We Doubled SMT Capacity. Why Aren't We Shipping More?
You doubled SMT capacity and dispatches barely moved because SMT was never the only gate: the slowest step after it, usually functional test, burn-in or box build, now sets your output. In our worked example (all inputs assumed), a second SMT line lifts dispatches by 11%, not 100%, and costs about ₹176 per extra board.
Last checked: 17 Oct 2026.
By the Mama Editorial Team. The method rests on Theory of Constraints and Little's Law, cited inline; every rupee figure is labelled "assume".
Key takeaways
- A line ships at the pace of its slowest step. Theory of Constraints, conceived by Eliyahu Goldratt, holds that "total process throughput can only be improved when the constraint is improved" (checked 17 Oct 2026).
- Extra SMT capacity in front of a slower step becomes racks of boards, not invoices. In our example the second line adds 1,120 boards a day of placing capacity and 124 boards a day of dispatch.
- The pile tells you where the constraint is. The step with the biggest queue in front of it and none behind it is the one setting your output.
- Fix the constraint before you buy anything. In the example, covering breaks at test and fixing one failure mode lift the test cell by 210 boards a day for about ₹39,000 a month; 196 of them reach dispatch before burn-in becomes the limit.
- When you fix it, the constraint moves. In the example it jumps from test to burn-in, then to box build. Plan for the next step before you spend on this one.
- A camera sees queues and idle stations; it does not see test results. Your test logs and yield data stay the source for pass and fail.
Why the second line did not double output
India's electronics production rose from ₹1.9 lakh crore in 2014–15 to ₹11.3 lakh crore in 2024–25, per the Press Information Bureau release of 11 Oct 2025. Under the components scheme (ECMS), production worth ₹10,34,751 crore is expected over six years, as of December 2025, per the PIB release of 3 Feb 2026. Buyers pass that volume down to mid-size EMS plants as bigger schedules.
Plants usually buy SMT because its speed is easy to quote. The steps after reflow, from wave solder to burn-in and pack, are not.
Each step has its own capacity in boards a day, and the line ships the smallest of those numbers. Doubling SMT hands the job of "smallest" to the next step down.
Theory of Constraints describes this in five focusing steps: identify, exploit, subordinate, elevate, repeat (checked 17 Oct 2026). The last step matters most here: once a constraint is broken, "recognize that there is a new constraint." A second SMT line breaks the SMT constraint. It does nothing for the next one.
The worked example: one plant, six steps
Every capacity and rupee input below marked "assume" is illustrative. Replace it with your own two-week count.
Assume a contract plant in Gautam Buddha Nagar (Noida) building a controller board that goes into a small enclosure. It runs 2 shifts × 8 hours = 16 hours a day, 26 days a month. Before expansion it had one SMT line placing 100 boards per placing hour at 70% line efficiency: 16 × 0.70 × 100 = 1,120 boards a day. It then added an identical second line.
| Step | How capacity is set (assume) | Capacity (boards/day) |
|---|---|---|
| SMT, 2 lines | 2 × 16 h × 70% × 100 boards/h | 2,240 |
| THT insertion + wave solder | Bench and wave throughput, assumed | 1,900 |
| Box build (enclosure, harness, screws) | Assumed from the operation bulletin | 1,600 |
| Functional test, 7 stations | 960 min × 80% usable ÷ 4 min a board = 192 tests per station; 1,344 tests ÷ 1.08 for an 8% retest loop | 1,244 |
| Burn-in | 480 rack slots × 3 cycles a day (4 h soak + 30 min load/unload) | 1,440 |
| Pack and dispatch | Assumed | 2,000 |
The smallest number is functional test at 1,244 boards a day. Before the second line, SMT at 1,120 was the smallest, so the plant shipped about 1,120. After it, the plant ships about 1,244.
That is +124 boards a day, or 11%, for a line that added 100% to placing capacity.
Two details in the test row carry most of the loss: the 80% usable time (stations stop at lunch, tea and shift change) and the retest loop (every failed board takes a second slot from a good one).
What the idle second line costs
Assume a semi-skilled operator earns ₹580 a day, ₹15,080 a month over 26 days (illustrative). Take the current notification for your trade and zone from your state labour department (Uttar Pradesh Labour Department); most states revise the DA part twice a year.
| Monthly cost of the second line | ₹ |
|---|---|
| Crew: 6 operators × 2 shifts at ₹15,080 × an assumed 1.3 loading (₹19,604 each) | 2,35,248 |
| Interest on an assumed ₹4 crore line at an assumed 10% a year | 3,33,333 |
| Cash cost a month (before power, spares, depreciation) | 5,68,581 |
| Extra boards shipped (124 a day × 26 days) | 3,224 |
| Cash cost per extra board shipped | ≈ ₹176 |
Assume a conversion value-add of ₹50 a board. The second line, as the plant runs today, loses about ₹126 on every extra board it enables. That is the test cell's fault, not the line's.
The combined SMT lines now use 1,244 ÷ 2,240 = 56% of their placing capacity. The other 44% is paid for and idle, or worse, it is running.
When SMT keeps running anyway
Running the idle line because the crew is there moves the problem from the wage bill to the store.
If both lines run flat out, 2,240 − 1,244 = 996 boards a day join the queue in front of test. Assume ₹600 of material in each board. That is about ₹6 lakh a day of components soldered into boards that cannot ship.
Little's Law puts a time on the pile. John Little's 2011 review in Operations Research gives the manufacturing form as TH = WIP/CT: throughput equals work-in-process divided by cycle time. It adds that "any increase in output requires either an increase in work-in-process inventory or a reduction in cycle time or both, unless some change is made in the manufacturing process itself."
Turned around, the queue in days is WIP divided by throughput. After six days of full running, about 6,000 boards wait in front of a step that passes 1,244 a day: roughly 4.8 days in queue for every new board, before any rework. A board that waits five days also waits five days for its fault to be found.
This is the "subordinate" step in Theory of Constraints: run SMT to the pace of the constraint, not to the pace of SMT.
Fix the constraint before buying
The "exploit" step means getting more out of the constraint with what you already have. At a test cell that usually comes down to two things.
Stop the station stopping. If test runs through lunch, tea and shift change with a relief operator, usable time rises. Assume 80% rises to 90%: 960 × 0.90 ÷ 4 = 216 tests a station, 1,512 for seven stations. Assume one relief operator per shift: 2 × ₹19,604 = ₹39,208 a month.
Shrink the retest loop. Find the single most common failure on your test log and fix it upstream. Assume it halves retests from 8% to 4%. Good boards: 1,512 ÷ 1.04 = 1,454 a day.
That lifts the test cell by 210 boards a day over 1,244 without a rupee of capex. Burn-in caps dispatch at 1,440, so 196 extra boards a day actually ship. At an assumed ₹50 value-add, 196 × 26 × ₹50 is about ₹2.55 lakh a month of extra conversion for a ₹39,208 relief crew.
Test-equipment vendor Keysight calls a chain of separate in-circuit, flash and functional-test stations "unsuitable for high-volume production due to its constraints on test throughput" (vendor blog, June 2024). Parallel test is an "elevate" option; price it after exploiting the stations you own.
Then the constraint moves
| Stage | Test (boards/day) | Burn-in | Box build | Output | vs 1,120 before expansion |
|---|---|---|---|---|---|
| After second SMT line | 1,244 | 1,440 | 1,600 | 1,244 | +11% |
| After test fix (breaks + retest) | 1,454 | 1,440 | 1,600 | 1,440 | +29% |
| After a second rack set at burn-in (4 cycles a day, assume) | 1,454 | 1,920 | 1,600 | 1,454 | +30% |
| After an eighth test station (1,728 ÷ 1.04, assume) | 1,662 | 1,920 | 1,600 | 1,600 | +43% |
The test fix made burn-in the constraint by 14 boards a day. A second rack set lets one set soak while the other loads, cutting the cycle from 4.5 hours to 4. At 1,600 boards a day the constraint is box build, a manual step, where the levers are balance and staffing rather than machines.
For balancing a manual step against a required pace, the takt and line-balancing calculator works the same arithmetic with your own minutes.
Only now does the second SMT line start to earn: at 1,600 boards a day the two lines run at 71% of placing capacity. The order of spending was the problem, not the line.
How to find your own constraint in two weeks
The first pass needs no software, only a count every two hours of boards waiting in front of each step.
- Count the queue in front of every step. The step with the largest and steadily growing queue is your likely constraint.
- Check whether that step ever waits. A true constraint is rarely starved. If test sits idle waiting for box build, test is not your constraint.
- Watch what happens at breaks. If a step stops at lunch and the queue in front of it jumps, that lost time is lost output for the whole plant.
- Count retests separately. A board tested twice uses two slots. Your test log has this; most daily reports do not.
- Convert each queue to days. Boards waiting divided by boards that step passes a day gives days of waiting (Little's Law).
For SMT line efficiency itself, see whether your SMT line can take PLI-scale volume; for hours lost in manual zones, where a mid-size EMS line loses hours; for the rack after AOI and ICT, the SMT rework queue. If SMT really is still the constraint, the ₹ choice is in second SMT line or a third shift.
What cameras can and cannot tell you here
What a camera can see. The queue. A camera over the area in front of test, burn-in and box build shows how many trays or racks wait and how that changes through the shift. It shows test stations unattended at breaks, burn-in racks sitting full, and whether the second SMT line ran or stood.
What a camera cannot see. A test result, a failure code or a burn-in log. It cannot tell a first test from a retest unless the retest rack is physically separate, or judge the cause of a fault. Test logs and failure analysis do that.
Counts, not persons. Measure stations and queues: waits, idle minutes, rack counts. Do not score individual operators. 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 the cameras a plant already has, or a few added over the queues in front of test, burn-in and box build. Each morning it sends the owner a short plain-language note on yesterday: where boards waited, which station stood idle at the breaks, and whether the constraint moved. It watches machines, stations and flow, not people. Leave an email or phone in the form below and Mama will get back to you with where your line's constraint appears to sit and where cameras would show it.
FAQ
Why did my output not rise after adding a second SMT line? Output is set by the slowest step on the whole line, not by SMT alone. If test, burn-in or box build passes fewer boards a day, the extra boards wait in racks. In our assumed example, doubling SMT raised dispatches by 11%.
How do I know which step is my bottleneck? Count the boards waiting in front of each step every two hours for two weeks. The constraint has the largest, growing queue in front of it and is rarely starved itself.
Should I keep the new SMT line running so the crew is not idle? Running it ahead of the constraint turns wages into work-in-process. In the example, full running adds about 996 boards a day to the queue and ₹6 lakh a day of tied-up material at an assumed ₹600 a board. Run SMT to the constraint's pace and move spare operators to the constraint.
Is buying more test stations the answer? Sometimes, but only after you exploit the stations you have. Covering breaks and fixing the top failure mode added 210 boards a day of test capacity in the example without capex, of which 196 reached dispatch. Then re-run the step table, because the constraint may already have moved to burn-in or box build.