Providers can take advantage of predictive analytics to improve common staffing challenges.
Irving Stackpole and Tim O’Rourke
Predictive analytics can be used to improve staff recruitment and retention in many settings, including seniors housing, assisted living, and skilled nursing. Operational and human resources experts have been studying these issues in the industry for decades, yet the track record of improving recruitment or retention is disappointing.
The Costs of Turnover
Almost no matter how it is viewed, the scale of the turnover problem in seniors housing and long term care is enormous and getting worse. The Institute for the Future of Aging Services said that turnover in the sector costs $4.1 billion, adding, The most commonly used, conservative rule-of-thumb for estimating the per worker cost of turnover in the overall U.S. economy puts the comprehensive cost of replacing a lost employee at 25 percent of his or her annual compensation amount.”
Applying this rule, the Employment Policy Foundation (December 2002) calculates that “[f]or the typical full-time employee who earns $38,481 and receives $50,025 in total compensation, the total cost of turnover would amount to $12,506 per employee.” The 25 percent rule-of-thumb applied to U.S. Bureau of Labor Statistics estimates of the annual wages of direct-care workers suggests a total cost of turnover per employee in the range of $4,200 to $5,200. This comes from a 2004 report from the Institute for the Future of Aging Servies titled, “The Cost of Frontline Turnover in Long-Term Care.”
Adjusting for inflation, this rule of thumb places the per-person replacement costs in 2018 at $5,739 and the national total at over $5.0 billion. Using this conservative model, at a 70-unit assisted living residence, with a turnover rate at 45 percent among frontline staff, the annual costs are conservatively $190,000. This center needs $1.05 million in revenue to cover that cost.* And at a 120-bed skilled nursing center, the frontline turnover costs are easily $335,000 per year.**
W. Edwards Deming, the patriarch of quality improvement, famously poked fun at massive, end of the line corrections as wasteful and inefficient. Yet this has been the sector’s response to the human resource hemorrhaging that turnover represents. Investments in training, culture, communications, and a host of other improvements have often been constructive and are occasionally effective. Yet these address the outcomes of the problem, rather than the root causes.
The classical recruitment process persists: the usual and customary perusal of a resume, or job application, the brief in-person interview which focuses on “fit” (personal and cultural bias), and the unrelenting pressure to make a decision. There’s little focus on so-called soft skills, such as leadership, humility, effective collaboration, adaptability, and willingness to learn and relearn, which are emerging as far more predictive of successful employment.
Churn—the rate of turnover—persists, and residential long term care providers continue to experience expensive and sometimes crippling turnover.
Stemming the Churn
Predictive analytics describes how to use large amounts of data from a variety of sources with statistical tools in order to make predictions. “Past results are no guarantee of future performance,” is a familiar caveat. Yet, in many ways, with large amounts of historical data, the future can be predicted. (How else would Amazon and Netflix know what people are interested in?)
Similarly, with contemporary statistical tools and methods, recruitment and retention effectiveness can be predicted, and improved. When the problem is properly scoped and appropriate resources are allocated, durable and desired retention can be improved by 25 percent or more from baseline.
This represents direct savings (cost avoidance) of over $47,500 per year for the assisted living residence, and over $83,500 for the SNF. These are the savings using the conservative model, and they accrue (i.e., the savings are permanent and enduring), falling directly to the bottom-line, year over year.
Applications of Predictive Analytics
This capacity for predictions based on historical information has significant implications in many facets of residential long term care segments. Predictive analytics in the sector has been described most often as a tool to predict clinical, psychological or behavioral needs among residents.
Sometimes referred to as “ambient assisted living,” collected sensors, monitors, and software help anticipate when residents are at risk so that additional observation or interventions can be taken. While both “analytic” and “predictive,” the growing practice of using these technologies to enable residents to live safer and healthier in their environments is not the benchmark use of “predictive analytics,” nor how the term is being used here.
Predictive analytics has enormous potential as a tool to predict and improve the recruitment and retention dynamics within the seniors housing and long term care professions.
There are two aspects to churn: recruitment and retention. These are distinct challenges; the reasons people accept or decline frontline jobs in residential long term care are different from the reasons they leave or stay in these jobs.
In many communities, employee attrition among frontline workers’ positions is highest during the first six months. This recent-hire attrition is expensive and inefficient for the provider, disheartening for the rest of the team, disruptive for management, and potentially disturbing for the residents and their families. Because of the scale, the direct and indirect costs of this recent-hire attrition is perhaps the most significant aspect of turnover in the sector.
The most sophisticated recruitment systems rely on careful profiling of the prospective staff members. All employers collect rudimentary data such as age, education, previous employment history, and required background checks, as well as information such as recommendations. When this information is accumulated and reviewed, the patterns that emerge are helpful for guiding recruitment.
If, for example, the best current employees are those who previously lived in a particular town, attended one of four schools, and were active at one of six churches or civic groups, these “correlates” not only signal that similar prospects are likely to be good employees, but also that these settings are opportunities for recruitment efforts.
In addition, many communities attempting to improve recruitment conduct pre-employment psychological profiling, according to a 2017 Walden University dissertation titled, “Employee Turnover in the Long-Term Care Industry”. And there are also best-in-class interviewing methods that help identify best-fit candidates.
From Insights to Strategic Advantage
Most insights about recruitment gained by combing through existing, available information about prospective employees stays with the person responsible for recruitment (recruiter or manager), and never becomes a tool or tools that can be used by other representatives in that same community or among other, affiliated centers. Individual recruiters may benefit from improved results, but these lessons stay with those recruiters, and don’t become a competitive advantage for the community at large. And the use of these tools differs far too greatly place-to-place, and within regions.
Predictive analytics converts insights into competitive advantage.
Where to Start
The foundation and the starting point for any successful application of predictive analytic tools is data – and lots of it! Here, the loosely used term “big data” and artificial intelligence are frequently used (and mis-used). While the amount or quantity of data is important, equally and perhaps more important is its quality and likelihood that the data will contribute to a relevant behavioral prediction.
The need for careful consideration about, and experience in the sector is therefore pivotal. Start with what will be “signal-rich” information, rather than randomly collecting and then attempting to aggregate data. This is where the experience and input of seasoned professionals are essential to good outcomes. The investment of time, money and effort to develop effective predictive models demands significant collaboration, especially in assembling the data sets and definitions needed.
Data” here means any facts and that can be converted into a computational symbol. The classic examples in seniors housing in long term care are age, prior employment history and recommendations.
Very recently, software tools have been developed that can convert other, far more extensive and varied information into data. Words, for example, recorded in notes from prior interviews can be converted into data using natural language processing (NLP) tools. NLP is a broad class of tools that includes speech to text, part of speech tagging, bag-of-words, and sentiment analysis.
These tools allow for data transformations and interpolations. If possible, data from other sources can be incorporated such as data aggregators and primary research. So the previous limitations of what could be used for computational analysis have been eclipsed by new software packages. This means that many more types of information can be considered as actual, or potential “data.”
Included in this analysis are the culture and characteristics of the place of employment. The “where” becomes as important and the “who,” with data suggesting that even within the same company a good hire at one location may not succeed at another, even for the same job title.
Building the Models
Information from the first contact with prospective recruits, all the way through employment and beyond is assembled and structured through a data dictionary into a large database. Statistical tools are then applied in order to identify the corollary and predictive patterns within the data. Decision Trees, Random Forest, Stochastic Gradient Descent Classifier are examples of the models used to establish predictive algorithms.
As with any good scientific approach, the processes are empirical through repetition, which begins to identify the combinations of factors that could be predictive. Tests of assumptions must be applied by sampling current employees within the community and others outside in the larger “community” to validate some of the insights produced by the earliest analysis.
The result of this significant undertaking is seldom merely an “answer.” The far more likely outcome is a new way of looking at prospective, current and previous employees.
*Assumes 98 frontline staff and 18 percent Net Operating Income
**Assumes 129 frontline staff