But why do it this way? Why not use “HAVING COUNT (DISTINCT product_name) = 1” instead?  This would look like the first query. The optimizer has to do a scan to count the product names, but the table statistics in every SQL compiler I know about keeps at least the min, max, and row count for computing execution plans. Once I see the min (or max) value for that column, I know that I have found the min (or max) value for each salesman and can ignore the rest of his values.