The Experiment
Can an algorithm using conversations/natural language data in Slack evaluate employee performance as well as a human?
People at companies who use Slack use it a lot. They communicate (almost all day), share work/accomplishments (actively) and consolidate third party data sources (like Google Analytics & Stripe). I’m interested in looking at ways we can mine such unstructured, natural language data to help companies understand things like: Who are your top performers? Who are your worst? Which relationships/teams are thriving? Which are breaking down?
To start, I’d get a few growing startups (50-100 employees) to give us full access to their Slacks. I’d then ask the CEO to confer with her leadership team and rank their employees from 1-100. Then, we’d try to reverse engineer an algorithm that can reproduce that order of employees using data in the company’s Slack channels & chats. I don’t imagine that would be enough to start scaling into a huge business, but it could be enough to comfort ourselves & others that an algorithm can at the very least produce the same results as high-touch humans.
Expected Duration: 8 Weeks
Ajay Rajani Dec, 09 2016 - 05:07 PM
Thanks for the comments, guys. I'm excited about it too!
Ash Gobindram Nov, 11 2016 - 02:44 PM
Great experiment idea Ajay
Christopher Chow Nov, 10 2016 - 07:22 PM
This is a wonderful ambitious idea. It would also be ironic if the ranking from your experiment was actually superior in predicting employee value over the subjective measurements from leadership that inevitably contain bias.