Where a Mid-Size Indian EMS Line Loses Hours: SMT vs Manual vs Box Build

By the Mama Editorial Team — we scope camera projects for Indian factories. Last verified: 2 October 2026, against the OEE and SMED literature cited inline and standard industrial-engineering definitions.

On a mid-size Indian EMS line the hours you can see and the hours you lose are not the same hours. The SMT machines log every placement, feeder stop and AOI reject, so their losses are already on a screen. The money leaks downstream — at the manual insertion bench, in box build, on the wire-harness bay and in the waiting around all three — where nothing logs anything and the loss is real but uncosted.

If you run a 200–1,000 worker contract-manufacturing plant — an SMT line or two feeding manual assembly and pack-out — you have already bought visibility into the automated half of your floor. What you have not bought is any record of what happens between the machines. This page splits an EMS line into three zones by a single question: when an hour is lost here, does anything write it down? The answer decides where you should look first.

Key points

• SMT is the data-rich zone. A modern placement machine can place more than 136,000 components per hour and logs each one. Its losses already have a number.
• Manual assembly, box build and wire harness are the blind zone. They leave a tick in a register, not a timestamped record — so their hours are lost twice: once on the floor, once in the accounts.
• Waiting is the biggest single thief. Operators idle on a kit that has not arrived, a test jig that is busy, a signature that is pending. None of it shows up as "downtime."
• The rework loop is rarely costed. Boards that circle back are twice-touched and, in many mid-size plants, counted nowhere — so yield looks better than cash.
• Video reads the blind zone, not the clean one. A camera reliably sees waiting, idle benches and line imbalance. It does not read a solder joint or a functional-test result — that is what AOI, SPI and your test jig are for.
• Start by counting, not buying. Two weeks of honest tallying at the manual benches usually finds more hours than a capex proposal does.

The three zones, and who logs each

Read the right-hand column as the honest one. These are structures, not benchmarks — do not accept a vendor's percentages for them. Your own two-week tally is the only number worth trusting.

SMT: already on a screen

The automated zone is where the industry's metrics already live. Overall Equipment Effectiveness (OEE) scores a machine's good output against its theoretical maximum, and the widely cited TPM benchmarks put world-class at about 85% with most discrete manufacturers running 60–75% (oee.com) — directional figures, not an India survey. If you want the formula and a rupee-worked example, start with OEE explained for Indian factories and what 'good' actually looks like.

The one SMT loss that hides in plain sight is changeover. High-mix Indian EMS means short runs and frequent job changes, and SMED — the method Toyota used to drive die changes from hours to minutes — exists precisely because setup time is recoverable. The machine logs the idle; it does not log why. A camera on the printer and feeder trolley turns every changeover into one number: stop to first good board. The rest of the ESD, traceability and changeover picture for electronics plants sits in cameras in electronics/EMS plants.

The blind zone: manual, box build and the waiting around them

After reflow, the honest part of the plant begins — heat-sensitive parts fitted by hand, connectors, mechanical assembly, functional test, box build and pack (SMT process). Nothing here logs a minute.

Three losses dominate, and all three are invisible to machine data:

• Waiting. The operator is present and paid, but the kit, the carton, the label reel or the test jig is not ready. This is the single largest uncosted loss on most manual lines, and it never appears as "downtime" because nobody rang a bell.
• Line imbalance. One station runs slower than the rest, so stations up and down the line alternately starve and pile up WIP. The line rate is set by the slowest bench, not the average.
• The rework loop. Units that fail test go back, get touched again, and return. In many plants that loop is not counted anywhere, so output looks cleaner than the labour bill says it is.

For the broader version of this blind spot across any floor, see the productivity loss a director can't see, and for the headcount side — are you paying for workers who are not on the line — contract labour headcount.

A worked example, in hours and rupees

One manual/box-build cell, 20 direct operators, single 8-hour shift, 26 days a month. Numbers are illustrative — plug in your own fully-loaded labour cost.

Take a fully-loaded operator cost of ₹18,000/month. Over 26 days × 8 hours that is about ₹86.5 per operator-hour.

Now suppose a modest 12% of each shift is waiting — kits late, jig busy, label reel not staged. That is well inside what a two-week tally typically finds on an unbalanced manual line, and it is conservative.

That is before a single rupee of rework or missed dispatch — and it never appeared on any machine report, because no machine was involved. A three-cell plant is on the wrong side of ₹15 lakh a year in waiting alone. The point is not the exact percentage; it is that this number is knowable, and today it is not known. For the schedule-slippage cousin of this problem, see why your shipments slip.

What a camera can and cannot do here

This is the section vendors skip, so here it is plainly.

A camera reads the blind zone well. Pointed at a bench row, the feeder trolley or the pack-out table, it reliably tells running from waiting and timestamps every transition — turning "the line felt slow today" into "bench 4 waited 38 minutes across the shift, all after the 2 p.m. kit." That is exactly the waiting, idle and imbalance that machine data misses.

A camera does not read quality or test. It cannot see a cold solder joint, judge a crimp, or tell you a board passed functional test. Anyone selling "CCTV instead of AOI/SPI" is wasting your money — optical inspection needs lighting, geometry and resolution a ceiling camera lacks. Keep AOI, SPI and your jigs; the camera's job is the time between them.

It is a process record, not a people score. The safe framing on an EMS floor is to count events at a station — waits, stops, rework entries — not scores against a name. We do not rank people, and neither should the owner's morning message.

That is the whole of what Mama does: send a short video walk of the floor and we return where the hours are being lost by zone, a camera plan showing which bench each camera should watch, and a proposal. If you prefer to use the cameras you already have, we will say so. Send the video; you get the map.

How to find your own lost hours

• Pick one manual or box-build cell and tally waiting for two weeks — kit-late, jig-busy, signature-pending. Paper and a pen beat a guess.
• Find the slowest bench. Line rate is set there, not by the average; rebalancing it is usually free.
• Count rework entries per shift. If the loop is not counted, your yield is flattering you.
• Only then decide on hardware. If the two numbers that matter — waiting minutes and rework entries — do not move after you fix staging and standard work, more cameras will not save you.

FAQ

Which zone should a mid-size EMS plant look at first? The manual and box-build benches, not the SMT machines. The SMT losses already have a number on a screen; the manual-zone losses — waiting, imbalance, rework — have no number at all, and that is where a first measurement usually surprises people.

Can a camera tell me my first-pass yield or defect rate? No. Yield and defect data come from AOI, SPI and your functional-test jig. A camera reads behaviour and time — waiting, idle benches, rework-table traffic — not the pass/fail of a board. Use both; they answer different questions.

Isn't waiting just part of manual assembly? Some is unavoidable. But most waiting on an Indian manual line is staging and kitting — kits that arrive late, jigs shared across cells, labels not pre-pulled — and that is fixable without capex once you can see it. The first step is making the waiting countable.

Will this work if we are high-mix, low-volume? Especially then. High-mix means frequent changeovers and constant re-kitting, which is exactly where waiting and imbalance compound. The metric to watch is stop-to-first-good-unit per changeover, and that is a time measurement a camera makes without touching your MES.

Do we need to wire cameras into our machines? No. The blind-zone measurement is deliberately machine-independent — it watches the bench, not the PLC. That is the point: it covers the half of the line your machine data cannot reach. Check your PoE budget before buying a switch (PoE power budget).