Why Your Precision Machining Orders Keep Going Wrong (And It's Not the Supplier)

The Problem You Think You Have

You send out an RFQ for custom CNC components. The drawings are clean. The tolerances are specified. The supplier confirms they can hit ±0.005" all day long. You place the order.

Three weeks later, the parts arrive. They look right. They measure right — mostly. But when your assembly team tries to fit them into the pump housing, something's off. Not catastrophically wrong. Just... off. A few thousandths in the wrong direction. Enough to cause a press-fit failure on 30% of the batch.

You call the supplier. They pull their inspection reports. Everything's within spec. On their end.

And that's when you realize: the problem isn't that they made bad parts. The problem is that you never actually told them what you needed — you just assumed the drawing said it for you.

What's Actually Going Wrong

I learned this the expensive way. In 2021, we ordered 2,500 custom molded plastic parts for a sump pump float assembly. Simple part — glass-filled nylon, two critical dimensions, nothing exotic. We'd used this supplier before. Good shop, ISO certified, reasonable lead times.

The first batch was perfect. The second batch failed dimensional inspection on a key diameter by 0.008". Not enough to reject outright, but enough that our assembly line started seeing intermittent leaks.

I assumed it was a tooling wear issue. Didn't verify. Turned out the supplier had switched resin pellets mid-run — same material spec on paper, but a different melt flow index from a different lot. Nobody flagged it because our purchase order didn't require lot traceability. We never asked. They never told us.

A $4,100 order became a $4,100 write-off plus two weeks of production delay.

That was lesson one. Here are the others that followed:

Your drawing is not your specification

A print with GD&T symbols is a starting point, not a complete spec. What it doesn't tell the machinist:

• Which dimensions actually matter for function vs. which are just reference
• Whether the tolerance stack-up against mating parts has been checked
• What the surface finish requirement really needs to be (Ra 63? Or just "not visibly scratched"?)
• Whether material certs need to be lot-specific or just grade-specific

On a precision stamping dies parts order back in Q1 2023, we specified A2 tool steel. The supplier sourced A2 from a distributor they trusted. The hardness came back at 58 HRC instead of the specified 60-62. Their fault? Not really — we never specified which A2 or whose A2. There are at least four major mills producing A2 with slightly different chemistry. The heat treat responded differently.

Result: dies wore out 40% faster than expected. Cost us $2,800 in premature replacement plus the downtime.

"Same material" means everything and nothing

This one still stings. We needed metalworks torsion springs for a utility pump check valve. Spring steel, 302 stainless, passivated. Supplier A quoted $0.38/unit. Supplier B quoted $0.52/unit. We went with A.

The springs worked fine in bench testing. In the field, about 8% failed within 90 days — stress corrosion cracking along the wire surface. Why? The passivation process on Supplier A's line was different. Not wrong, just different. Their bath chemistry produced a slightly thinner oxide layer. Looked identical. Caliper measurements matched. But the corrosion resistance was measurably worse.

That one cost us a customer. A commercial building in Ohio had three pumps fail in one heating season. They switched to a competitor's system. I don't know the exact dollar figure, but it was way more than the $0.14/unit we "saved."

First article approval is a snapshot, not a guarantee

The biggest lie in precision manufacturing: "FAI passed, so we're good." The first article inspection tells you the process can produce a good part. It doesn't tell you it will produce good parts at scale, after the tooling has run 5,000 cycles, after the material lot changes, after the operator on second shift does things slightly differently than the guy on first shift.

For washing machine parts we produced in 2022 — a stamped stainless bracket, nothing fancy — the FAI was flawless. The next four production runs were flawless. Run five had 12% dimensional drift on the bend angle. The tooling had simply worn past the point where the original setup could compensate. Nobody noticed until our incoming inspection caught it.

We got lucky that time. Caught it before assembly. But it was pure luck, not process.

What This Actually Costs You

Let me add it up. Not to be dramatic — just to be honest about what "going cheap" or "assuming it'll be fine" actually means:

• 2021 molded plastic parts: $4,100 write-off + 2 weeks delay
• 2023 stamping dies parts: $2,800 replacement + 4 days downtime
• 2022 torsion spring failure: lost customer (~$15K/year account)
• Plus the soft costs I can't quantify: engineering time spent firefighting, assembly line rework, customer trust erosion

Roughly $18,000 in hard costs over 7 years, and one customer relationship I can't get back. (Should mention: we caught at least $12K more in potential problems using our current checklist — so it's not all bad news.)

The real cost isn't the dollar amount. It's that every one of these could have been prevented with a 15-minute conversation before the PO went out.

What I Actually Do About It Now

I'm not going to pretend this is complicated. It's not. It's just tedious, and most people skip it because they're busy and the supplier seems competent.

Here's the checklist we use before any custom CNC, molding, or stamping order:

1. Critical dimension callout sheet. Not the print — a separate document listing which 3-5 dimensions actually matter for function, and what happens if they drift. Attached to every PO.
2. Material lot requirement. If the part matters, specify lot-specific certs. Costs nothing. Prevents the "same spec, different behavior" problem.
3. Process parameter freeze. For molded and heat-treated parts, require the supplier to document their process window and notify you of any changes. Some will push back. That's fine — the ones who do are telling you something.
4. Periodic capability checks. Not every batch. But every 5th batch, request a Cpk on critical dimensions. If they can't provide it, that's your answer.
5. Torsion spring / spring-specific spec. For any spring application: require load testing at 20%, 50%, and 80% of rated deflection. Wire diameter alone tells you almost nothing about performance.

That's it. Five items. Takes 15 minutes to prepare. Has caught at least 47 potential issues in the past 18 months.

The Honest Limitation

This approach works when you're sourcing from suppliers who want to do good work and just need better communication from you. If you're dealing with a shop that's cutting corners intentionally or doesn't have the capability to hold tolerance even when properly specified, no checklist will save you.

I recommend this level of specification for anyone sourcing custom CNC components, precision stamped parts, or molded plastic parts where the application matters — pump assemblies, appliance components, anything that goes into a product with a warranty. If you're prototyping or making one-off brackets for a jig, skip it. Not worth the overhead.

The gap between "supplier made bad parts" and "I didn't tell them what I actually needed" is where all the money goes. Close the gap. Save the money. Simple.

— Give the supplier a fighting chance to succeed. That's the whole lesson.