We didn’t make it out of the blue. All assessments that Pymetrics builds as part of our approach to data collection and ingestion of data in this analytics lifecycle are drawn from cognitive neuroscience. We have researched the literature and developed a series of assessments based on this literature that we can then adjust and adapt to in the employment landscape. To do this, we need to take these assessments, which were originally used in a lab context, and transform them into data that we can use in a recruitment prediction algorithm.
We do this using a variety of ML solutions. The key is that as we build these solutions and as we build our machine learning platform, we should have a tool that predicts success but is also fair. Because it is really important in the field of employment that we focus on ensuring that everyone has equal opportunities in the context of work. This is the core machine learning part of what we do.
We have to do two things at once: We have to make predictions, and those predictions have to be fair. And it’s not just prediction that we have to focus on. It’s making sure to maximize the value of the people we recommend hiring. This is not just fairness. We also have another layer. We have to add to the explanation element.
When you make decisions about people’s lives like we do in employment, it’s really important that those decisions feel within someone’s reach. It’s not just a black box. It is not just an arbitrary decision. For a candidate applying for a job, they must understand why we make the decision we make. For a recruiter who is evaluating a candidate, he must understand where the decision comes from. These seem like features of a machine learning solution, but they are actually the key features of business value as we go through the customer lifecycle.
These three measures of predictability, fairness and explanability, we can then quantify in different ways, which we do for all of our clients. We define things like the efficiency we provide to the customer’s employment lifecycle. If they need to hire 1,000 or 10,000 people, how much time can a solution like Pymetrics save? We can define things like diversity. How much diversity are we adding to the company that can help offset some of the historical marginalization of underrepresented groups? We can also determine things like people’s performance on the job, how effective they are, and the type of candidate and client experience over time. Over the past five or six years, we’ve developed a whole set of metrics that we measure for everything. We have a truly valuable, scalable, proprietary data set that we use to drive our hiring decisions. We have a basic set of machine learning algorithms that we use to make predictions and recommend who should be assigned to a particular job. Finally, we have well-proven methods for demonstrating business value. So, this is in a nutshell what Pymetrics does.