Pymetrics – Using Neuroscience & AI to change the age-old hiring process

Pymetrics leverages decades of neuroscience research and machine learning to match quality job candidates with the right companies and careers. It does this through a gamified process that provides companies with data on behavioral traits of candidates applying for the role.

It was co-founded in 2013 by Frida Polli and Julie Yoo from HBS and MIT. They were frustrated with the subjective, inefficient and biased nature of the job recruitment process. She wondered about how in the day and age of Netflix, Spotify and Amazon — platforms that take in information about you and give you personalized recommendations that seem to know you better than you know yourself – why was there no equivalent platform to find jobs? [1]

How Pymetrics is using AI and Neuroscience to change the hiring process?

  1. Gamified Neuroscience –

Pymetrics developed games based on years of well-established neuroscience research. They have a set of 12 neuroscience mini-games that take less than half an hour to measure 90 cognitive, social and emotional traits of candidates. While many traits are said to be acquired while on the job, Pymetric focuses on measuring the intrinsic traits that do not change over time.

Some of the games are filling animated balloons with water without them bursting, clicking the space bar every time a green dot appears and weighing how much money to trade with an imaginary partner ina scenario akin to the prisoner’s dilemma. There are no victories in this game, but rather they serve to measure the candidate’s various behavioral traits to help map them to the best-fit job.

Pymetric Game #1

When the multi-trait games are over, candidates receive a report which will let them know pymetrics’ evaluation of 90 different traits including –  attention duration, processing consistency, flexibility, creativity, decision making, learning from mistakes and more.

Pymetric Game #2

  1. Custom AI Trained Model on Top Performers –

Pymetrics makes custom algorithms for companies by running their mini-games on at least 50 of the organization’s top performers. It then uses this model to compare and find applicants with similar traits. Job seekers play different games when applying for a job and a matching algorithm is used to select the one which would be the best fit for the role or have similar skills as the top performers at the company. This model has been mostly employed by companies to recruit for standard entry and midlevel corporate positions.

  1. Ethical AI: De-Biased Algorithms –

Resume reviews lead to women and minorities being at a 50-60% disadvantage. [2]. Pymetrics works to create a more ethical AI-enable de-biased algorithm. The games played by the candidates are conducted in the form of a blind audition for job candidates. Candidates move through the platform anonymously, and the prediction algorithm does not use any demographic information to assess career fit. With AI that doesn’t see race or sex, underlying skills specific to the job can shine through. Before any algorithm is deployed, pymetrics checks each algorithm and removes any bias through their open-sourced algorithm auditing tool, Audit-AI.

Employing statistical methods to actively de-bias the dataset and validate the method, on which the predictive models lie, is imperative to ensure the selection procedure is promoting fairness rather than perpetuating barriers. Only then is the final result of a bias-free prediction model that recommends future best performers.

  1. Saves resources spent on recruiting –

Recruiters typically spend an average of six seconds on a resume; often arbitrarily cutting many candidates out of this phase. Pymetrics can help companies to screen candidates in a more systematic way based on matching skills required for the job using their games, rather than traditional CV, cover letters and self-reported questionnaires. Recruiters can spend time on outreach to improve their candidate pool, rather than haphazard resume-scanning.

It uses a software-as-a-service model, and charges based on the number of applicants a company receives each year.

Results of these shifts:

Pymetrics has managed to become a part of the hiring process for many high-profile companies like Unilever, LinkedIn, and Accenture to name a few. Polli says that some companies have more than doubled the percentage of candidates they hire out of those they invite for in-person interviews. She also noted that the platform also has helped companies increase their diversity. She says Pymetric’s algorithms constantly test for and remove ethnic or gender biases that arise, leading to more women and minority hires. It also helps companies expand their scope beyond just those who can afford expensive college educations. [3]

Challenges and future:

Pymetric has adopted an interesting model of using games to collect large sets of data about candidates. Despite, it having a large set of high-profile clients there are some challenges which it might face in the future:

  1. Most companies would still have an interview process after the shortlist using pymetric where human judgment would be used, cause biases to play out. Polli agrees with this point – “We offer another data point that’s free of bias and subjectivity, and hopefully people will trust it as an objective data point.” [4]

  2. The algorithm developed would need to be job-specific for companies to hire for a specific role. Where do you keep getting data sets of high-performance candidates?

  3. It is difficult to have a de-biased dataset. Datasets usually replicate a company’s existing setup and they have some form of bias.

  4. Would the usage of pymetrics cause the diversity of ideas to decrease due to the skill set and personality traits of employees hired being similar?

  5. People might start gaming the system, by developing, playing and practicing such games to get a higher score on the traits required for the job role they want.

Pymetrics has $17 million in funding from VC firms including Khosla Ventures and Jazz Venture Partners. The startup has about 70 employees based in New York, London, and Singapore. [5]

They have decided to tackle a problem where there is friction but only time will tell if they will be able to deliver all that they promise.



[2] Ibis







Hopper: Using AI to Plan Your Next Vacation


Upstart: Using machine learning to transform the personal loan experience

Student comments on Pymetrics – Using Neuroscience & AI to change the age-old hiring process

  1. Very fascinating – thanks for sharing! The interview process is an imperfect screening tool that hasn’t been thoroughly tested for effectiveness in hiring decisions. Two questions popped into my mind while reading through the article: 1. If there is a limited set of games that are available, how can Pymetrics prevent users “gaming” the system by practicing those same games elsewhere? 2. The set of games that you described seem to have optimal solutions (e.g. there is probably a “right” or “better” way to play the prisoner’s dilemma). Why does the company need to gather data about high-performers within a company as a first-step? As an extreme example, an HBS MBA grad might perform very well on the games – wouldn’t they be a good fit for most companies? Or would they appear to be a poor fit for some companies that just happens to have lower-quality employees?

  2. Thanks for the interesting piece! As you mentioned, there is a definitely a space for this type of technology to cut through the fluff of some recruiting processes. There are also, of course, some concerns. From your piece, I learned that pymetrics puts together the games by having current top performers play the games then training the pymetrics for the company. I wonder if pymetrics takes into account the importance of having variance in hired employees. With some smaller companies, their top performers may be very similar in how they think and work. Training models to identify candidates based off of how they match with top performers could exclude innovative thinkers.

  3. Great post – thanks for sharing! I think you represented the strengths and weaknesses of this technology comprehensively. I had the same reaction as Kemi and Tommy… this is well intended and has the potential to make hiring managers aware of some of their own biases, but I probably would not go so far as to market the tool as a “de-biased” algorithm. Particularly when it is trained to identify future leaders with similar personality types and instincts as past leaders. There is still a lot of human judgment involved when this system is used for hiring, and I think there is a risk that we actually automate the status quo, rather than boosting diversity.

  4. Thanks for sharing this really interesting AI start-up, Riddhi! I have similar concerns and questions as Kate, Kemi and Tommy pointed out in their comments. I think the idea is very creative, and definitely has significant potential to help companies have a less biased, less haphazard and more effective recruiting process. But they really need to gather a large enough pool of data to make the algorithm more accurate and precise. Plus, there’s a lot of work that needs to be done on customization for the client companies by industry, company size, stage of company development, geography and other things the recruiting companies want to factor in. And to achieve such high level of customization, I believe the upfront investment would be significant.

  5. I like the fact that Pymetrics focuses on measuring intrinsic traits, as I feel like these are unlikely to be evaluated in an interview process even though they are very important. Since Pymetrics focuses on traits that were likely to be taken into account (since they couldn’t be measured), I wonder how seriously firms will take this evaluation. What weight will they assign to experience versus intrinsic traits?

  6. I recently had an interview with Linkedin, and I was tested using Pymetrics. It was the very first time I had ever taken this sort of interview and I loved it! I thought it was much more engaging than the typical surveys that make you pick specific choices, and it felt more natural. I’m curious to see how the company connects the tests for “honesty” or “equity” with the metrics that I provide them, because the results do not seem to be based on a typical numerical basis.

Leave a comment