STEP models from vendors are a big plus. Not just for mechanical integration, but also just as a reasonable check that you didn't mess up the footprints.
The reason I underscored connectors is that 3D models allow to verify mated configs to confirm which pin connects to which pin on the other side. I've lost count on how many times I got that part wrong
![Banging Head |O](https://www.eevblog.com/forum/Smileys/default/bangheadonwall.gif)
, so this is a particular pain point for me. But in general, with the clear trend towards custom-built packages for power parts (like DC-DC converters and especially modules - nearly all of them have unique package and footprint designs), I also try to pick parts for which there are 3D models available. But if not, I always make one in freeCAD or OpenSCAD before making a footprint as it not only helps to confirm the footprint, but also allows DFM checks to make sure the board can actually be assembled (and reworked if neccessary).
It's very hard to "feel" the scale of things on a board while it's in eCAD on your screen and so it's very easy to pack things too tight and end up with a PCB which can not be assembled (or it's going to be super-hard and/or not very reliable). For example, on my recent FT2232H-based Xilinx FPGA programmer prototype I specified PUSB3BB2DF ESD protection diodes for USB data line, which seemed like a good idea for me because they were cheap, have super-low Cd and so are really high-speed - so I figured I will buy a big-ish batch of them and use them everywhere. Little did I think then about exactly HOW this device achieves such impressive stats! And so here I was, with a PCBs manufactured and delivered, no stencil (because I figured the board is rather simple so I can assemble it manually with a soldering iron and hot air gun), and a part that is just a bit larger than 0201 cap and yet has 3 pads, which are all underneath device! I did manage to solder it using a hot air gun, but to say it was a pain is a gross understatement.