From RFQ to Pilot Run: How FPV OEM Prototyping Actually Works

Most FPV buyers have seen the marketing version of OEM: send a logo, receive a drone. The real process has more steps, more failure modes, and considerably more paperwork — and every step is an opportunity for the product you approved to quietly become a product you didn't. We run OEM builds out of our own Shenzhen lines, and we've also received enough samples from other factories to know where projects go wrong. This article is the process as it actually happens on the floor: what each stage produces, what it costs, how long it takes, and which of your decisions need to be locked before mass production, not after.

The five stages, and what each one produces

A serious OEM programme moves through five gates. Skipping any of them doesn't save time; it just moves the discovery of problems to a stage where they cost more.

1. RFQ and feasibility review — you send requirements, the factory confirms they're buildable and quotes. Output: a priced, spec-anchored quotation.
2. DFM review — the factory's engineers push back on your spec. Output: an agreed build spec (the document everything later references).
3. Sample / golden sample — first physical units built on prototype tooling or existing tooling. Output: the golden sample, signed and sealed.
4. Pilot run — a small batch built on the production line, with production jigs and production operators. Output: pilot data and a signed-off QA standard.
5. Mass production — where every earlier shortcut surfaces.

The vocabulary matters when you compare quotes. A factory that offers "free samples in 3 days" is usually shipping you a hand-modified unit from its own product — fine for a logo check, useless as a production commitment. That's a sampling exercise, not prototyping.

What a working RFQ actually contains

We've covered the component-level detail elsewhere — see our flight controller sourcing guide and the motor spec decoder. At programme level, a usable RFQ answers five questions:

• What is it? Product category, size class, cell count, target retail tier. "5-inch freestyle BNF, mid-range" is a starting point; "6S 5-inch freestyle BNF with analog VTX, target landed cost under $180" is a spec.
• What stays and what changes? Are we modifying an existing platform (fast, low NRE) or building from a clean sheet (slower, real tooling costs)? This single sentence changes the timeline by six to ten weeks.
• Volumes and ramp. Pilot quantity, first production quantity, and expected monthly volume after ramp. A factory plans differently for 200 units a year than 2,000 a month — and the quote differs too.
• Target markets. FCC or CE or both drives VTX power limits, labelling, and firmware regional configuration. If you don't say, we have to ask, because shipping 800mW into the EU is our compliance article come to life as a customs problem.
• Branding scope. Shell colour, silkscreen, retail box, manual language, stickers — each is a line item with tooling or MOQ implications.

DFM review: the pushback you should hope for

DFM — design for manufacturability — is the stage where a factory tells you things you don't want to hear, and you should be grateful. On a recent cinewhoop programme, the customer's spec called for a 3D-printed TPU duct mount bonded to a carbon plate with an adhesive we knew failed above 60°C — a realistic airframe temperature in summer. Flagging it at DFM cost one email. Finding it after 3,000 units shipped would have cost a warranty programme.

Typical DFM findings on FPV hardware:

• Component placements that can't be reflowed or reworked with standard tooling.
• Fastener choices requiring a driver nobody's customers own.
• Weight targets that ignore the actual mass of the wiring harness — commonly 8–15g on a 5-inch build, a figure that surprises people who spec airframes to the gram on CAD.
• Motor and prop combinations that will run hot, the failure chain described in our motor article.

If a factory returns your RFQ with no questions and no objections, that's not efficiency — it's a salesperson accepting a spec no engineer has read.

The golden sample: what it locks, and what it doesn't

The golden sample is the reference unit both parties keep. Once signed off, it becomes the arbitration standard: production units are compared against it for appearance, weight, workmanship and measured performance. We photograph, measure and seal ours, and the customer keeps a matched unit.

What it locks: build quality, materials, colour, weight, measured parameters within stated tolerances — thrust on a stated prop and voltage, current draw, RF output, flight behaviour.

What it doesn't lock: anything you didn't specify. A golden sample proves the factory can build your product correctly once. It doesn't prove the production line will — that's what the pilot run is for. It also doesn't lock components unless the build spec names them by part number. If your spec says "quality ESC" and the golden sample had a particular MOSFET, you have no claim when production substitutes. Write part numbers into the spec; that's the entire trick.

Pilot run: the stage buyers try to skip

The pilot run is a production rehearsal — typically 30–100 units for complete aircraft, 50–200 for components — built on the real line with real jigs and real operators, not the sample room. Its purpose is to find the failures that only appear at production pace.

On our own lines, pilot runs consistently surface a predictable short list:

• Assembly-sequence problems. Something that took a technician 20 minutes with tweezers takes the line 4 minutes and a different result.
• Tolerance stack-ups. Each part individually within tolerance, but the combination jams, rattles or misaligns. Fix is usually a jig or one part's tolerance tightened.
• Firmware and config drift. Units flashing correctly on the bench but missing a parameter set in volume flashing — the classic cause of "different batches fly differently".
• Packaging damage. The retail box survives one careful shipment; the pilot run tests the actual carton configuration.

A pilot run should end with a signed QA standard: AQL level for the incoming inspection, the specific bench tests each unit passes, and the acceptance criteria — the same test philosophy we apply to our own retail products, described in the factory QC article. If you skip the pilot and fold it into the first production order, your first container is the pilot. That's an expensive way to learn.

Timeline reality, in weeks

Across the programmes we run, these numbers hold up well:

The biggest schedule killer isn't engineering — it's decision latency. Every round of samples restarts a clock measured in shipping days plus queue days. Buyers who batch their feedback into one consolidated revision per round routinely finish two rounds faster than buyers who drip comments one WhatsApp message at a time. Component lead time is the second killer: a specific MCU or gyro in shortage can add 8–12 weeks regardless of enthusiasm, which is why the sourcing guide recommends asking about second-source options up front.

Tooling, NRE and who owns the mould

Two cost categories confuse first-time OEM buyers. NRE (non-recurring engineering) covers design, programming and jig-making — usually quoted separately, sometimes amortised into unit price above a volume. Tooling covers physical moulds: plastic shells, canopy parts, retail boxes. Mould costs run from a few hundred dollars for simple silicone parts to $3,000–15,000 for injection-moulded canopies or shells.

The question to ask in writing: who owns the tooling? If you paid for it, it's yours — the factory should tag it, store it, and not run it for anyone else. If the factory absorbed it, the arrangement is usually that it runs exclusively for your volumes, but get that in writing too. We've seen buyers discover their "exclusive" canopy mould producing a suspiciously similar product for a competitor two booths down at an expo.

The five failure modes we see most

1. Spec-by-sample. Approving "the one that worked" without a written spec, then losing the thread when people change jobs on either side.
2. Skipping DFM. Overriding factory pushback without engineering review, then paying for it at pilot or in the field.
3. Compressing the pilot. Merging pilot into first production to hit a season — then recalling the season.
4. Unpriced changes. "Small change" to the shell after golden sample sign-off, discovered at invoice time as a tooling charge.
5. No change-control. Six months in, a component EOL notice gets handled by email consensus, and nobody can later say which units shipped with which part. A change log with signed approvals costs nothing and answers everything.

Bottom line

OEM prototyping is not a favour a factory does you between production runs — it's a quality instrument, and the buyers who treat it that way get products they can stand behind. The sequence that works: a specific RFQ, a DFM round where you listen, a golden sample backed by a written spec with part numbers, a pilot run on the real line, and change control from then on. Every stage you skip, the product skips too — later, at your expense.