Design that works

DFM is not making it worse to make it cheaper

Design for manufacturing has a bad reputation, and it's earned by people who use it as a euphemism. Real DFM doesn't ask you to give up a feature you need. It asks whether the way you drew that feature is the way the process wants to make it — and most of the time there's a version that keeps the function, loses nothing you care about, and costs a fraction of the original.

The parts of your design that matter are usually not the parts driving your price. That's the whole problem, and it's fixable.

get a quote. The fix is to ask early, while changing the model is still free.

The five things that set what a part costs

Every process, every material, every shop. The same five levers.

Geometry complexity. Not how the part looks — how many distinct operations it takes to produce. A shape that a machine can cut in one pass is cheap. A shape that needs the part flipped, re-fixtured, re-referenced and cut again is not.

Tolerance. How tightly a dimension is controlled. This one climbs the fastest, and it's covered in its own section below.

Material. Stock cost is the obvious part and often the smaller part. How the material behaves under the tool matters more: some alloys cut fast and clean, others work-harden, load up cutters and slow everything down. Whether the size you need is a standard stock size matters too.

Finish. Anything done after the part is made — deburring, bead blasting, polishing, powder coating, plating (that last one through a partner). Each is a handling step, and handling steps don't get cheaper with practice the way cutting does.

Quantity. Setup is paid once and spread across the run. One part carries all of it. Fifty parts each carry a slice. This is why a second unit almost never costs what the first one did, and why telling us your realistic quantity changes the answer.

Principles that hold no matter the process

Every process has features it makes cheaply and features it fights. A machined part loves flat faces and round holes. A sheet metal part loves bends on parallel axes. A printed part loves geometry that would be impossible to cut. Ask what your chosen process is naturally good at, then check whether your design is asking for that or the opposite.

Tolerance costs money at an accelerating rate. Not linearly. Each step tighter constrains the process you can use, the tooling, the number of setups, the feeds and speeds, and how the part gets verified. Those constraints stack on top of each other.

Cosmetic requirements are a separate axis from functional ones. "It has to hold 200 pounds" and "it has to look good on a desk" are two different jobs with two different cost structures, and a part can easily pass one while failing the other. Tell us which surfaces are seen and which aren't. A part that's cosmetic everywhere gets treated as cosmetic everywhere.

Setups and orientation changes cost more than cuts. Cutting metal is fast. Taking the part out, turning it, clamping it down, finding the reference again — that's the expensive part. A feature on a fifth face may cost more than the four faces before it. Large or awkward parts add to this: they often can't sit in a standard vise or chuck at all and need workholding built for them.

Standard sizes and standard fasteners beat custom. Standard stock thickness, standard tube, standard screws in standard sizes. A custom thread or an odd plate thickness introduces sourcing, lead risk and often a second operation, and it rarely buys anything the standard part wouldn't have done.

Tolerances: specify the few that matter

A tolerance is the allowable range on a dimension, not the target. And every dimension on your part already has one, whether you wrote it or not — if you didn't state it, the shop is inferring it, and inference is where expensive misunderstandings live.

The single most common and most expensive habit we see: a tight tolerance applied to every dimension because the CAD template came that way. It doesn't communicate precision. It communicates that nothing has been prioritized, which forces the whole part to be made to the tightest thing on it.

Do this instead:

• Find the dimensions that control fit or function. Usually a handful. A bore that takes a bearing. A hole pattern that has to line up with something else. A face that seats against another face.
• Tolerance those, and only those. Tightly if they need it.
• Leave everything else to a general note covering the rest of the drawing.
• Say what the part mates with. This is the most useful sentence you can write and it's almost always missing. "This bore takes a 608 bearing" tells us more than a number does, because it tells us what has to be true, not just what you calculated.

Two notes on standards, since drawings inherit them. ASME Y14.5 is the US standard for geometric dimensioning and tolerancing, and if your part has real fit requirements, GD&T states them better than a stack of plus-minus dimensions. And if your title block carries a general tolerance note citing ISO 2768, be aware the geometric part of that standard was withdrawn in 2021 and replaced — worth confirming rather than inheriting from an old template.

What we'll do with what you tell us: tolerances are reviewed per project. Ordinary work is made and verified in-house. When a drawing calls for something tight, the job goes to a partner who can supply a CMM report — which usually costs less than trying it twice. Tell us which dimensions matter and we'll tell you which path your part takes and what comes with it.

Where designs run into the process

Machined parts. Internal corners can't be sharp, because the cutter is round. A modeled sharp internal corner has to become a radius, and the radius comes from the smallest cutter that can reasonably reach it — so specifying a generous internal radius is a gift to yourself. Deep narrow pockets and deep holes lose accuracy for the same physical reason: the deeper the tool has to reach, the more it overhangs, and an overhung tool deflects. Thin walls and tall ribs push back against the tool and chatter; the fix is shorter tools and lighter cuts, which means more passes and more time.

Printed parts. A printed part is not the same strength in every direction. In material-extrusion printing the layers are the weak plane, so the build orientation decides which axis your part is weakest along. That makes orientation a design decision, not a shop-floor detail — tell us how the part gets loaded. Wall thickness is the other one: thin or unclosed walls are the leading practical reason additive orders get sent back before they are made.

Molded and cast parts. Two rules dominate. Draft: vertical walls need a taper so the part can leave the tool, and how much depends on the resin and how deep the draw is rather than one universal number. Uniform walls: thick and thin sections cool at different rates, and that difference shows up as sink marks, voids and warp. Where thickness has to change, change it gradually. Ribs are made thinner than the wall they stiffen — commonly about half to two-thirds of it — with a fillet at the base.

Sheet metal. Almost every constraint comes from the bend. Material stretches on the outside of a bend, which is why flat patterns aren't just the sum of the faces, and why a hole placed too near a bend distorts. Flanges have to be long enough for the brake to hold. Bend radius, minimum flange length and the allowance built into the flat are all governed by the tooling and the material, not by the model — so keep bends on as few axes as you can and give features room to stay away from them.

What to do about it

Ask for a design review with your quote, before you finalize.

That's the whole recommendation. Send the model, tell us what the part does, what it mates with, what it has to survive, which surfaces are seen, and how many you expect to need. We'll come back with what's driving the number and what we'd change to move it. Changing a model is free. Changing a part is not.

Uploading a file is not a commitment. It starts a review — a person confirms feasibility, pricing, capacity and timing before we commit to anything.