RECRUITER FAQ

About pymetrics
Diversity and adverse impact
Interpreting pymetrics reports
Using pymetrics
ABOUT PYMETRICS

What is pymetrics?
pymetrics is a neuroscience game-based talent assessment platform. Using a series of
12 quick games that measure 77 cognitive and emotional traits, we help companies
hire in a more objective and efficient manner and individuals find career paths that
capitalize on their inherent unique traits.

The core pymetrics offering is a talent assessment model. Company-specific models

utilize data collected from current company employees. The company-selected
employees then play our games, and an overall model is created.

Why do these traits matter?

In the initial development phase of the games, pymetrics determined the traits and
constructs to be measured following research into the key characteristics of
successful performers across a range of occupations and industries. The constructs
were then mapped to cognitive and personality traits that are rooted in neuroscience
and psychology research. From this pool of research, the pymetrics test was created
with job success behaviors and the widest possible application of occupations in mind.
The data driven approach then helps to identify which aspects of the traits within the
games are unifying and distinctive of the organisations’ high performing population.
Anything that is not useful for prediction or not relevant to the group will not find its
way into the model.

What traits does pymetrics measure and how should they be applied? Are they the same as
competencies?
Pymetrics traits are a collection of cognitive, social, and emotional characteristics
which are inherent to an individual and descriptive of how they learn, process
information, and react to interpersonal situations involving trust, altruism, and
fairness. The traits are like discrete behavioural units that, when combined together,
can form the basis of a competency. In this way, they are more granular, discrete
behaviors than competency descriptors. They are the ‘building block’ behaviours. For
instance, a competency for ‘problem solving’ would be underpinned by the pymetrics
building block traits of ‘planning’, ‘attention’ and ‘processing’.

Surely some traits are ‘bad’ for certain roles and you’d want to remove them if you found
them?
Our traits don't necessarily measure "good" and "bad", and we have success stories
which speak to that. Here’s one scenario: Hiring a salesperson.

Many organisations hold an ideal view of their ideal salesperson: organized,

professional, focused, and generally high IQ. However, we can empirically show that
great salespeople are often in reality, not this idealized version. We find that many
salespeople have short attention spans, prefer risky decisions, and respond quickly (if
inaccurately) to questions like those that appear on IQ tests. When we find applicants
similar to current high performers we see robust returns on investment. It turns out
that some salespeople need to be energetic, to rapidly shift their attention to find
the best selling point, and can't be bogged down by their mistakes in search for a
greater reward, even though none of these traits would appear as traditionally "good".

Ultimately, what's good for one company is not necessarily good for another.

The take home message is that it's difficult to discern why someone is good at their
job. This is why taking a data driven approach can reveal more surprises and valuable
insight. It’s also why we do not pick, choose, or manipulate certain traits, as it can
lead us to bias.

What does pymetrics not assess?

There are some behaviors that we do not measure as they are not suited to
individually focused game-based assessment. For instance, team working,
collaboration, personal impact are very difficult to assess via individual gamification
exercises and therefore should form part of a later stage of the assessment process,
should these be important to role success (e.g. assessed via a face to face method
such as interview or assessment centre). We are also not designed to assess the hard
technical skills and functional experience of roles.

How does pymetrics validate the models?

Prior to utilizing models in the selection process, pymetrics ensures that all models
have an accuracy/recall rate of 65% or higher and pass our requirements for a bias
free model:
• Recall: This is the actual success rate of the model at identifying/recognising
top performing employees from a pool of unknowns.
• Estimated Accuracy: This is an estimate of how well the model is likely to
generalize to a pool of unknown candidates.
• Bias free models: According to the United States Equal Employment
Opportunity Commission (EEOC) and the ⅘’s rule, each demographic group
must have a success rate of at least 80% of the rate of the highest performing
demographic group.

Why is it critical for organizations to look beyond the traditional hiring profiles?
Many organisations understand the value of increasing the diversity of their
talent. Organisations in the top quartile for gender diversity are 21% more likely to
enjoy above-average profitability than companies in the bottom quartile. Companies
in the top quartile for ethnic diversity, meanwhile, are 33% more likely to see higher-
than-average profits than companies in the lowest quartile (McKinsey & Co).

In many markets there is also a war for talent. Due to demographic shifts and other
factors it can be very difficult to recruit talent in certain regions and within certain
occupations. Organizations need to be more open minded to candidates with
varied backgrounds and skill sets that may prove to be extremely valuable despite
being different than ‘usual’ hires.
Traditional methodologies such as CV applications tend to be much more susceptible
to unconscious bias. pymetrics helps to de-risk sourcing from wider talent pools,
distinguish potential candidates that may be better suited to ‘you’ rather than your
nearest competitor, and reduce human bias.

Using a multidimensional AI based tool you can be sure that your pymetrics model of
success is unique to you as no two models are the same. This helps you target those
who are most suited to you and give you added confidence to take a ‘risk’ on someone
with a less typical background. The AI process can represent the complexity of
behaviours amongst your high performers easily, and allow for a complex range of
individuals to fit your model (thereby improving diversity more than when using more
linear measures). It also helps you to measure traits that are relevant but very
difficult to define via other measures (e.g. attentional focus, processing style,
memory, emotional intelligence).

DIVERSITY

If the pymetrics models are built from the data of top-performers, how does pymetrics
ensure its models are not bias?
Once a model is complete, pymetrics conducts a rigorous de-biasing process before
deployment. The purpose is to ensure that the model does not unfairly select a
certain gender or ethnicity more so than another.

During the de-biasing process, pymetrics will push hundreds of thousands of users
through the model whom we know the gender and ethnicity of. If we see that the
model is unfairly selecting a certain gender or ethnicity, we can de-weight or remove
the traits that are causing this bias. Because the model is so complex, removing or
deweighting 1 or 2 traits has little effect on the overall accuracy of the model. We
actually see that accuracy/recall tend to increase after the de-biasing process - we
theorize that this is because gender and ethnicity are not true predictors of
performance but rather noise in the data.

We use the EEOC 4/5ths rule as our standard because they have the highest threshold
for what they consider to be adverse impact. The EEOC 4/5ths rule states that each
demographic group must have a success rate of at least 80% of the rate of the highest
performing demographic group.

If models are created based on the traits of current high performers, how does pymetrics
support diversity?
Clients use pymetrics to determine the cognitive, social, and emotional trait profile
for top-performers in a specific role at a specific company. We then design a model
off of that trait profile and use that model to identify candidates with a similar trait
profile. Before sharing any models with our clients, we de-bias the model ahead of
launching with candidates (see: How do we debias? What does it mean?).
Building models based on top-performers does not mean that all selected candidates
will be homogenous when using pymetrics. On the contrary, as many potentially
successful candidates are screened out because they don’t ‘look the part’. pymetrics
unlocks the collection of deep-level, meaningful characteristics that are the drivers of
success for employees in a particular role, at a particular firm. Using this approach, it
is possible to select more diverse candidates who share only the most relevant,
important indicators of success in-role while demographic features like gender, race
and SES, are left to vary naturally.

Do pymetrics debias for White, Asian, Hispanic and Black, and how does pymetrics handle
populations such as "chinese-american"?
We currently debias for white, asian, hispanic, black as outlined by the EEOC. At this
stage our data set does not include data of the appropriate sample size to support
statistical analysis of any broader populations, for example ‘Chinese-American’.
However, we are constantly expanding our data set to include additional ethnicities.

INTERPRETING PYMETRICS RESULTS

How to interpret the Recruiter Report

The purpose of the Recruiter Report is to provide qualitative context for the
pymetrics results. The Recruiter Report is broken into two sections. The top section
shows the candidate’s top three traits as it relates to the model of the role to which
they applied. These top three traits also identify individual differences between top
performers. The second section highlights the traits to probe, which helps identify
where the candidate behaved a bit differently to the top performers who went into
building the model; we provide behavioral interview questions for each candidate to
help recruiters probe these areas further.

How to apply reports to interview

Pymetrics provides basic behavioral questions under each trait to probe - the purpose
of these questions is to help you probe a candidate’s profile further in areas where
they behaved differently to your top performers.

Candidate takeaways
Each candidate receives a personalised trait report. The purpose of the trait report is
to provide feedback to the candidate after completing the games. The traits on the
trait report do not outline any kind of fit to the role the candidate has applied to,
rather the trait report outlines in general where that person falls on a spectrum
across each trait we measure. The candidate has access to their trait report within
the pymetrics platform and is able to download in pdf format.

Please note the candidate does not have access to their pymetrics recommendation
based off of the model they are being compared to, or the recruiter report.

USING PYMETRICS
How do we encourage high completion rates?
We work with recruiters to encourage high completion rates by focusing on the
candidate experience and making pymetrics more visible to the candidate throughout
the process. We also partner closely on the communication that candidates receive
(e.g. invitation and reminder emails) as well as the welcome video that candidates
see prior to starting the games.

In addition, we work with recruiters to revise the candidate workflow to ensure that
candidates complete the pymetrics assessment at the most suitable time in the
assessment process. Typically this will be upon application for direct applicants and
following phone screen for those applying for more senior, specialist or hard to fill
roles.