For example, CAT bonds are generally tied to the specific natural hazard ("this bond triggers if a hurricane of Category 3 or higher land falls in this segment of Florida") or to industry losses, as estimated by an agreed upon source.
This means that a CAT bond is correlated with, but not directly informed by an insurer's actual loss experience. Traditional reinsurance (so an insurer themselves getting insurance) will usually be tied to specific policies, so their experienced loss is what determines payout.
However, depending on the insurer's policies, traditional reinsurance may be unavailable or much too expensive (either due to the large limit needed, the risk level of the policies, or any number of other reasons). Depending on the trigger, a CAT bond can also pay out faster because you don't have to wait to see the claims from 100k home insurance policies.
From the technical side, most large reinsurers license CAT modeling software from one or both of the same two vendors: Moody's RMS or Verisk. The biggest reinsurers will develop their own models, and there are other modeling vendors that they may license for particular perils (EQEcat for earthquake and KatRisk for flood come to mind), but the big two are pretty widely accepted in reinsurance markets.
That means if your policies are "odd" in some way (uncommon construction type, power facility, etc.), depending on how a reinsurers chooses to model them (or how the model specifically handles them) can have a big impact on your reinsurance pricing. If you know something about your policies that can't be incorporated into a vendor model very well, you may get better pricing on a CAT bond.
These are just some of the considerations! There are so many more things that go into it. But I think it's super interesting to think about.
Source: I work in this side of the industry, specifically in natural catastrophe modeling.