Costly Mistakes

The Costliest Mistakes in LED and Fixture Manufacturing (and How to Catch Them)

A field guide to the mistakes that wreck LED fixture quotes, reserves, and test schedules, each with its symptom, root cause, and a numeric fix.

The single most expensive mistake in LED fixture work is quoting a finished luminaire off the populated board cost alone. Symptom: your quote wins every bid and margin evaporates by ship date. Root cause: the LED Board Assembly Cost figure, say $18.56 per board, ignores housing, optics, driver, wiring, and final assembly labor, which together often run 3 to 5 times the board. Fix: build the fixture from the full stack. A $18.56 board plus a $22 driver, $14 housing and coating, and 6 minutes of assembly labor lands near $70 to $90, not $19.

Driver failure rate is where reserves go wrong by thousands. Symptom: warranty claims run double what you accrued. Root cause: someone plugged a supplier's headline 0.5 percent figure into Driver Failure Cost when field returns show 2.5 percent. Because variable cost scales linearly, that fivefold error turns a $5,625 reserve on 5,000 fixtures into $28,000. Fix: never use the datasheet number. Pull the rate from your own RMA data or accelerated life testing, and remember a flat rate understates the end-of-warranty spike since drivers follow a bathtub curve, so pad the last two years by 20 to 30 percent.

Confusing gross burn-in capacity with good, shippable output starves the line. Symptom: assembly builds 180 fixtures a shift but shipping is short and WIP piles up in front of the chamber. Root cause: planners read the 192 gross number (48 positions times 4 cycles) as throughput. After 88 percent chamber uptime and 98 percent first-pass yield, Burn-In Test Capacity actually clears about 166 good units. Fix: always derate gross by uptime and yield. Every point of uptime here recovers roughly 1.9 good fixtures, usually cheaper than buying a second chamber.

Photometric scheduling collapses when the stabilization allowance is dropped. Symptom: an LM-79 job booked for 8 hours runs 10 and blows the DLC submission slot. Root cause: someone scheduled raw scan time, 24 fixtures at 3 per hour, and forgot each unit needs 30 to 60 minutes of thermal stabilization plus mount, align, and report time. Fix: apply a 20 to 40 percent allowance in Photometric Test Workload before booking. The 25 percent default turns 8 base hours into the 10 you actually need. Split long goniophotometric scans out of the same blended rate, since they run far slower than a quick sphere reading.

Unit errors on efficacy and energy inputs quietly kill rebate eligibility and payback claims. Symptom: a fixture listed at 150 lm/W fails DLC verification, or a retrofit proposal shows payback that the customer's meter never confirms. Root cause: using LED-only lumens instead of luminaire output, or ignoring driver and thermal losses that cut system efficacy 10 to 20 percent below the diode spec. Fix: divide measured luminaire lumens by total system watts, not module watts. A 6,000 lumen fixture drawing 42 system watts is 143 lm/W, not the 165 you get if you forget the 8 watts of driver and thermal loss.

Carton and packaging math fails on the waste allowance and the mixed-order ratio. Symptom: the pack line stops mid-run with no cartons, or a mixed pallet arrives short. Root cause: ordering exactly one carton per fixture with no loss factor, or running Fixture Packaging Cube once for an order that holds three fixture sizes. Fix: apply 92 to 97 percent fill efficiency so 400 fixtures pull about 421 cartons, and calculate each SKU separately. A loss allowance above 10 to 15 percent is not normal buffer, it points to a die, print, or crush-damage problem worth fixing at the source.

SKU proliferation hides as a costing error long before anyone names it. Symptom: per-board cost creeps up across the catalog and the plant never seems to hit economic batch size. Root cause: every board variant carries the same fixed setup, so a $280 stencil and program load spread over a 60-unit variant adds $4.67 per board versus $0.56 at 500. Fix: run SKU Variant Complexity before adding a variant and consolidate low runners. If the setup premium exceeds 5 to 10 percent of the variable rate, the batch is too small and either the run gets combined or the variant gets dropped.

Two data-hygiene failures cut across all of it. First, stale rates: a per-board or replacement cost left unchanged after component prices or line speed moved will quietly misstate every quote by the drift, often 5 to 15 percent a year. Second, counting reworked units as first-pass good, which inflates burn-in yield and Warranty Return Rate and masks an upstream solder or lot problem. Fix both with a quarterly input review and a hard rule that only true first-attempt passes count as yield. Catching a yield slip from 98 to 96 percent early is the whole reason burn-in data exists as an early-warning signal.

Published 2026-07-01.