Teaching Talent Analytics executives how to use computer simulations to complement the predictive modeling work in HR. Simulations allow you to examine multiple scenarios and examine their end states and consequences before taking action.

1. Q3 2018 19
Real-World Talent
Insights From
Computer Simulations
By Andrea Kropp

2. 20 Talent Analytics Quarterly © 2018 Gartner, Inc. and/or its affiliates. All rights reserved. CLC181576
I
t’s one of the most natural questions human
being ask about ourselves, our environments
or our organizations. Yet too often the answer
is some version of “we’ll never know because
we aren’t going to do X” or “we’ll know only
after we try X.”
Analytics professionals spend lots of time examining the past
and perhaps extrapolating those trends into the future, but
they spend very little time considering the multiple alternate
futures that could emerge from where their organization is
now. Extrapolating an organizational trend line or applying
a regression model derived from past data is a distinctly
different activity for peering into the future than creating
computer simulations where thousands of employees
interact and make decisions.
In this article, we encourage teams to add computer
simulations to their analytical toolkits and provide a way
to get started.
The Benefits of Simulation
Simulations aid HR leaders in two ways:
• They provide access to insights that would be
expensive, risky or time-consuming to obtain via real-
world experiments.
• They can uncover insights not obtainable in any other way.
For example, a change in the compensation plan for
salespeople can be expensive and carry substantial risk. It
could also take many months to determine if such a change
makes a positive difference. With simulations, the proposed
compensation change (as well as alternatives) can be
modeled numerically before being implemented.
This experiment could be run in the real world, but the
simulation reduces the time required to obtain an answer
and could prevent a poor management decision that would
be difficult to reverse.
Other times, simulations are the only path for gaining insight
into a complex process. Such is the case with emergent
phenomena, where the whole is greater than the sum of its
parts and has properties that the individual parts do not. For
example, while total sales is exactly the sum of all individual
sales made by salespeople, an organization’s total capacity for
innovation or overall culture is more than a sum of individual
contributions. Instead, culture and innovation emerge from
a combination of the environment, the individual people and
the people’s interactions with the environment and each other.
In analytical work, simulation results are used to improve
models and theories without needing to run actual
experiments (see Figure 1). For example, the HR team
may have a set of theories about leadership development,
internal mobility, organizational turnover or satisfaction with
HR service delivery. If a simulation applying those theories
I wonder what would
happen if we do X?”

3. Q3 2018 21
to a model system doesn’t match real-world results, the
team knows that they have an incomplete or inaccurate
understanding of the phenomenon and need to revisit their
thinking as well as beliefs on the topic.
Simulation Terminology
Simulation is the process of running a computer model.
The computer model itself is the set of equations and event
probabilities that define the attributes and behaviors of the
real-world system. In discussing simulation work with others,
it would be correct to say the team is building a model and
running a simulation, but it would be incorrect to say they
are building a simulation.
In talent analytics, a model system could be the entire
workforce, a specific team, all managers or any other
segment such as expatriot workers, knowledge workers
or campus hires. Ideally, the model system is as simple as
possible while capturing all the essential complexity of the
real system (see Figure 2).
Introducing Agent-Based Models (ABMs)
Agent-based models (ABMs) are computer simulations
comprised of individual agents interacting with one an-
other to study the overall system. We focus on ABMs in
this article because they are the most relevant for social
science and organizational research.
Source: CEB analysis.
Environment
Agents
Agent-Agent Interactions
Agent-Environment Interaction
Figure 2: A Model System of Agents,
Environment and Interactions
Source: CEB analysis.
Make
a Model
Real
System
Perform
Simulations
Compare and
Improve Model
Model
System
Construct
Approximate
Theories
Theoretical
Predictions
Compare and
Improve Theory
Perform
Experiments
Simulation
Results
Experimental
Results
Figure 1: Role of Model Systems and Simulations in Analytical Work

4. 22 Talent Analytics Quarterly © 2018 Gartner, Inc. and/or its affiliates. All rights reserved. CLC181576
In an ABM, each agent (person for organizational research)
is represented directly and possesses a set of character-
istics, internal states and behavior heuristics. Agents have
their own goals and act in a self-directed way to pursue
those goals. The organizational-level patterns emerge from
the agent-level dynamics.
Specifying the model system requires creating the agents
themselves (the people), the agent-to-agent interactions,
the environment and the agent-to-environment interac-
tions as shown in Figure 2. ABMs are particularly useful
to analytics teams interested in organizational dynamics
because they:
• Allow complex individual agent behavior rules — In
addition to linear behavior, agents can act nonlinearly,
follow if-then rules, couple their behavior to other
agents or behave erratically.
• Allow complex time-based effects — Agents can
have memory of their own past choices, display path
dependence and learn from the environment as the
simulation progresses.
• Explicitly incorporate the spatial and network aspect
of social process — Distance between agents (whether
physical or org chart distances) are an important
consideration when defining if and how agents interact
with one another.
• Show the path, not just the end state — The designer
obtains insight of the path taken to arrive at each
simulation run’s end state. Was the path linear and
mostly smooth or circuitous and chaotic?
• Work well with Monte Carlo methods1 to introduce
chance and randomness.
When agent interactions are heterogeneous in this way,
the simulation can generate network effects leading to
deviations from predicted aggregate behavior. By running
the simulation multiple times, the team gains insight into
whether a certain end state is inevitable (the same thing
happens each time) or whether the possible end states vary.
Agent-Based Simulation for Organizations
As with most analytical techniques, simulations are late to
arrive in HR departments compared to other disciplines.
Fortunately, this means substantial work and examples exist
to be built upon. Unfortunately, few of these are HR-spe-
cific, and it therefore requires some imagination to see the
connection between the existing work and talent topics.
That said, peer-reviewed journals such as the Journal of
Artificial Societies and Social Simulation are starting to see
more submissions that talent analytics leaders could use
directly. For example, the issue for the first quarter of 2018
featured the article, “Model of Knowledge Transfer Within
an Organization,” along with simulations on discourse pat-
terns within small groups and balancing competence and
capacity in organizational task formation.
As an introduction to the ABM technique, we focus the
remainder of this article on one famous example with an
outstanding online interactive demonstration relevant for
diversity and inclusion executives. We strongly encourage
talent analytics professionals to visit the Parable of the
Polygons site themselves as well as “The Petrie Multiplier:
Why an Attack on Sexism in Tech is NOT an Attack on Men”
blog or any of the over 300 working examples compiled
by the NetLogo user community.1,2,3
Parable of the Polygons: A Diversity and Inclusion
Example of ABMs
Parable of the Polygons is based on Nobel Prize-winning
game theorist Thomas Schelling’s 1971 paper, “Dynamic
Models of Segregation.” In this simple ABM, triangles and
squares living on a two-dimensional grid start out in an
integrated environment. Each shape is its own agent. All
the agents follow the same single behavioral rule. They will
move into a new neighborhood (an empty space on the
grid) if less than one-third of their neighbors are similar
to them.
Running the simulation from start to finish reveals that the
initially mixed society of triangles and squares becomes
more and more segregated as each shape acts to find a
place where it isn’t in a one-third minority (see Figure 3).
The shape-segregated society is an emergent characteris-
tic of the entire population stemming from the behaviors
of the individual agents. No matter how many times the
simulation is run, the end state is always highly segregated.
The simulation also clearly shows that if you start with a
segregated society no movement will occur unless the be-
havior heuristic itself is changed. This is significant because
it allows researchers to ask an infinite number of “what
if” questions, demonstrating the power of agent-based
modeling. This particular ABM allows users to adjust the
starting ratio of triangles to squares to empty spaces and
the shape’s move behavior rules to see what group-level
characteristics emerge for each scenario.
In addition to clearly teaching what an ABM is and how it
can be useful to talent analytics professionals, the Parable
of the Polygons also gives diversity and inclusion executives
a way to demonstrate to audiences that a biased organi-
zation can emerge even when the individuals within it are
not biased. Talent analytics professionals should share it
with them.

5. Q3 2018 23
Figure 3: One Run of the Simulation Using
the One-Third Rule
Source: Vi hart and Nicky Case, “Parable of the Polygons,”
https://ncase.me/polygons/.
Start
Midpoint
End
¹
Monte Carlo methods are a broad class of computational algorithms
that rely on repeated random sampling to obtain numerical results. Their
essential idea is using randomness to solve problems that might be
deterministic in principle.
1.
Vi hart and Nicky Case, “Parable of the Polygons,” https://ncase.me/
polygons/.
2.
Ian Gent, “The Petrie Multiplier: Why an Attack on Sexism in Tech is
NOT an Attack on Men,” Ian Gent’s Blog, October 2013. http://blog.ian.
gent/2013/10/the-petrie-multiplier-why-attack-on.html.
3.
“NetLogo Models Library,” NetLogo, https://ccl.northwestern.edu/
netlogo/models/.
Tools for Agent-Based Modeling
• NetLogo is free and open-source ABM environment with
commercial licenses available in the Logo programming
language. It was designed to teach children as well as
domain experts without a programming background
and comes with extensive sample model libraries in
economics, psychology and the natural sciences. Several
massive open online courses use NetLogo for demos and
assignments.
For users who already know the R or Python language, the
open-source RNetLogo and PyNetLogo packages provide
an interface to the NetLogo ABM platform.
• Multi-Agent Simulator of Neighborhoods (MASON) is
fast and portable simulation environment developed
in Java and available for free download from George
Mason University. MASON comes with 2D and 3D
visualization options built in and an extensive user
manual and set of online tutorials.
• The Recursive Porous Agent Simulation Toolkit
(Repast) is a family of free and open-source ABM
platforms with an active developer community. It
currently comes in two versions: Repast Simphony for
Java and Repast for High-Performance Computing in
C++.
• Mesa is an ABM framework for Python users. It allows
users to quickly create models, explore their results
using Python’s data analysis and visualization tools and
work exclusively in a browser via iPython notebooks if
they prefer.