2017

Using Predictive
Analytics to Reduce
Employee Turnover
SANJAY K RAJ
Why is Employee Turnover important to Organisations?

Employee Turnover is a major cost to any organization. There is a great deal of cost
associated with employee turnover isolated to not just financial costs from the impending
recruitment and training processes, but also brings with it a high degree of intangible costs
associated with the loss of knowledge capital. Building and maintaining a talented, and
contented employee base can prove to be a challenge as well as an eventual source of
success to organizations. Given the well-known direct relationship between employee
dissatisfaction and employee turnover, and happy employees and happy customers, it
becomes of utmost importance for organizations to understand drivers of employee
dissatisfaction and other contributing factors leading to employee turnover. For these
reasons, it is imperative for organizations to understand the root causes of turnover so that
it may aid them in measuring and predicting turnover, to control and prevent it within
targeted talent and knowledge levels.

Using Analytical Tools to Influence Churn Rate:

The evolution of analytical tools and technology has empowered HR professionals to be able
to influence retention rates within an organization. HR Analytics tools are a focused
extension of analytics and Human Resource Systems that aggregate not just HR data, but
also data from across the company. HR Analytics helps design, and build models that mine
and analyze data for unusual patterns helping to discover new trends and/or statistically
validate behavioral patterns. Traditional HR support systems do provide some insights into
employee turnover, however, HR analytics can point towards the root cause of such
phenomenon more accurately.

Effective Human Resource analytical system enables the organization to analyze the data
generated or collated by the HR systems key metrics to identify trends and perform
predictive modeling. These tools can also be used to help generate the standard turnover
reports that traditional HR systems churn out such as the nature of turnover, whether
voluntary separation or involuntary separation, the demographics of the employees (age,
ethnicity, tenure, skill specialization), the employee-cited reason for separation, and more
but also they are capable of generating a more comprehensive report on the identified
trends and relationships between the employee-data set and turnover.
Need for Understanding the Root Cause of Employee Turnover:

Understanding the root cause of turnover is not an easy task, as its complexity is
compounded by factors like the need for access to accurate, comprehensive data about
employees, their compensation and benefit structures, performance evaluation, training
results, and more, in a timely manner. While many organizations usually do document such
data, however:

i. The sense-making of such data often depends on analyzing the right data sets
in the right way, rather than the focus on collating and adding more
(unnecessary) data into the mix.

ii. The data is usually stored in isolated clusters that each department is
responsible for, making it harder to access them all at once for any analytical
process.

iii. Such data is stored in a format to facilitate transactional and operational

functions of the company, and not the decision support systems or a business
intelligence tool system.

Furthermore, the data itself has to be valid because they form the foundation that analytics
depends on. For example, a meta-analytic study by Rodger Griffeth and Peter Hom (2000)
on turnover found that the employee-stated reason under the construct of intent to leave
for an organization was only true 12% of the time. This casts a doubt on the validity of the
stated intentions, and their use in any analytics in the future. It also implies that there is a
need to understand the true, root cause of the employees reason to quit an organization
before venturing into the domain of predictive analytics.

Analytical Techniques used in Employee Turnover Analysis:

The commonly used methods for analyzing Employee Turnover are based on statistical
methodologies, some of them being:

1. Logistic Regression:
 This form of regression is used in scenarios that have binary outcomes, in essence, a
result of 1 or 0, which aligns perfectly in the case of employee turnover people
either quit the organization or they do not. For this reason, this regression technique
is the most used in creating predictions for a turnover.
In simple terms, logistic regression helps to understand the past by clarifying the
impact that different variables, factors, or events had on a particular outcome of
interest and provides an insight into the future by using those variables, events, or
choices into an equation that can be used to predict the probability that an outcome
of interest will occur.
A downside to logistic regression is the difficulty associated with interpreting the
output of this statistical approach. Said output is usually derived in the form of an
odds-ratio comparison of the likelihood of one groups outcome (quitting an
organization) to the likelihood of another group using the same outcome as a
measure. To convert the derived odds of outcome into a percentage likelihood of the
outcome of interest requires further steps, which can be confusing for those not well-
versed in the language of advanced statistics.

2. Survival Analysis:

This method is also known as an event history analysis and is an advanced statistical
technique used to estimate the probability of an outcome occurring over time.
Historically used in medical science to predict survival rates of patients, this
statistical technique can be applied to estimate turnover as well. Unlike logistical
regression, survival analysis can be used to generate the likelihood of turnover a
specific point in time for a specific employee. This technique can also be used to
analyze and make predictions via tools like R-studio, which helps to visualize the data
for easier sense-making when compared to the logistical regression technique.

3. Decision Tree Model:

Unlike the previous two techniques that dive into an micro-level analysis of an
employee/a small group of employees, the decision tree model begins with taking
into account every employee within an organisation, starting at the macro level, and
 then proceeds to sort them into smaller groups based on their likelihood of attrition,
creating a branched tree-like map with a central node and branches for each path.
In a particular study on the attrition analysis of SanDisk, using the decision tree
model, found:

Current Analytical Tools in the Market:

There exists a variety of business intelligence tools in the market that utilize and build upon
the above statistical techniques to help an organization predict turnover. A few of the
leading, and more commonly used analytical tools are:

i. IBM Watson
ii. IBM Kenexa
iii. Visier
iv. OrgVue
v. Talent Analytics
vi. HiQ Labs

Challenges with Employing Predictive Analytics for Employee Turnover:

1. Data Quality:
The most rudimentary problem associated with any kind of analytics is the
dependency on the quality of the data being fed into the analytical process. The
validity of outputs depends on the authenticity of the inputs, which underlines the
 importance of the quality of the foundational data.

2. Need for Satisfaction of the 3 Criteria for Analytical Performance:

i. Data must be accurate and should make statistical sense
ii. The analytical models in use must be easily understood by the Human
Resource manager in order to completely understand the drivers of employee
turnover. Any analytical model is useful, only as long as those models make
sense to the relevant stakeholders, and it is for this reason that complex,
black-box models are less preferred over simple, easy-to-understand
analytical models.
iii. The selected analytical models should be operationally efficient, meaning that
it should be easy to run these models on a periodic basis, routinely, to provide
predictive estimates before the outcome of interest can occur.

3. The Outcomes Should be Actionable:

The predictions of employee turnover models should be feasible to action on, and
only then would any predictive model can even be considered useful. An analytical
model that delivers an un-actionable outcome would render the entire process into
an exercise in futility, considering the time, resources, and efforts invested into the
analytical processes.

4. Monitoring of the Quality of the Predictive Model:

The drivers of employee turnover may change over time. Different trends may
influence the different generations very differently (millennials, baby-boomers, Gen
Z). For example, a lack of work-life balance might cause the highest proportion of
millennials in an organization to quit, as opposed to employees from other
generations. Even factors like economic shifts could influence the drivers of
employee turnover. It is therefore imperative that organizations constantly monitor
the authenticity of their predictive models, and adapt their models to relevant
changes in the external environment when required.

Exhibits
Exhibit 1.1: IBM Watson
Exhibit 1.2: Visier
Exhibit 1.3: OrgVue
Exhibit 1.4 Talent Analytics
Exhibit 1.5: HiQ
 References:
1. Analyzing Employee Turnover - Predictive Methods. Rosenow, Richard. May 9, 2016.

2. Who's Afraid of Logistic Regression? Dalen, Paul. April 8, 2016.

3. Mine Employee Data to Reduce Employee Turnover. Love, Gary, Canadian HR

Reporter. April 5, 2004.