This is a tough one....
Pete, it is really tough working this out, especially if you are using the current throughput to determine the actuals versus the possible.
Here's an example. In a recent project I deployed intelligent workload distribution; effectively wrapping TASKS as if they were interactions in a contact centre and then ROUTING them to back office staff. The very simple process of presenting work items to the back office resulted in a 300%+ increase in throughput! Why? Because back office staff weren't cherry-picking jobs from a shared 'queue' in ERP/Task Management but were being given tasks without prejudice. Heck, we even found 1 of the 100 or so back office staff had been unable to handle any tasks for weeks because their login to the ERP wasn't working, yet they sat there, getting paid; whilst actually doing nothing.
So, with that cautionary tale out the way try to look at what is current based on actual throughput of a representative user. Not someone who is brilliant, and not someone who is brand new and unfamiliar with it. Use them by monitoring closely for a few days to get your 'actuals'. Shoulder surfing back office tasks often surfaces a whole range of other activities that these users get involved in that are not encapsulated by looking at the raw metrics. Remember, this is always where PEOPLE, PROCESS and TECHNOLOGY overlap.