What a pilot run is meant to prove
A pilot production run tests the process, not the design. By the time it starts, the design should be frozen: DVT has closed, the bill of materials is released and any remaining changes are small and controlled. The pilot run answers a different question: can the factory build that design on its production line, at the planned rate, with acceptable yield and consistent quality, using parts from the suppliers that will deliver in mass production?
Most schedules place the pilot run at PVT, the last of the EVT, DVT and PVT builds. Some factories also use “pilot run” for a smaller trial build before PVT, for a pre-production build after it, or for the first build on any new line. The label matters less than the conditions. If engineers are hand-building units, parts are sorted before they reach the line, or the design changes from one day to the next, the build is another engineering build whatever it is called, and its yield will not predict mass production.
A pilot run should exercise everything mass production will use:
- Production tooling, with every mold cavity in use
- Production-intent parts received through incoming inspection (IQC)
- Released work instructions, fixtures and test programs
- The operators who will run the line, not the engineering team
- Traceability, labeling, packing and outgoing inspection
Pilot output usually has other jobs too: units for reliability testing, certification samples that match production, retail and channel samples, and early field or customer units. Plan those quantities into the build rather than taking whatever is left at the end.
Deciding the build size
There is no standard pilot quantity. The right size is the smallest build that answers the questions you need answered, so start from the questions.
The first is detection. If a defect affects 1% of units, the chance of seeing it at least once is 1 − 0.99 raised to the number of units built: about 63% in a 100-unit build and about 95% in a 300-unit build. A related rule of thumb, the “rule of three”, says that if you build n units and see no failures of a given kind, you can be roughly 95% confident the true rate is below 3/n. Zero failures in 100 units therefore only tells you the rate is probably under 3%. If your quality target sits below that, a 100-unit build cannot show you have met it.
Then add coverage and demand:
| Input | Effect on build size |
|---|---|
| Lowest defect rate you need to be able to detect | Pushes the build up, as shown above |
| Number of mold cavities, component lots and parallel test stations | Each needs enough units to show it behaves like the others |
| Units reserved for reliability, certification, samples and field trials | Adds a fixed quantity on top of the yield sample |
| How long the line must run to show its real rate | A build that ends in an hour never sees a shift change, material replenishment or operator fatigue |
| Unit cost, scrap cost and component MOQs | Pulls the build down, or limits how much you can afford to risk |
| How much changed since DVT | Many changes argue for two smaller runs rather than one large one |
For many consumer electronics programs this lands somewhere between a few hundred and a few thousand units; low-volume industrial products may run far fewer. Planning two pilot builds, with time between them to fix what the first one finds, is common and usually cheaper than finding the same problems during the ramp.
Line readiness before the first unit
A pilot run started before the line is ready measures the readiness gaps, not the process. Hold a readiness review shortly before the start date and check each item against evidence, not verbal confirmation.
- Tooling. Molds and other tools have passed their trials, dimensional and cosmetic results are approved, and every cavity is qualified. See tooling and injection molding.
- Materials. Every part matches the released BOM revision, comes from an approved source and has passed IQC. No hand-picked, reworked or engineering-stock parts.
- Documentation. Work instructions with photos are posted at each station in the operators’ language and match the current revision. Inspection criteria and the cosmetic standard (viewing distance, lighting, time per surface) are written down.
- Fixtures and test. Assembly fixtures and ICT and FCT fixtures and programs are at release revision, with test limits agreed in writing. The firmware loaded on the line is the release version, and the test sequence checks it.
- Equipment. Torque drivers, dispensers, test instruments and gauges are calibrated, with current records.
- People. Operators are trained and signed off station by station, and the line is balanced to the target cycle time.
- Traceability. Every unit gets a serial number linked to its test results and to the component lots it used.
- Packing and outgoing. Production packaging, labels and accessories are on hand, and the outgoing inspection plan (AQL levels and defect classes) is agreed.
Many of these checks overlap with a factory audit. If the factory has not been audited for this product and process, do it before the pilot, not during it.
Yield and defect tracking
Agree how yield is measured before the run, then measure it the same way every day.
First-pass yield is the share of units that pass a station the first time, without rework or retest. Rolled throughput yield multiplies first-pass yield across stations and gives the better picture of the line: three stations at 98%, 97% and 99% give a rolled throughput yield of about 94%, so six units in every hundred need attention somewhere. Final yield after rework is a separate number, useful for cost planning but not for judging the process.
Log every failure, not only the ones that end in scrap. A useful defect log records:
| Field | Why it matters |
|---|---|
| Serial number and station | Ties the failure to test data and component lots |
| Symptom, as observed | Keeps the raw evidence separate from the diagnosis |
| Root cause category | Design, part or supplier, workmanship, equipment, or test false fail |
| Disposition | Reworked, scrapped, or held for analysis |
| Owner and status | Shows what is still open each day |
Review the log every day of the build with the factory’s engineering and quality leads. Rank defects by frequency, and give each of the top ones an owner, a containment action and a date for root cause. Significant or repeating defects should go through a structured process such as an 8D report.
Two habits distort pilot data in ways that are hard to see afterward. The first is retesting failed units until they pass. A station that fails good units is a test problem; a station that sometimes passes bad ones is a quality problem. Either way, the retest policy needs a limit and a record. The second is fixing problems quietly on the line. A rework step added on day two that nobody writes down becomes a hidden cost and a hidden risk in mass production.
Golden sample sign-off
A golden sample is an approved reference unit that defines how an acceptable product looks and performs. Many teams sign a reference sample at the end of DVT. The pilot run is where that sample is confirmed, or replaced, with units built on the line by production operators, because a unit assembled by engineers at a bench says little about what the line will produce.
To sign it off:
- Pull units at random once the line has stabilized, not from the first hour.
- Inspect them against the drawings and the cosmetic standard and measure key dimensions. A first article inspection report on critical parts supports this.
- Run the full functional test and confirm the firmware version.
- Check labels, packaging and accessories as the customer will receive them.
- Record the BOM revision, firmware version and build date, then sign and seal the sample on behalf of both parties.
Store sealed sets on both sides, in the factory’s quality office and with your own team, and give a set to any third-party inspector. Add limit samples for features that need judgment, such as gate marks, color variation or texture: units that mark the boundary between accept and reject. Any later change that affects appearance or function needs a new golden sample. Samples with batteries, printed finishes or soft materials also age, so plan to replace them.
Exit criteria
Write the exit criteria before the run starts, with numbers, so the decision afterwards is a check rather than a negotiation.
| Criterion | Evidence |
|---|---|
| Yield at or above the agreed target, per station and rolled | Daily yield reports for the whole run |
| No open critical defects; top defects have a root cause and a verified fix | Defect log, 8D reports, data from units built after the fix |
| Line holds the target cycle time for a sustained period | Output per hour by shift |
| Reliability tests on pilot units passed or on track | Test reports on production-line units |
| Certification complete, or unaffected by any change made | Comparison of the tested and production configurations |
| Golden and limit samples signed | Sealed samples and the sign-off record |
| Documentation released at final revision | BOM, work instructions, test specifications, packaging specification |
| Materials for the ramp secured | Clear-to-build report against the first production orders |
| Change control in place | An agreed ECN process |
Yield targets vary widely by product: a simple accessory and a multi-board device with radio calibration will not share a number. What matters is that the target is agreed in advance, and that the trend across the run, not only its best day, supports it.
The review ends in one of three decisions. Go to mass production ramp-up; go with conditions, listing each open item with an owner and a date; or repeat the pilot after fixes. Going with conditions is reasonable, but only when no open item involves safety, certification or a defect that would reach customers.
Where Zefinity helps
For the brand, the pilot run is the last point where a process problem is still cheap to fix, so the readiness review, the daily defect review and the exit decision are worth staffing properly. Zefinity’s Launch service supports those steps with OEM/ODM partners in Asia, coordinated from Hong Kong, through NPI and technical project management, manufacturing engineering support, and QA and corrective action follow-up.
Frequently asked questions
Is a pilot run the same as PVT?
Often, yes. In the EVT, DVT and PVT sequence, PVT is the first build on the production line with production tooling and processes, which is what most teams mean by a pilot run. Some factories also use the term for a smaller trial build before PVT or a separate pre-production build after it. Agree the name with your partner, and more importantly the build conditions and exit criteria, in writing.
Can units from a pilot run be sold?
They can be, if they pass the same outgoing inspection as mass production units, match the certified configuration and carry no rework that changes the product. Many teams instead use pilot units for reliability testing, certification, retail samples and field trials. Decide before the build which units are sellable, so they are tracked and inspected accordingly.
What yield should a pilot run reach?
There is no universal number. Targets depend on product complexity, process maturity and how much automated test the line has. Set first-pass yield targets per station and for the whole line before the run, then judge both the final result and the trend during the build. A rising trend with known, fixed causes is a better sign than one good day.