Giving voice to office customers: Best practices in how office handles verbatim text feedback

Microsoft Office users submit hundreds of thousands of pieces of verbatim feedback per month. How can an engineer or manager in Office find the signal in this data to make business decisions? This paper presents an overview of the Office Customer Voice (OCV) system. OCV combines classification, on-demand clustering and other machine learning techniques with a rich web UI to solve this problem. In this paper, we describe the different types of feedback received. Next, we outline the architecture used to build OCV. We then detail the text processing, classification and clustering done to reason on the data. Finally, we present challenges, future plans, and best practices that may be relevant to other teams analyzing customer feedback. We argue that this multi-pronged approach to handling customer feedback presents a pattern that other organizations can use to mature their handling of customer feedback.