If you’ve applied to a big company in the last few years, you’ve probably had the experience of assessment tests, sometimes “factual” number tests and sometimes – and this is where the company from this blog article comes in – random appearing games that are subsequently supposed to map your personality and soft skills. Pymetrics is a company that uses Big Data and Machine Learning to create an individual candidate profile and predict the fit to the respective company, but step by step:
Pymetrics was founded in 2011 in New York, NY, USA and has a current funding of approximately $57m. The company’s mission is, “To help people and companies unleash their true potential today and tomorrow.” To do this, they rely on algorithm-based hiring and talent management. The company was recently acquired by harver, a recruiting software company. Pyremtics aims to make the hiring process more objective and efficient, using their behavioral science assessments to evaluate the soft skills of potential candidates. Customers include BCG, Kraft Heinz and McDonalds.
Pymetrics advertises with pretty impressive numbers: Employees stay longer with the company (+198%) and the duration of the application process is reduced by around 60%. How exactly to classify the other two numbers is questionable, do people who perform well in these games have better sales skills? And how can the female representation be increased when the main USP of the software is actually to reduce bias? These remain open questions, feel free to post your opinion in the comments, do you think bias is completely reduced by these games?
The founder story
Frida Polli, who studied at Harvard herself, was fed up with subjective, inefficient and lengthy recruiting processes, most of which failed to recognize the person behind the CV. Through her studies in neuroscience, she knew the theory and the academically researched experiments on personality analysis. As Spotify and Netflix became increasingly popular, she realized that matching algorithms not only made sense in these areas. Netflix, after all, chooses recommendations not based on description, but on data analysis of user behavior and the corresponding media. Polli therefore wanted to bring a similar matching algorithm to the human resource field.
By extending the applicant funnel and incorporating predictive data, pyrementics can contribute to improvements in efficiency, performance, diversity and fundamental predictiveness. Furthermore, Polli says that it is easier to remove bias in AI than human bias.
How it works
Pyrmetrics relies on gamification and behavioral science, more specifically these are 12 different games, these range from bursting balloons to money exchange games to quickly confirming the space bar for time. In total it takes 25 minutes to complete and achieves a 98% completion rate across all participants. Each of the games analyzes a variety of data points based on factors such as Learning, Attention, Effort, Decision Making, Risk Tolerance, Focus, Fairness, Emotion and Generosity, developing a unique and “holistic” candidate profile. The tasks themselves come from the research area and were not developed by pymetrics itself.
Since the games just mentioned seem very “random” to most participants and it is not recognizable how and which factors are tested, the “social desirability” or the “impression management”, i.e. the manipulation of the answers by the candidate because this would seem to be better received by the recruiters, can be counteracted. The created profile is also visible for the candidate, as well as suggestions for other positions that might suit him/her. On the company side, the analyzed factors are exactly aligned with their “company DNA”. This results when more than five employees have played the games.
Pymetric’s success strategy is based on the following three pillars:
1. “Better data”, through evenly-distributed soft-skills data
2. only release algorithms when accurate and unbiased
3. independent audit
I think I can definitely agree with the first point, the more data there is about the candidate, the more informed you can make a hiring decision. However, I believe that even this data can be biased due to stress, technical inconsistencies, environmental disturbances, etc. when performing these tests. One could also argue that the interpersonal, what a human counterpart feels at the first moment is not really seen, but how objective is this human impression? I have asked around in my circle of friends and the opinions are quite different, some find it good that there is an objective tool, which is more fun through games than any posed sentences to the personality to classify on a 5 point likert scale. The others, however, complain about frustration when they are rejected because of these games at a company and finally never talked to a human counterpart of the company…. So how measurable are soft skills like empathy or sympathy? Feel free to write your opinion in the comments!
It is hard to say if assessing traits and avoiding bias is really accurate. The underlying research is not comprehensive enough to make the games applicable to all specific situations. Also, AI can only be as good as the data it is fed, how can Pymetrics be sure that it is fed with the right data and that it is not also based on old historical biases?
Another hurdle is the regulation on automated hiring tools, so software like pymetrics has to perform regular audits. Which regulation will be published in the future in different countries remains to be seen. Finally, Pymtrics advertises that it brings fun for the candidates through the games. What do you think? I feel like the keknobel for the perfect score will never stop… is there an optimal typing speed for pressing a space bar over a minute? and to what extent does popping balloons say anything about my risk tolerance?