That's probably their thinking, but unless their aggregates include all of the other products that each customer who buys an item also buys (which it should, they have the data thanks to fly-buys schemes) they can't properly assess the overall impact of a given product.
As a trivial example, just looking at the profitability of individual items would lead them to discontinue selling milk (I mean, there's no profit in it so why bother?) because it's generally used as a loss leader.
That's a fair point. But I'd imagine it's a multi-step process. Step 1: group products into families, like "milk" or "ketchup". Step 2: optimize shelf space with whatever rules you have applied to product families. Within a product family, pick the N best-selling products.
The loss leader in this scenario is the "milk" family, not a particular brand of milk.
As a trivial example, just looking at the profitability of individual items would lead them to discontinue selling milk (I mean, there's no profit in it so why bother?) because it's generally used as a loss leader.