in ten minutes is 12 minutes an hour lost on that machine alone, if the pattern holds through the shift.
₹ method. Stops per hour × average stop length × run hours = lost minutes. Lost minutes ÷ cycle time = lost pieces.
3. Operator waiting: the machine is ready, the person is not, or the other way round
What it is. The operator is present but waiting: for material from the store, for the forklift, for first-piece approval from QC, for the setter to finish another machine. The register shows a manned machine.
On the walk. Count, at one moment, how many operators are standing at machines that are not cycling, and ask each what they are waiting for. The answers that repeat tell you which system to fix. This is a flow problem, not a verdict on people.
₹ method. The wages are already paid, so the cost is the output not made. For context, ASI 2023-24 gives average emoluments of ₹3,65,748 per person engaged a year (PIB, ASI 2023-24, Table 2): about ₹91 lakh a month for 300 people, paid whether or not pieces come out.
4. Rejects and rework counted late
What it is. The counter at the machine credits a piece the moment it is made. Final inspection rejects it hours or days later, or sends it back for rework that takes machine time again. OEE.com splits this into startup rejects ("from startup until stable … production") and process defects during steady running (OEE.com).
On the walk. Go to the rejection bin and the rework corner, not the machine. Read the date tags. Ask one question: at what point does the register count a piece, before or after final QC? If before, your plan is being measured against pieces that may still be scrapped.
₹ method. Scrapped pieces cost their material and processing as well as the lost contribution. Reworked pieces cost machine time twice. Price both before you price the lost contribution.
5. Running empty: machine on, nothing loaded
What it is. Motors, conveyors, heaters and hydraulics run with no part in the machine: waiting for the next coil, the next mould, the next trolley of blanks. The classic TPM loss list files it under "Idling and Minor Stops" (OEE.com).
On the walk. Look for conveyors moving with gaps, presses cycling with no blank, ovens on with empty belts, and spindles turning with no tool in cut. Note which machine upstream is starving them.
₹ method. Lost pieces as above, plus the power drawn while empty. If you want hours rather than impressions, see how to see machine idle hours from your phone.
Worked ₹ example: splitting a 20% gap in a 300-person plant
Assume a 300-person engineering plant whose production plan is worth ₹15 crore of sales a month. That is ₹60 lakh per person a year, below the ASI 2023-24 all-manufacturing average of ₹78.2 lakh, because we assume a more labour-heavy mix.
Assume a contribution margin (sales minus materials, power and other variable costs) of 20%. For scale, ASI 2023-24 gross value added was 16.0% of output across registered manufacturing (₹24,58,33,605 lakh of GVA on ₹1,53,27,16,609 lakh of output, PIB, Table 1). GVA is not contribution (it deducts all purchased inputs and no wages), so use your own costing figure.
Actual output is ₹12 crore a month, 20% below plan. Each point of plan is worth ₹15 lakh of sales and ₹3 lakh of contribution. The split across the five causes is an assumption for illustration:
| Cause | Points of plan lost (assumed) | Sales value lost (₹ lakh/month) | Contribution lost at 20% (₹ lakh/month) | Also costs |
|---|---|---|---|---|
| 1. Speed loss | 6 | 90 | 18 | Nothing extra, so it is often ignored |
| 2. Micro-stops | 5 | 75 | 15 | Operator attention, line stoppages downstream |
| 3. Operator waiting | 4 | 60 | 12 | Wages already paid for idle minutes |
| 4. Rejects and rework counted late | 3 | 45 | 9 | Scrapped material and repeated machine time |
| 5. Running empty | 2 | 30 | 6 | Power drawn with no output |
| Total | 20 | 300 | 60 |
Read the table this way:
- ₹60 lakh a month, about ₹7.2 crore a year, of contribution is at stake, but only if customers would have bought the extra output. If they would not, the gap is idle capacity: the loss is the wages and power you paid for, not the contribution.
- Speed loss and micro-stops make 11 of the 20 points here. Both need a stopwatch or machine data, not a meeting.
- Rejects counted late look small in points but cost the most per piece, because the material is gone too.
Your split will differ. The method is what transfers: measure each cause for one week on your bottleneck machine, then rebuild this table with your own numbers. To price downtime properly once you have the hours, see the real cost of factory downtime.
Start at the bottleneck
A 20% gap in plant output is usually a gap at one or two constraint machines that everything else waits for. Five percent more speed on a machine that already waits for parts adds nothing. Walk the five checks at the bottleneck first, then at the machine feeding it.
Put the five numbers on one weekly sheet: cycles against rated, stops per hour, operators waiting, pieces rejected after counting, machines running empty. Four weeks of it beat a year of "all machines running".
What cameras can and can't tell you here
What a camera can see. Whether a machine is cycling or stopped, minute by minute, including stops of under a minute that no one writes down. Whether an operator is at a stopped machine. Whether a conveyor or press is moving with nothing on it. That covers most of causes 2, 3 and 5, and part of cause 1 where the cycle is visible.
What a camera can't see:
- The speed setpoint. Video can count visible cycles, but whether the machine is set to 80% of rated speed lives in the controller or the HMI, not in the picture.
- Quality inside the part. Porosity in a casting, a weak weld, a seal that leaks under pressure, a dimension off by 0.05 mm: these need gauges, a QC lab or the customer's test.
- Anything outside the frame. Racks, poor lighting or a lens aimed at the wrong place leave holes.
Keep it about flow. "Press 4 waited 40 minutes for blanks after lunch" is a feeding problem. Count people in aggregate and rank no one. Footage of recognisable workers is personal data under s.2(t) of the DPDP Act, 2023; tell workers in writing what is recorded and why (see the worker CCTV notice under DPDP).
Where Mama fits
Mama reads a factory's existing cameras, or cameras it adds, and every morning sends the owner a short plain-language note on where hours leaked yesterday: which machines stood in short stops, where operators waited, which lines ran empty, valued at the owner's own numbers. It watches machines and flow, not named people, and it leaves setpoints and part quality to the controller and the QC lab. Mama can draw a camera plan for your bottleneck machines. Leave an email or phone in the form below.
FAQ
My supervisors say all machines ran the full shift. Are they wrong? Usually not. They are reporting time, and the time was spent. The missing output is in speed, short stops, waiting, late rejects and empty running, which a shift register is not built to record.
Which of the five causes is usually the biggest? No such split by cause appears in the MoSPI ASI 2023-24 or 2024-25 releases (checked 17 Oct 2026). It depends on the process. Run the five walk tests at your bottleneck for a week and rank the causes by lost pieces.
Should I value lost output at sales price or margin? At contribution margin, if you could have sold the extra pieces. If demand was the limit, value the gap at the wages and power you paid for no output, which is a smaller but still real number.
Can OEE software fix this? Machine-connected OEE systems measure speed and micro-stops well when the machine has a controller to read. Older machines without a PLC need sensors or cameras. None of them will fix a loose cycle-time standard or late QC counting; those are process decisions.