Program analysis can assist software development but its use islimited by a number of usability issues, including false positivesand integration issues. Previous systems have show significantprogress addressing these by using a data-driven approach totrack the 'not useful' results, and integrating the results into codereview. However, these systems were very context dependent. Inthis paper, we motivate the application of the same data-drivenprinciples and code-review based integration to the design of theMEAN system, an analysis infrastructure running on an opentools stack. We then evaluate its implementation within a localsoftware company, and present implications for the design offuture program analysis systems.