Company culture was identified as the most significant barrier to effectiveness. The first part of a talk offers lessons of how NLP can be used to break through first steps of solving issues such as quantifying organizational culture without surveys. I will demonstrate scoring engine that was created to help with it. Second part will cover insights that we found applying this engine.
Company culture was identified as the most significant barrier to effectiveness. Traditional survey-based psychometric approaches usually do not help team leaders to manage high-performing teams successfully or to hire the talent with the best cultural fit. However, NLP techniques applied to communication data with the help of Python-friendly libraries make it possible. The first part of a talk offers lessons of how NLP can be used to break through first steps of solving issues such as quantifying organizational culture without surveys. I will demonstrate scoring engine that was created to help with it. Second part will cover insights that we found applying this engine to different text data.
I will start with a quick real time demo of a labeling system that was created to identify pattern of behavior in teams. It is associated with a psychological dimension that has been established by C. O’Reilly (2015) at Stanford University as a foundation of organizational culture. It measures the following dimensions: adaptability, collaboration, customer-orientation, detail-orientation, result-orientation. I will demonstrate how this technology can be applied though examples based movie character communication. Then, I will briefly cover how this engine was created and why. I will explain how - instead of running surveys and long time-consuming manual labeling of texts – we created an engine that label text automatically. I will provide examples of how unsupervised methods (topic extraction and word2vec) can be useful when researchers need to modify existing knowledge from other areas to their specific research question and how these algorithms can be used to make a scoring engine richer and more sensitive to the topics, lexicon that are determined by data generating process/ context. And the last part of my talk will provide interesting insights that we have found applying our scoring engine to different data sources. Firstly, the engine was used to understand the differences in cultural behavior between different teams. After the analysis of half a million slack messages that have been generated primarily by tech startup, we found out that specific cultural dimensions are associated with certain roles in a team. In particular, we found that a significant number of teams that can be defined as detail-oriented based on O’Reilly are associated with design and front end roles. On the other hand, backend teams, while discussing releases and code reviews, contributed a lot to the terms that refer to result-orientation and integrity terms. Moreover, leadership positions are usually filled with people who demonstrate adaptable and principle-focused behavior. Secondly, we applied our scoring system on Glassdoor reviews to develop detailed cultural profiles of 335 startups based in the world’s leading startup ecosystems: Silicon Valley, New York City, London, and Berlin. The analysis included 41 unicorns, extraordinarily successful startups with $1 Billion+ valuations such as Slack, Uber and Airbnb, whose data provided key insights about what factors contribute to a startup’s success. The results of the investigation of “ecosystem cultures” shows differentiating factors of companies in Silicon Valley and New York from those in London and Berlin. Examining the unicorns, their significantly higher adaptability proofed to be the distinguishing factor that set them apart from less highly-valued companies.