Surprisingly one of the areas I use my comp sci knowledge the most is in one of the most widely applied areas - relational databases. Every programmer spends lots of time reading and writing data, and understanding a lot of the low level operations, plus linear algebra, can really help with performance and assessing alternatives.
The way I think about it is that what you learn in class is valuable for understanding the abstractions that you will use in industry. If you’re working in C# or Java, most of the time you don’t care about the cost of memory allocation, method calls, reflection voodoo, file system access, etc... BUT, in those rare situations where the abstraction causes a performance or correctness issue, then all that academic knowledge becomes valuable. I find that the instances of these problems are very rare but when they occur you have an opportunity to deliver a lot of value.
For example, at my job we have a rather slow build. Looking at the logs, it’s because we spend a lot of time doing I/O. Someone had the idea to use symlinks instead of file copies. Badabing, badaboom, we got something like a 3x speed up from doing that.