This is a mistake. Disk space is cheap. A raw image is only about twice the size of a good jpeg image. For my Nikon D7100 the comparison is about 30MB vs 15 MB.
Here's what is going to happen: Marketing is going to get a Jpeg. They are going to say, "that sky isn't blue enough. Let's saturate it more" And they edit it. And Lo and Behold because you're mapping 8 bits of information into 8 bits of information there are rounding steps, and the sky is banded, or becomes mottled. And that expensive model's flawless skin now is pixelated at looks like it's made from coarse sandpaper.
Back into photoshop. Mask the sky. Introduce noise into the saturation channel. Now increase saturation. Ok, it worked this time. But it took 15 minutes of an expensive person's time. (Good photoshop techs don't come cheap.) Or worse, they just blur the sky. No bands, but it loses something. Cloud edges don't pop anymore.
Never throw information away.
46,000 images at 30 MB each would be 1.38 TB. Buy a pair of enterprise quality 2 TB drives, and mirror them. You're set up for a few years.
A larger problem is keeping the versioning in sync. The JPeg image should show up in your system as being a derived image from the Raw master, and keywords applied to the master should propagate to the JPeg. Whether you can do this is a function of the DAM software you got.
Tips: You need unique IDs for images in the system.
Look at using exiftool and using metadata to rename images. I would suggest naming them
This guarantees you a unique number even if you are a local newspaper with 11 Canon cameras on staff. Note: Use a naming scheme that does not include spaces or characters that have special meaning to various operating systems. Avoid /@3&<>!?* at least.
Note that this fails big time with scanned images. Scanned images in DAMs are a difficult proposition. You need to run a salvage operation for metadata.
Your dam should be set to write this into any image on export as a keyword. That way 2 years from now, when the Marketing department says, "We need a 3000 pixel version of this image for a billboard instead of the 256 pixel version used or our mobile website, you can actually find it. (Yes this happens. I'm doing it now for my website. For 2000 images.)