A queue at the checkout does not automatically prove that another checkout is needed. The slowdown can arise from price searches, the handling of an exception, or packaging. Observing the individual steps allows you to identify which job prevents you from serving the next customer.
I propose to note duration and cause of interruptions at certain representative moments. Detection should be about the process, avoiding personal rankings built on a few episodes. The distribution of arrivals also counts: a sudden concentration requires a different response from a continuous problem.
In a hypothetical example, ordinary operations are quick but codeless products require lengthy research. Correcting the available information can have a more pertinent effect than adding a person who encounters the same obstacle at the adjacent checkout.
Subsequent verification measures the wait and checks for errors. Speeding up payment by losing accuracy can transfer the problem to corrections. The improvement should therefore be read over the entire service, including clarity for the customer and the ability of the staff to resolve exceptions.