New Product Introduction (NPI): from prototype to mass production

Guide · New Product Introduction New Product Introduction (NPI): from prototype to mass production NPI is the disciplined path from a working prototype to a product you can build ten thousand of, reliably and at cost. It is where most hardware programs win or die. This guide explains the NPI stages, the role of a no-MOQ partner, and the scaling risks to manage before volume. What NPI actually is NPI is the structured handoff from “we proved it works” to “we can build it forever.” It is not a single step but a managed sequence that keeps cost, quality and schedule visible the whole way. Skipping it is the most common reason a promising prototype becomes an unbuildable or unprofitable product — the design is fine, the production system around it is not. The stages, in order Concept & feasibility. Can it be made at all, and at what cost? DFM starts here. Prototype (EVT). Prove the design works; surface the first manufacturability issues. The three gates. Design validation (DVT). Validate under real conditions; lock the DFM and tooling changes. Process validation (PVT). Prove the line can build it at target yield and cost, documentation frozen. Mass production (MP). Run it, with supplier management and continuous yield improvement. The three gates, in detail EVT, DVT and PVT are the three validation gates every hardware program passes between a working prototype and mass production. They are not paperwork — each one answers a specific question, and each one should produce a specific deliverable before you spend on the next. EVT — Engineering Validation Test At EVT you prove the design functions at all. You build a small batch — often 5 to 20 units — to confirm the concept, exercise the critical features, and surface the first manufacturability problems: tolerances that fight the process, features that need special tooling, materials that behave differently than the CAD assumed. The deliverable is a closed list of design changes, not a pretty part. This is where DFM begins in earnest. DVT — Design Validation Test DVT proves the design holds up under real conditions. You build a larger batch and run it through functional, environmental and reliability testing — temperature, vibration, life-cycle. DVT is where the design is frozen: DFM changes are locked, tooling is released, and the BOM is set. Changes after DVT get expensive fast, so the gate’s job is to make sure nothing major is left open before you commit to tooling. PVT — Process Validation Test PVT is the quiet gate, and the one teams skip. Here you stop testing the design and start testing the production system: can the line build the product at target yield and cost, repeatedly, with the real process and the real inspection? Documentation is frozen at PVT — control plan, parameters, test plan. If a marginal step exists, PVT is where it shows up as a yield cliff rather than as a surprise at volume. GateQuestion it answersTypical volumeKey deliverable EVTDoes it work?5–20 unitsClosed design-change list DVTDoes it survive real use?20–200 unitsFrozen design, released tooling PVTCan we build it forever?200–1,000+ unitsFrozen process & documentation Read together, the three gates are a funnel: EVT asks whether the idea is buildable, DVT asks whether it is robust, and PVT asks whether the factory can repeat it. Skip one and you import its question into the next stage — usually at higher cost and lower control. Mass production: the fourth gate MP is not the absence of gates — it is the gate that validates the whole system under sustained load. Its deliverables are the ramp plan, the supplier-management cadence, the continuous yield and Cpk improvement loop, and change control that keeps a running line stable. Why the gates matter: the change-cost rule Every gate exists for one reason — the later a defect is caught, the more it costs to fix. The familiar $1/$10/$100/$1000 curve says a change costs a dollar at concept, ten at EVT, a hundred at DVT, and a thousand at mass production. The numbers are illustrative; the shape is real: each stage multiplies the cost of a fix roughly tenfold because tooling, inventory and certification are already committed. Why a no-MOQ partner fits NPI It is a volume that changes every stage — 5 parts at EVT, 50 at DVT, 500 at PVT, 10,000 at MP. A partner with a minimum order forces you to over-buy at stages where you should be learning, not stocking. A no-MOQ shop built for NPI lets you order exactly what each gate needs, so the learning is cheap and the scaling is deliberate. The EMS build path. How no-MOQ matches the NPI ramp NPI volumes are not a smooth curve — they are a staircase. You need a handful of parts at EVT, a few dozen at DVT, a few hundred at PVT, and only then do you commit to thousands. A supplier with a minimum order quantity forces you to over-buy at every early gate, turning learning into inventory you may later scrap. A no-MOQ partner lets each gate order exactly what it needs, with no volume commitment carried forward. At our Dongguan plant the floor is one piece — MOQ = 1. That means an EVT revision, a DVT confirmation batch, and a PVT validation run can each be ordered on their own terms, so a design change never strands a warehouse of parts. Lead time scales with the part, not with a quota: straightforward machined parts turn in about 3 days, standard work in about 7, and complex parts that need many operations and post-processing run up to 30 days. A machined housing with surface treatment is a 30-day case, not a 3-day one. The cost of a no-MOQ ramp is not just the parts you do not buy — it is the scrap you do not own. A design change at DVT that would have stranded a 1,000-piece minimum instead strands a 30-piece confirmation batch. Over a typical NPI that difference compounds at every gate, and the capital stays in the product instead of in inventory. The practical effect is a deliberate ramp. You spend on learning when learning is cheap, and you only commit volume once PVT has frozen the process. The partner that does this well is built for NPI from the ground up — small batches, fast turns, and a quality system that does not require a production minimum to engage. The documentation that survives scaling What you build at PVT is not just parts — it is a controlled record: the locked BOM, approved alternates, the DFM change log, the test plan, the process parameters, and the inspection criteria. This documentation is what lets a second line or a second site build the same product. Programs that skip documentation discover, at MP, that only one person knows how it is made. Control plan and PFMEA A control plan states how each characteristic is held in spec — the dimension, tolerance, measurement method, sample frequency, and the reaction when it drifts. PFMEA is the reverse view: how each step can fail, how severe the failure is, and which control catches it. Together they turn a documented process into a controlled one. On a production program the PFMEA named incoming-material and machining failure modes; the control plan assigned 4-hour sampling to safety-critical features. PPAP for automotive programs PPAP — Production Part Approval Process — is the evidence package proving the production line builds to spec: process flow, control plan, PFMEA, dimensional report and capability study, submitted before volume ships. Not every product needs it, but the discipline behind it — a frozen process and a characterized line — is worth borrowing for anything scaling past a few thousand units. Pilot run and yield ramp The pilot run is the dress rehearsal for volume: a PVT batch built on the real line with real fixtures, operators and inspection, measured against a yield target. The yield ramp is what follows — first production lots where first-pass yield climbs to the steady-state target and every dip is chased before it becomes a habit. Two numbers matter: yield proves the line works; Cpk proves it stays in spec. The DFM review: where cost is actually set Design for Manufacture is not a single meeting — it is a running review that starts at concept and closes at DVT. The point is simple: most of a product’s cost and risk is locked the moment the design is frozen, so the cheap place to change it is early, before tooling and volume commit you to it. A real DFM note captures more than “make it easier to machine.” It records the specific changes proposed, the reason each one matters, and the cost before and after. Typical wins: relaxing a tolerance the function does not actually need, splitting a hard-to-hold feature, choosing a stock size over a custom one, or moving a feature to a face the machine can reach in a single setup. Each change is logged so the learning survives the handoff to production. Done well, DFM is a conversation between design and the floor, repeated at every gate. EVT surfaces the obvious fights; DVT locks the last compromises; PVT assumes the note is closed. If you reach PVT and the DFM note is still open, you have not finished NPI — you have postponed a cost you will pay at volume. A useful DFM note is concrete, not a wish list. It typically records: Features where a tolerance can be opened without hurting function. Geometry that needs a second operation or a special fixture. Materials or stock sizes that simplify the setup. The cost delta before and after each change, so the trade is visible. The DFM feedback loop DFM only pays off as a closed loop: the floor flags a problem, design changes the model, and the next build confirms the fix held. A one-way note is a suggestion; a loop is a control. The loop shrinks at every gate — several rounds at EVT, the last compromises at DVT, and empty by PVT, when design and process finally agree. Scaling risks to manage before volume Yield cliff. A process that yields 98% at 50 units can collapse at 5,000 if a marginal step was never characterized. PVT exists to find it. Single-source components. Fine at prototype, fatal at volume. Qualify alternates during NPI. Test coverage lag. A test plan sized for prototype does not catch field defects at volume. Scale the test with the product. Cost underestimate. Prototype quotes hide volume costs (tooling, NRE, test fixtures). Model the MP cost during NPI, not after. Buyer risk management during NPI The buyer’s job in NPI is not to chase parts — it is to keep four risks visible and close them before they become volume problems. Yield cliff. A process yielding 98% at 50 units can collapse at 5,000 if a marginal step was never characterized. Require a yield number from PVT, not an assumption borrowed from EVT. Single-source components. Acceptable at prototype, fatal at volume. Qualify alternates during NPI and record them in the BOM with approved substitutes. Test-coverage lag. A test plan sized for prototype will not catch field defects at volume. Scale the test plan with the product’s risk, not with its stage. Hidden volume cost. Prototype quotes hide tooling, NRE and test-fixture cost. Model the mass-production cost during NPI, before you commit. The discipline that contains all four is documentation plus measurement. A frozen control plan, a characterized process, and a capability index you can actually read — for critical characteristics we hold Cpk ≥ 1.67 and general characteristics Cpk ≥ 1.33 — turn “we think it works” into “we can prove it repeats.” That is the kind of evidence that survives a scale-up. None of these risks is a reason to avoid scaling — they are reasons to scale on evidence. A partner that hands you a yield number, a frozen control plan and a readable capability index at each gate turns NPI from a leap of faith into a managed ramp. NPI cost modeling: what prototype quotes hide A prototype quote is a poor proxy for production cost, and treating it as one is how programs blow the budget at MP. The prototype price usually covers a single setup, hand-finishing, and a relaxed schedule — none of which survive contact with volume. Model the mass-production cost during NPI, not after. The line items that appear only at volume are predictable: tooling and fixtures amortized across the run, NRE for test and inspection, per-part process time at the real cycle, scrap built into the yield number, and the overhead of supplier management. A credible MP model separates piece price from these one-time and recurring costs so you can see where the money actually goes. The discipline pays off at PVT. When the process is characterized and the yield is known, the model stops being a guess and becomes a number you can quote against. Programs that skip this arrive at MP with a prototype price in the business case and a production price in the invoice — and the gap is rarely small. The supplier handoff problem Many programs use a prototype shop, then move to a production shop — and lose everything learned in between. The handoff is where DFM notes, fixtures and tribal knowledge disappear. The cleaner path is one partner from EVT to MP, or at minimum a documented transfer package that travels with the product. A partner that runs the whole sequence keeps the learning in one place. How to vet a CM for the long run. Supplier collaboration: one system, EVT to MP The cleanest NPI is one partner that runs the whole sequence under one quality system, so the learning never leaves the building. When the same team does EVT, DVT, PVT and MP, the DFM note, the fixtures and the tribal knowledge stay in one place — and the buyer is not reconstructing context at every gate. For programs that demand automotive-grade discipline, the evidence is the APQP/PPAP package. On a production program that meant a full process flow (incoming inspection → CNC turning → two CNC milling operations → QC → post-processing → carburizing → cleaning → outgoing inspection → packing), a control plan with key (SC) characteristics and 4-hour sampling, a PFMEA covering incoming material and machining deviations, and live SPC on the line. It is the difference between “we can make it” and “we can prove we make it the same way every time.” Collaboration also means traceability. With an MES/ERP-linked system you can follow a part from raw-material lot through process parameters, machine, operator and inspection data — so when volume scales, a problem is isolated in minutes, not weeks. Our Dongguan plant (6,800 m², 100+ staff, operating since 2006 (Zhuohang), certified to ISO 9001 / IATF 16949 / ISO 14001) is set up for exactly this: small-batch NPI and volume production under the same roof. Good collaboration also shows up in how problems are reported. When a deviation appears, the question is not “who shipped it” but “what in the process let it through” — and the traceability data answers it the same day. That loop is what keeps a scale-up from repeating the same defect at ten times the volume. One line, prototype to mass: the Nex-G capability The practical way to keep NPI learning intact is to run every gate on the same floor. Nex-G’s Dongguan Hengli plant — 6,800 m², 100+ staff, since 2006 (Zhuohang) — pairs EMS (PCBA/SMT, component sourcing, box build and test) with CNC machining across 80+ machines (Mazak, Brother, TSUGAMI, Sodick) holding ±0.005 mm and ±0.002 mm where the part demands. One roof, one quality system (ISO 9001 / IATF 16949 / ISO 14001), MOQ = 1 from the first EVT part to the last MP lot. NPI questions to ask any partner Can you take me from EVT to MP without a handoff? If the answer is “we only do prototype” or “we only do volume,” plan the transfer now. What do you deliver at each gate? You want a DFM note, a yield number and a cost model, not just parts. How do you document so it survives scaling? The answer reveals whether MP will be repeatable. Frequently asked questions What does NPI stand for?New Product Introduction — the structured path from a working prototype to repeatable mass production, keeping cost, quality and schedule visible at every stage. What is the difference between NPI and NPD?NPD (development) covers the design itself; NPI covers the introduction of that design into production. It is the “can we build it forever” discipline that follows “can we design it.” Why use a no-MOQ partner for NPI?Because NPI volumes change every stage. A no-MOQ shop lets you order exactly what each gate needs, so learning stays cheap and scaling is deliberate rather than forced into over-buying. What is the biggest NPI risk?The yield cliff — a process that works at prototype volume collapses at production volume because a marginal step was never characterized. PVT exists to find it before MP. Do you support NPI?Yes — from EVT through MP under one quality system, with DFM notes, yield tracking and cost modeling at each gate, so the learning is not lost in a handoff. When should NPI start?At concept. The earlier manufacturability and cost are considered, the fewer expensive changes later — and the smoother the climb from prototype to volume. More NPI questions What is your minimum order quantity for NPI?One piece. MOQ = 1 across EVT, DVT and PVT, so you order exactly the batch each gate needs without carrying inventory forward from an earlier stage. What lead times can I expect?Simple machined parts turn in about 3 days, standard work in about 7 days, and complex multi-operation parts with post-processing up to 30 days. Lead time follows part complexity and process-step count, not a fixed calendar. Do you run automotive-grade quality for NPI?Yes. We hold critical characteristics at Cpk ≥ 1.67 and general characteristics at Cpk ≥ 1.33, and we have executed full APQP/PPAP packages — process flow, control plan, PFMEA and SPC — on programs such as a production program. Can one supplier take me from EVT to MP?Yes — from our 6,800 m² Dongguan plant (100+ staff, since 2006 (Zhuohang), ISO 9001 / IATF 16949 / ISO 14001), under one quality system with MES/ERP traceability, so the learning is not lost in a handoff. When does DFM start?At concept — the cheapest place to change a design. DFM runs as a loop through DVT and should be closed before PVT freezes the process. Do I need PPAP for a non-automotive product?Not formally, but the discipline behind it — a frozen process, a control plan and a capability study — is worth applying to any product that will scale past a few thousand units. NPI readiness checklist Before you scale, confirm these are in place — each is cheaper to fix at this stage than in production: Locked BOM with approved alternates for every non-trivial part. DFM note closed, with the cost before/after captured. Test plan scaled to the product risk, not the prototype. Process parameters and inspection criteria documented. Yield characterized at PVT, not assumed from EVT. One owner of the learning from EVT through MP. If all six are true, the climb to volume is a managed ramp. If not, it is a gamble — and the gaps show up as scrap and stoppages, not as a warning. Vetting a partner for the long run. Catching circuit problems at the NPI gates Validation gates exist to find problems when they are cheap to fix. The circuit layer is where the most expensive surprises hide, so it belongs in the EVT and DVT review — not left for the customer's first field return. Decoupling and stability at EVT. A prototype that "works on the bench" can still oscillate under load if decoupling is wrong. EVT should verify local decoupling at every power pin and check loop stability — the kind of issue that passes a smoke test and fails a production run. (Source: The Art of Electronics, 3rd ed., pp. 51, 729)Buyer takeaway: a partner that reviews your schematic for decoupling before building EVT is saving you a re-spin. Testability planned at NPI. Boundary scan (JTAG) lets you test assembled boards and configure programmable logic in-system — but only if the test access port is designed in from the start. Fold it into DVT planning so ICT and flying-probe gaps close early. (Source: The Art of Electronics, 3rd ed., p. 802)Buyer takeaway: ask whether DFT (JTAG, test points) is in the NPI plan, not bolted on after DVT. Starting a new product?Email your concept and target volume to [email protected] — we will map the NPI path, run DFM from EVT, and carry you to MP without a handoff.Request a quote Related articlesEVT → DVT → PVT: The Three GatesThe Complete Guide to Electronics ManufacturingContract Manufacturing in China: A Buyer's Guide