Mistakes

NPI, DFM/DFA and Engineering Change: Costly Mistakes and How to Catch Them

A troubleshooting guide to the errors that wreck NPI budgets and change estimates: wrong assumptions, unit slips, missed costs, and the process failures that hide them.

The most frequent NPI mistake is quoting a launch off part cost alone and ignoring the launch load. Symptom: your NPI Launch Cost total lands 40 to 70 percent under actuals. Root cause is leaving out non-recurring engineering, tooling, first-article inspection, and pilot scrap, which on a 5,000 unit consumer launch routinely add 180,000 to 320,000 dollars on top of piece price. Fix: itemize NRE separately from per-unit cost and load a launch amortization line. If tooling is 90,000 dollars over 5,000 units, that is 18 dollars per unit you cannot bury inside a 6 dollar piece price.

A classic DFM error is counting part-reduction savings without counting the requalification cost to get there. Symptom: DFM Savings shows a glowing 4.20 dollar per unit gain, but the program never breaks even. Root cause: merging two brackets into one casting saves machining and two fasteners, yet triggers a 35,000 dollar mold and a 12 week PPAP cycle. Fix: divide the one-time change cost by annual volume and subtract it. A 47,000 dollar change over 20,000 units per year is 2.35 dollars per unit in year one, which halves that 4.20 dollar claim to 1.85.

Unit and time-base errors dominate DFA Assembly Time work. Symptom: a line quoted at 62 seconds per unit actually runs 95. Root cause is entering handling and insertion times per operation but forgetting the 15 to 25 percent allowance for fatigue, reorientation, and tool changes, or mixing seconds with the TMU basis where 1 TMU equals 0.036 seconds. Fix: apply a personal, fatigue, and delay allowance of 18 percent as a default, and confirm every entry is in the same unit. Skipping a 20 percent allowance on a 62 second cycle understates real time by 12 seconds.

Engineering Change Cost estimates fail when they stop at engineering hours. Symptom: an ECN scoped at 8,000 dollars settles at 40,000. Root cause: the estimate captured 40 drawing hours but ignored scrapped inventory, in-transit WIP, retooling, re-inspection, and documentation. A single dimensional change on a purchased part can obsolete 6,000 dollars of stock and 3,000 dollars of WIP before one new part ships. Fix: force five line items on every change: engineering, inventory write-off, tooling, revalidation, and documentation. Changes that touch a fit dimension should assume at least 2.5 times the raw engineering hours.

ECO Workload planning breaks on the assumption that every change is equal. Symptom: your queue shows 30 open ECOs and leadership expects them cleared in a month, but throughput is 6 per week. Root cause: treating a Class III cosmetic redline (2 hours) the same as a Class I form-fit-function change (30 to 60 hours) in the same backlog count. Fix: weight the queue by change class before loading the ECO Workload model. Thirty ECOs at an average of 14 hours is 420 hours, which at 2 engineers and 30 productive hours each per week is 7 weeks, not 4.

Prototype and pilot estimates get sandbagged by the yield assumption. Symptom: Prototype Build Cost or Pilot Run Cost looks lean, then the build consumes twice the material. Root cause: assuming production yield during a pilot, where first-pass yield realistically runs 50 to 75 percent, not the 95 percent you will hit at maturity. Fix: buy material to a pilot yield. A 200 unit pilot at 60 percent first-pass yield needs kits for roughly 333 units, so a 90 dollar kit means 30,000 dollars of material, not 18,000. Underbuying stalls the pilot and blows the launch date.

Launch Readiness Score gets gamed into a false green. Symptom: the gate reads 92 percent ready, yet the line stops in week one on a missing gage R&R. Root cause: scoring completion of documents rather than verified capability, and weighting all criteria equally so 10 easy checks mask 2 critical gaps. Fix: make critical-path items (PPAP approval, process capability Cpk above 1.33, supplier qualification) mandatory gates, not weighted line items. A single unmet critical item should cap the Launch Readiness Score at yellow regardless of the other 90 percent.

Design Review Workload and Manufacturability Score both suffer from stale or averaged data. Symptom: reviews run long and the manufacturability rating never predicts the real defect rate. Root cause: scoring a design against generic guidelines instead of your own process limits, for example allowing a 0.5 mm wall when your molder holds 1.0 mm reliably. Fix: calibrate the Manufacturability Score to plant-specific capability and size Design Review Workload off part complexity, not part count. A 40 feature assembly needs roughly 6 review hours, while a 400 feature one needs 20, and averaging them buries the risky designs.

Published 2026-07-01.