2019 Volume 22 Issue 3

EDITORIAL: On the Transformation of HR

EDITORIAL: On the Transformation of HR

Machine Learning to the Rescue!

“This design has been instrumental in employee retention.”

Organizations, both large and small, continue to face growing challenges in employee retention and recruiting as the unemployment rate declines to record lows. While many in the punditry world are in denial, today a good employee is king. In today’s economy, employee retention is the number-one issue on many HR managers’ plate. Employee turnover is costing organizations billions of dollars each year in direct replacement costs (e.g., recruiting, training, and onboarding). These staggering sums, however, represent only some 30 percent of the total costs associated with turnover. The remaining balance can be attributed to lost work productivity and opportunity costs such having a less-experienced, less-knowledgeable replacement employee.

The healthcare industry is one sector that has been particularly hard hit by employee turnover. Domestic spending in the healthcare industry is projected to reach over 19 percent of U.S. GDP within the next five years. Over that same period, global spending on healthcare will rise to around $10 trillion, which constitutes over a five percent annual growth rate. These trends will continue to create more healthcare jobs at a time when existing positions are going unfilled. To that end, the projected demand for healthcare workers will grow at a rate three times that of the rest of the domestic economy over the next decade. In 2018 hospitals had experienced the highest turnover rate in over a decade, and since 2014, the average hospital has turned over nearly 90 percent of its workforce, truly an amazing statistic! These high turnover rates are one of the main factors driving up healthcare costs.

Both the senior leadership and HR staffs throughout business and government are well aware that it is much more cost-effective to retain a quality employee through a variety of incentives,  than to recruit and train a replacement employee of the same capabilities and that is assuming quality replacements can even be found, given the current economic environment.

In my experience, people don’t leave their organizations; they leave their managers. Maureen Swick, CEO of the American Organization of Nurse Executives

Many in the HR community suggest that a technological approach may be the only viable solution to the current retention challenges. Specifically, some HR gurus are proposing that machine learning-based algorithms are an ideal vehicle for identifying employees at “risk,” that is, those who are either considering or planning to leave. Machine learning, a subset of artificial intelligence, has already seen some success in the HR community of practice (e.g., employee recruiting). At its core, machine learning is a process for building analytical models that can detect patterns and make decisions with limited human engagement.

Typically, analytical-based models are developed from only a portion of the available data (usually 75 percent) and then tested using the remaining data. This two-step process helps ameliorate a phenomenon known as overfitting. Usually, overfitting occurs when the model matches the data used in developing the algorithm but does generalize well on new data.

Machine learning can be classified into three broad categories: supervised, semi-supervised, and unsupervised. In supervised learning, models are constructed from databases that contain both inputs and outputs, for example, determine the degree of association between certain employee characteristics and their corresponding compensation. In this application, employee characteristics are the inputs and compensation is the output. Classification and regression trees, neural nets, and Bayesian belief networks are three of the more popular machine learning algorithms in use today for supervised applications.

Semi-supervised learning involves developing models wherein a significant portion of the sample does not include outputs, for example, employee fraud detection where most of the fraud cases remain unidentified. In this scenario a predictive analytics model is generated with the dataset that contains both inputs and outputs, and then the developed model is used to produce outputs for the remaining non-assign inputs.

In unsupervised learning, the task is to analyze data that contains only inputs such as data from employee satisfaction surveys and  employees’ written descriptions of their short-and-long-term goals. The learning algorithms used in unsupervised analysis are designed to group unsorted information according to differences, similarities, and patterns,  typically in the form of clusters. Upward of 80 percent of a typical organization’s data can be characterized as unsupervised.

HR analytics in the systematic identification and quantification of the people drivers of business. Sjoerd Van Den Heuvel & Tanya Bondarouk.[1]

So, the question at hand is how can machine learning be used to help retain current employees?  Understanding why employees decide to stay at or leave lies at the core of this question. Identifying attrition risk factors can be addressed using advanced pattern recognition technologies. Perhaps somewhat surprisingly, machine learning has also been employed to identify students at risk in an academic setting. Here, early detection is essential to provide meaningful interventions. These student-at-risk algorithms can also be  applied for detecting employees at risk.

Employee data base systems are usually hierarchical in nature in the sense that they capture information from individual computer keystrokes and voice commands through to the completion of a given task or project. The complex nature of many of these factors is where machine learning can earn its keep. For example, using personal information (e.g., years of experience), performance assessments (e.g., above average), and job types (e.g., salesman)  can classify each employee as to their propensity to leave.  If the analysis was based on historical turnover data, this exercise would be classified as supervised. If historical turnover data was not available, then an unsupervised approach would work best.  Having identified which employees are at risk, the next step, which is the fun part, is to generate an appropriate response. This is like identifying specific remedial actions for students at risk in an academic setting, for example, providing specific customized content based on the student’s background and characteristics. In a HR setting specific strategies include compensation, mentoring, work environment, enhanced retirement and health benefits, and new job opportunities. The specifics of each of these strategies can be customized, once again, using machine learning.

Machine learning is not limited to identifying employees at risk and developing customized retention strategies.  Some HR applications, beyond reducing employee turnover, include predicting post-hire performance, recruiting and onboarding, and job design optimization. It is fair to say that the future of machine learning is bright and growing brighter by the day. The HR community of practice needs to adopt this technological approach to employee relations, since it will provide a comparative advantage like the use of machine learning in academe and professional sports.

Nearly ninety percent of managers believe employees leave their jobs due to compensation, but also the same percent of employees report leaving for other reasons including too little support, little opportunity for growth, and a loss of confidence in management.  Leigh Branham, CEO of Keeping the People.


[1] Van Den Heuvel, S., & Bondarouk, T. (2016). The Rise (and Fall?) of HR Analytics: The Future Application, Value, Structure, and System Support. Academy of Management Annual Meeting Proceedings2016(1), 1. https://doi-org.lib.pepperdine.edu/10.5465/AMBPP.2016.10908abstract

Print Friendly, PDF & Email
Author of the article
Owen P. Hall, Jr., PE, PhD
Owen P. Hall, Jr., PE, PhD
Owen P. Hall, Jr., PE, PhD is a former Corwin D. Denney Academic Chair and is a Professor of Decision Sciences at Pepperdine University’s Graziadio School of Business. He is a Julian Virtue Professor and a Rothschild Applied Research Fellow. Dr. Hall received the Harriet and Charles Luckman Distinguished Teaching Fellow in 1993, the Sloan-C Effective Teaching Practice Award in 2013, and the Howard A. White Teaching Excellence Award in 2009 and 2017. He is the vice-chair of the INFORMS University Analytics Programs Committee. Dr. Hall has more than 35 years of academic and industry experience in mobile learning technologies and business analytics.
More articles from 2019 Volume 22 Issue 3

Creating a Mentoring Program That Works

Growing interest in mentoring programs is largely due to the benefits mentoring provides to participants and their organizations. Mentoring can offer significant benefits in corporate settings. In this paper, we outline a process for how your organization—large or small—can create a mentoring program that meets your unique needs.

Related Articles