You scanned a part, you have an STL, and your manufacturer wants a STEP file. A converter turns one into the other in about ten seconds, so it looks like the problem is solved. Then the shop opens the file and tells you they can't work with it.
The conversion succeeded. The file is the wrong kind of file. Here is what is actually going on, and what to do about it.
A 3D scanner does not measure a shape. It measures points on a surface, hundreds of thousands or millions of them, and software stitches those points into a mesh: a shell of tiny triangles that approximates the object. An STL is that shell.
CAD works differently. A STEP file from a CAD system describes a part as geometry with intent behind it: this face is a plane, this is a cylinder of 12.7 mm diameter, this edge has a 2 mm fillet. The software knows what the features are, which means a person can change them.
A mesh has none of that. It knows there are triangles at certain coordinates. It does not know that a region is a hole, or that two faces were meant to be parallel.
Converters do work, and for some jobs they are the right answer. What they produce is a solid body wrapping the same triangles you started with. The file extension changes and the geometry is now technically solid, so it will open in CAD software.
What you get is sometimes called a dumb solid, and the limitations show up fast:
For a one-off print, or a visual reference, or checking clearance in an assembly, that is often fine. For manufacturing a part, it usually is not.
Reverse engineering uses the scan as a reference and rebuilds the part as real CAD on top of it. An engineer identifies the actual features, a boss here, a bore there, a fillet along that edge, and models them as those features with real dimensions.
The output is a STEP file that behaves like one built from scratch: editable, dimensioned, reasonably sized, and free of scan noise. A shop can quote from it, a mold maker can tool from it, and you can change a dimension next year without starting over.
It takes longer than ten seconds. That is the trade.
Auto-conversion is usually enough when the part is going to be 3D printed as-is, when you need a visual or spatial reference rather than a manufacturing input, when the geometry is organic enough that no one would ever edit a dimension anyway, or when you only need to check whether something fits.
You need reverse engineering when a machine shop, mold maker, or casting house is going to make the part, when anything about the design has to change, when tolerances matter to how it functions, or when the part has to mate with something else.
The short test: if a person needs to open the file and change a number, a converter will not get you there.
Manufacturable is not a compliment, it is a checklist. A file a shop can actually work from has clean feature geometry rather than triangle soup, dimensions and tolerances on the surfaces that matter, a form the intended process can produce (draft angles for molding, accessible features for machining), a neutral format like STEP that opens everywhere, and often a drawing alongside it so a quote is possible without a phone call.
A fishing tackle company came to us through Protolabs with a hand-finished lure and no CAD file at all. Protolabs needed a clean 3D model before they could produce it, and no converter was going to help, because the part had organic curves and a fine surface texture with real design requirements behind them.
We scanned the part and reverse engineered it into editable CAD. We also said up front that one textured region might not model cleanly, and delivered ten business days after the part reached us.
Thirteen months later he emailed to say he had built a business on the file, and that he is manufacturing at volume through Protolabs off a lightly modified version of that same model. The full story is in our case study.
Send the mesh, tell us what the part has to do, and say where it is going to be made. Those three things determine how much reverse engineering the job actually needs, and sometimes the answer is less than people expect.
If you do not have scan data yet, short-range scanning starts at $260 and includes the reverse engineering that turns the scan into a usable model. You get the price and the timeline in writing before anything starts.