I've practiced usability and human factors since around 1995. Last month, I added an MBA to my credentials. I've come to the conclusion that the two groups do not effectively talk to each other. Business executives want to know what the ROI is for a usability investment, usability people end up spending a lot of time developing flawed ROI numbers.
The usability people are talking to the wrong department. It is certainly the case that good usability can decrease support calls, decrease returns, increase sales, and increase efficiency, depending on the exact situation. But it's the Marketing department that has the tools to determine the value of usability (and interaction design, and user-centered design in general) to the customer.
One thing marketers and economists understand (that engineers and accountants do not) is that the cost is unimportant, as long as the price that must be charged to make a profit is somewhere less than, but near, what the customer values the product or feature. At one point in time for me, this was a truism that was irrelevant because it is unmeasurable.
It turns out that the marketing profession has developed tools to measure the unmeasurable, in dollars. My favorite, invented by psychologists, is conjoint analysis (also see Wikipedia). In this process customers can determine the relative value of different features, which can include things like size, overall design (like usability!), and brand.
The magic of conjoint analysis is revealed when you include price as one of the variables being studied. Since you can reasonably assume that the customer values $15 at approximately $15, you can see whether the improved design is valued at more, or less, than $15. You can even interpolate and get an exact value for the improved design.
Once you have a value for the improved design, you can work backwards. If your customers value your design at $25, you price the product at an extra $24, and your company requires a minimum ROI of 20%, then the investment is good if you can deliver the design at no more than $20.
Obviously this analysis needs to be done after a test program is complete, otherwise you won't have any designs to compare. But the analysis can be generalized for future decisions.