Check out this awesome post (by Keith Harrison-Broninski)  on RQNG that focuses on improving project success rates by capturing user interactions.
From the post:

 In the end, software applications are only there to support human work. Even a low-level, highly automated software application for (say) car numberplate recognition or payroll calculation is only there to meet the needs of the police officers or HR staff who ultimately set its initial parameters and use its output.

Yet most approaches to understanding and modelling human work have a major weakness – they offer reasonable support for capturing H2S interactions (between humans and systems), but are extremely weak when it comes to capturing H2H interactions (between humans and humans). Further, mainstream modelling techniques provide little of the context required to understand what truly goes in knowledge work.

Read the full post here.