Such a formula has so many caveats that it is not terribly useful in revenue estimation, except perhaps to provide a range within an order of magnitude. For example, Starbucks has $60,000 in revenue per employee, while DirecTV has $1.6mm. Surely no formula that would make predictions for household brands that would make errors anywhere from +66% to -94% is useful.

The main reasons for the variance is (1) how much value add the company generates for its customers, and (2) what portion of that value add is labor.

Some companies' products are very labor intensive, but they outsource most of that labor, so the company's product has high revenues per employee because there are "invisible" employees not on the payroll due to outsourcing arrangements. This is where the company is providing low value-add. For example, consider a trucking broker, who provides transportation services to its customers, but outsources the actual trucking to owner-operators who are not employees.

Other companies have high value-add, but very little of it is labor. For example, a television cable company has high costs for licensing media content, which is a big part of the value it provides for consumers, but that licensing content requires low levels of labor (unless they vertically integrate).

Companies that have high value-add, mostly in labor, tend to have moderate levels of revenue per employee. A consulting company is an example of this. It requires a lot of labor, but almost the entirety of the value add is that labor.