Extensions are the worst part of XBRL's design; dimensions are the best part. Extensions are clunky and undermine the central purpose of a common business language. Dimensions are elegant and preserve XBRL's central purpose - comparability. So why are filers and software vendors opting for extensions when dimensions are better? (and yes, software can and should play an important role in the choices made by filers)
Here's an example, compliments of Streamline Health Solutions, Inc. (the filer) and Merrill Corporation (the software vendor). The filer has five components of revenue and of cost of revenue. But there are not corresponding standard elements in the US-GAAP taxonomy for all ten of these line items. So the filer chooses to tag four items to standard elements and create six extensions, or custom tags. It has chosen a one-dimensional data structure, the line item. And it's filing is no longer standardized. It's a hybrid of standard and non-standard elements. It's income statement is not comparable to statements from other companies without modification, i.e. mapping the extensions. Analysts interested in the gross margin by product component are forced to do the calculation manually since the revenue and cost components don't match up.
So in this case, the XBRL filing is machine-readable, but not machine-understandable. It could be, though, had the filer decided to use a second US-GAAP dimension called 'Products and Services' (or create a new dimension), and add five members to the domain corresponding to it's revenue components. This new axis can then be applied to two standard line items - Revenue and Cost of Revenue, as shown below. Member values can be rolled up to totals and the two line items can be easily analyzed along this dimension. No extensions, no mapping.
This is very simple, even trivial. There are many opportunities like this for choosing dimensions over extensions and significantly increasing the usability of XBRL data.
