The maybe dozen NUC's I needed were not a very substantial hardware cost. You bought a Celeron NUC and they just worked even if you didn't have the most experienced staff supporting them or the best software running on them.
If you're going to be deploying a thousand such systems, sure, of course you shouldn't use NUCs.
I'd probably argue that, barring specific requirements that these might not cover, the NXP i.MX6/7/8 would probably be my default for something like this in volume. ~Motorola~/~Freescale~/NXP have been incredibly great stewards of the platform and provide significant lifetime guarantees for their parts (like 10+ years often). Plus you can generally run mainline kernels on them unless you're doing something special that requires something from their vendor branch.
I 100% agree that it’s not the technically optimal solution. But don’t discount the idea that, given an existing set of people and skills, it can definitely be the local optima organizationally.