Ask five engineers which process to use for a metal component and you'll get five confident answers. Most of them will be right — for a part that isn't yours.
I've been on the receiving end of those answers for a while now. I'm the quality and brand compliance manager at a mid-size flow-control manufacturer. Every batch of sheet metal components, machined parts, fabricated weldments, and springs we buy crosses my bench before it goes into an assembly — roughly 280 active part numbers a year. In 2024 I rejected about 11% of first articles on dimensional non-conformance. Not one of those rejections happened because the supplier used the "wrong" process. Every single one happened because the process didn't match the part.
So here's the honest version: there's no best process for metal parts. There's only the process that fits your volume, your geometry, and the tolerance you actually need — which, most of the time, is looser than what's on your print.
Four scenarios cover the majority of what I see:
- Scenario A: high volume, thin gauge, flat or formed geometry → a custom metal stamping supplier
- Scenario B: tight tolerances, 3D geometry, low-to-mid volume → a precision CNC machining supplier
- Scenario C: large, low-volume, welded or assembled structures → metal fabricated parts
- Scenario D: anything that has to stretch, push, or return → large extension springs (and yes, this deserves to be its own scenario)
The three variables that actually decide it
Before you call anyone, pin these down:
- Annual volume — and how confident you are in it. Not "we think we'll sell a lot." A number you'd bet money on.
- Geometry. Flat and uniform, or three-dimensional with pockets and undercuts?
- Real tolerance. What does the assembly actually need? If you wrote ±0.05 mm because it looked good on the drawing, you're paying for something nobody's going to measure.
That third one is where most of the money leaks. ISO 2768-1 exists precisely so you don't have to tolerance every dimension individually — classes f, m, c, and v cover general linear and angular dimensions, and a stamped bracket is usually fine at class m or c. I watched a project go from a $1.80 stamped part to a $4.60 machined part because someone specified ±0.1 mm on a hole position that the mating bushing genuinely did not care about. The assembly worked identically either way.
Scenario A: You're buying volume
If you're above roughly 20,000 pieces a year, with stable geometry and material under about 3 mm thick, stamping is almost always the answer. A progressive die amortizes its tooling across the run, cycle times are measured in fractions of a second, and per-piece cost drops fast.
What you're really buying from a custom metal stamping supplier isn't the part — it's the die. So the questions that matter are about the die. Who builds it? Is it built to your print with your gauge points, or to their house standard? What's the maintenance interval, and who pays for it?
From the quality side, here's what I check. First article inspection matters, obviously. But the failure mode I see most is die wear drift — the part is perfect at 5,000 pieces and out of tolerance at 1,200. In early 2024 we received a run of aluminum sheet metal parts where the hole-to-edge distance had walked 0.9 mm past print over the course of the order. The vendor's position was that it fell "within industry standard." There isn't an industry standard that overrides your drawing. They re-ran the batch at their cost, and now every purchase order we issue includes a mid-run inspection requirement at 50% of quantity.
If your aluminum parts carry a temper spec, ASTM B209 is the reference document. 6061-T6 and 5052-H32 behave very differently in a forming die, and suppliers occasionally substitute because the alloy on the floor was what they had that week. Ask for the mill cert. It's a two-minute request and it has saved me at least three arguments.
Scenario B: You need precision, not volume
This is the counterintuitive one, so bear with me. Everyone assumes CNC machining is the expensive fallback you use until you can "graduate" to stamping. Often it's the reverse — and jumping to stamping too early is the expensive mistake.
Run the arithmetic. Say the stamped part lands at $0.85 at volume and the machined one at $3.40. Tooling for the die is $14,000, plus a first-article cycle, plus engineering changes. If your volume is 4,000 units a year and your design has any real chance of changing in the next 18 months, the stamped part costs you $14,000 plus $0.85 a unit — and locks you into that geometry. The machined part costs nothing up front and lets you revise in a week.
I've been on the wrong side of that one. We tooled a bracket in 2022 at what looked like obvious volume, then revised the mounting pattern four months later. The die rework was $6,800 and cost us six weeks (which, honestly, hurt more than the invoice did). We'd have been better off machining for another year and tooling when the design settled.
When you do go to a precision CNC machining supplier, the differentiator isn't the tolerance on their capability sheet. Almost anyone can hit ±0.005 mm in a temperature-controlled room on a single part. The differentiator is holding it on part 400 — and that comes down to in-process probing, tool-wear compensation, and whether they're checking parts or checking their setup.
So ask about GD&T literacy. If they work from ASME Y14.5-2018 and can talk through datum reference frames without deflecting, that's a good sign. If they tell you "we hold everything to ±0.01," they aren't reading your print — they're reading a number they like. And if you're in automotive, IATF 16949 plus PPAP documentation is table stakes; if a supplier can't produce a PPAP package without three follow-up emails, that's information.
Scenario C: Big, welded, and low volume
Metal fabricated parts — laser or plasma cut, press-braked, welded, sometimes finish-machined — live in a different tolerance universe. The single most common mistake I see is customers applying machined tolerances to a weldment.
Welding moves metal. Heat distortion on a 1.5 m frame shows up in millimeters, not microns. If your print says ±0.13 mm on a welded assembly, you'll get one of two things: a rejected part, or a very expensive fixture. Possibly both, and you'll pay for the fixture.
My rule: tolerance the finished features you actually locate from, leave everything else at ISO 2768 class c or v, and then tell the fabricator in writing which features are critical. Nobody argues with a clearly marked critical-to-quality callout.
Sheet metal components that get formed and then assembled are the same story. A 90° bend is not 90°. The parts we buy come in at around ±1° on bend angle unless we specifically pay for a tighter setup — and if that bend sits in the middle of a tolerance chain, one degree can eat your entire budget.
For anything headed into aerospace, AS9102 first article inspection is the expectation. Honestly, it's a useful discipline outside that industry too. Set it up once and you have a documented baseline for every rework conversation you'll ever have with a fabricator.
Scenario D: Springs are their own problem
Large extension springs get treated as an afterthought. They get bundled into the sheet metal order because, well, it's all metal. Then they show up and don't behave, and everyone stands around looking at each other.
Here's the counterintuitive part: on an extension spring, the coils are rarely what fails. The ends are. Hook type, loop orientation, bend radius, and whether the end is stress-relieved all matter more than coil count — and none of that shows up if you only spec wire diameter, free length, and rate.
Material spec matters too. Music wire (ASTM A228/A228M) and stainless spring wire (ASTM A313/A313M) have very different fatigue and corrosion behavior. If you're in a damp or washdown environment and spec music wire because it's cheaper, you'll find out at the worst possible time — usually in a customer's hands.
Rate tolerance is the other trap. The commercial springs we've spec'd generally come in around ±10% on rate (confirm that with your supplier, because it varies). If your application needs ±3%, say so up front — it changes the wire selection and the price. Discovering it after the fact means a redesign, not a rework.
And unless your sheet metal supplier genuinely does wire forming in-house, don't assume they can "just add it to the order." Subcontracting a spring line through a stamping house adds a handoff, and every handoff is a place where a spec quietly gets dropped.
Which one are you?
Answer these in order and stop at the first "yes":
- Does it need to store or release energy? → Scenario D. Springs, spring pins, and anything with a rate callout go to a wire specialist. Don't blend them into another process.
- Is any single dimension tighter than roughly 0.15 mm and critical to function? → Scenario B. Precision machining. You can machine a flat bracket; you can't stamp a 3D contoured one.
- Is the part larger than about 300 mm in any direction, or does it get welded? → Scenario C. Fabrication — and revisit your tolerances before the RFQ goes out.
- Are you confident in 20,000+ pieces a year against a stable design? → Scenario A. Stamping. If you hesitated on "stable design," go back to Scenario B.
If you answered no to all four, you're probably somewhere in the middle. Machining is the safe default — it's the only one of these four that tolerates design uncertainty without charging you a tooling penalty for it.
A word on the cheapest quote
This part matters more than the process choice. In my experience, the lowest quote has ended up costing us more in the end maybe 60% of the time. Not because cheap suppliers are dishonest — most aren't. It's because the quote is priced against a version of your part that doesn't exist yet.
That $200-per-order saving usually reappears later as one of three things: a mid-run dimensional drift you now have to inspect around, an incoming inspection you've had to move to 100% because you can't trust the AQL sample, or a six-week rework that pushes a customer commitment you'd already confirmed.
I'm not saying buy the most expensive part. I'm saying price the whole thing. Ask what the supplier's first-article process looks like, what happens when parts come in out of spec, and how long a rework actually takes. Those three answers are worth more than the delta on the quote sheet — and most suppliers will tell you straight if you ask directly.
One caveat, and it's a real one. This is written from the perspective of a mid-size manufacturer with reasonably predictable ordering patterns and an ISO 9001:2015 quality system driving the process. If you're a startup prototyping one-off assemblies, or you're sourcing internationally with long lead times, the math changes — tooling might cost less than air freight, but your rework loop is measured in months instead of weeks. Your mileage will vary.
And verify your own numbers. The figures above reflect what we saw through Q1 2026; metal pricing and tooling quotes move, and your part isn't our part.
