Well, I often tell those new to VRFs, they are (somewhat) the L3 equivalent of L2 VLANs. I.e. we can run totally different L3 domains side by side, on the same equipment, but for L3, they are logically separate.
For example, in L2, we can reuse the same MAC for different hosts in different VLANs. We "know" they are different because we append (logically or physically) a VLAN ID to the MACs.
For a VRF, we append a RD to a L3 IP prefix, so we can have the
prefix 192.168.1.0/24
in two side-by-side L3 domains, each "known" to be not the "same"
192.168.1.0/24
(i.e. each belongs to its own VRF).
I'm guessing, you might already understand all that, and RT just looks like a "redundant" RD. Well it's not, because there's only one RD per VRF prefix, but we can have multiple RTs per VRF prefix, as they are BGP extended communities. Like other communities, we can use different communities for policy purposes.
With RT, groups of communities are often used to share portions of VRF prefixes between different VRFs! I.e. we can have a composite L3 domain that's composed of portions of different VRFs.
Try reading this article and see if it help explains the "why" behind RTs.
Perhaps the closest analogy to a composite L3 domain drawing from different VRFs using RTs would be the Internet which has multiple private IP networks interacting with each other, but each of those "behind" a public IP that's NATed.
Another analogy, but further from RD/RT, would be think of RD as a VLAN tag, and RT as a second VLAN tag for Q-in-Q (and why you might use Q-in-Q when you already have VLANs). Again, this latter example is worst, because Q-in-Q tags are push/pop access, not so much a list of tags you can manipulate/examine for policy purposes (like RT BGP extended communities).
For me, it's when I learned about the forgoing, the "light bulb" went on, why there's both a RD and RT.
