What Is Low Volume Production on Xometry? A Quality Fail That Started With a CAD File

-
Why We Needed Low-Volume Production in the First Place
-
What Is Low Volume Production on Xometry?
-
Xometry's Injection Molding Service Did Its Job
-
Then the CNC Brackets Failed Inspection
-
We Almost Solved It With a Laser Welder. Then We Didn't.
-
The AI 3D Printer Didn't Catch It Either
-
How Should I Prepare CAD Files for CNC Machining?
-
The Real Cost of Skipping CAD Prep
I don't enjoy rejecting parts, but it comes with the title. For the past four years I've been the quality gate at a small instrumentation manufacturer, which means I review every incoming production part before it reaches our assembly line. Roughly 2,000 individual components a year. I expect our end-of-2024 audit to show that I rejected about 11% of first deliveries. Most of those were legitimate supplier misses. One of them wasn't. The lot of 200 CNC-machined brackets that failed inspection last March was, at its root, a mistake in the CAD package I approved.
Why We Needed Low-Volume Production in the First Place
Last March we launched a pilot run of a redesigned wireless temperature transmitter for industrial cabinets. The pilot quantity was 200 units, which is the manufacturing version of a no-man's-land. Too many parts for our prototyping 3D printer. Too few for a traditional injection molder to take us seriously.
Our regular injection molding supplier quoted a steel mold at around $30,000 and wanted a 5,000-piece minimum order. We wanted 200 housings and 200 small aluminum brackets. That's when a colleague said, just upload the files to Xometry and get an instant quote.
What Is Low Volume Production on Xometry?
If you're asking what is low volume production on Xometry, the short answer is: the space between prototyping and mass production. In practice, that meant a few hundred parts made with production processes, not 3D-printed prototypes, but without the cost of full-scale production tooling.
For the plastic housing, Xometry recommended low-volume injection molding with aluminum bridge tooling. Aluminum tooling costs a fraction of steel production tooling and is fine for a few thousand shots. The quote engine came back with a price and lead time in about an hour. I uploaded the bracket model separately for CNC machining and got that quote too.
Honestly, the part that surprised me wasn't the speed. It was the clarity. The quoting screen forced me to choose material, surface finish, quantity, and tolerance assumptions before I sent anything. Everything I'd read about low-volume manufacturing said the fight would be about minimum order quantities and cost per piece. In practice, the fight was about file quality. Nobody warns you about that.
Xometry's Injection Molding Service Did Its Job
I'll get the boring good news out first: the Xometry injection molding service delivered exactly what it promised. The ABS housings showed up around five weeks later, and I went through them with the same calipers I use on everything else. Gate vestige was clean. Ejector pin marks were visible but within reason. I found two parts with minor cosmetic blemishes and one with slight flash near a boss.
Not perfect. But workable. For bridge tooling, the consistency was better than I expected, and the dimensional check on critical mounting features was well within tolerance. That part of the order cost us about $4,800 for 200 pieces, if I remember correctly, including the aluminum tool. The housings passed.
Then the CNC Brackets Failed Inspection
The mounting bracket was a simple 6061 aluminum part: a milled face, two clearance holes, and two blind M4 tapped holes on the edge. My drawing note said, 2X M4 x 0.7 tapped hole, 12 mm deep.
In my head, that meant 12 mm of full thread. In the machinist's head, it apparently meant drill a 12 mm deep hole and tap as much of it as physically possible. A tap cannot cut threads all the way into the flat bottom of a blind hole, so the actual full-thread depth came out around 8.5 mm.
During incoming inspection, I pulled a sample of eight brackets. Two failed a thread depth check. Then we checked all 200. Thirty-four had full thread depth below 10 mm, and several were worse. Our assembly team used screws to hold a mounting plate. Anything under 8 mm of engagement was a pullout risk in a vibration environment.
I rejected the lot. The machining partner, which I found through Xometry's marketplace, pushed back:
We followed the drawing. The hole is 12 mm deep. The tapped depth is within industry standard for that hole depth.
According to Xometry's published CNC manufacturing standards, the default linear tolerance is ±0.005 inches, so hole position wasn't the issue. The issue was drawing intent. I opened the original PDF and stared at my own note for a long time.
The note didn't say minimum full thread depth. It didn't specify drill depth separately. It just said the tapped hole was 12 mm deep. The shop was technically not wrong, and that made the rejection much more expensive, because it meant the fix was on my side.
We Almost Solved It With a Laser Welder. Then We Didn't.
While I was deciding what to do, our operations manager forwarded a machinery dealer's newsletter with the subject line Laser Welder Sale — 20% Off Pulse Welders. His idea was to buy a small laser welder and use it to repair future defective parts in-house, starting with these brackets.
I asked him what exactly he planned to weld. The brackets didn't have cracks or porosity. They had missing thread depth. No amount of laser welding was going to create threads, and welding on a 6061 bracket would create a new set of metallurgical problems. Tempting. Also wrong.
The AI 3D Printer Didn't Catch It Either
Here is the part of the story that still bothers me. Our R&D team had printed prototype brackets on a 3D printer with AI defect detection, and the software had flagged zero issues. The model looked perfect. The AI check passed. The prototype assembled fine.
The surprise wasn't the printer. It was the gap between what an AI 3D printer monitors and what a CNC machinist needs. The AI watches for layer shifts, warping, and print failures. It does not read your drawing notes, because an additive model doesn't need them. Subtractive manufacturing does. If the design intent exists only in your head, no AI on your printer will catch that.
How Should I Prepare CAD Files for CNC Machining?
After the rework request, I sent the corrected package to the machining partner. This time the drawing said, M4 x 0.7 — 6H. Minimum 12 mm full thread depth. Drill depth 16 mm. They drilled the holes deeper, re-tapped all 200 brackets, and shipped them back. The second batch passed inspection completely.
So, how should I prepare CAD files for CNC machining? Here is the practical answer I now give to anyone on our team:
- Export a STEP file (AP214) plus a 2D PDF with annotations. A STEP file carries geometry but not necessarily your drawing notes. For machined parts, the PDF is where intent lives. Do not send an STL and expect threads or tolerances to survive.
- Check units before uploading. Confirm the export is millimeters. A part scaled by 25.4 is an expensive way to learn this.
- Define your critical dimensions and tolerances. If you don't mark them, the machinist assumes general tolerances. Xometry's standard CNC tolerance is ±0.005 in. Ask yourself if that is good enough for every feature.
- For blind threaded holes, specify both full thread depth and drill depth. Write it like this: M4 x 0.7 — 6H, 12 mm minimum full thread, 16 mm drill depth. Ambiguity here is exactly what caused my rejection.
- Add material, finish, and edge break requirements. 6061-T6 clear anodized is a different part than 6061-T6 as-machined. If you care about burrs, say so.
The Real Cost of Skipping CAD Prep
The rework added about $2,300 to the project in extra shipping and vendor charges, plus eleven days of delay. What stings is that a twenty-minute review of my own drawing before the initial quote would have prevented all of it.
I've been doing quality control for long enough to know that prevention beats inspection. But I applied that logic only to physical parts. I checked incoming goods while sending out vague CAD files. That's like installing a security camera and then leaving the door open.
Since March, I run every CNC file through a short pre-flight review before upload. It takes about twenty minutes per file. The low-volume production parts we have ordered through Xometry since then have been delivered without a single full-lot rejection. The injection-molded housings are still running on the same aluminum tool. The discipline, it turns out, is just another inspection. It just happens earlier, on my screen instead of theirs.