Have you been avoiding predictive analytics? If so, does one or more of these statements sounds familiar?
Excuses for putting off your organization’s embrace of predictive analytics seem endless, but did you know that applying predictive analytics enables you to upgrade workforce planning (WFP) to the more-impactful strategic workforce planning (SWP)? Engaging in SWP allows organizations to forecast more accurately and to plan more effectively and completely for varying scenarios, which, in turn, helps them ensure they have the right workforce with the right skills at the right time to meet a new or evolved mission.
The first step toward leveraging predictive analytics is understanding what that is. Predictive analytics involves using various statistical techniques to, first, identify trends based on current and historical facts and, second, to assign probabilities to specific events. Simply stated, doing predictive analytics allows an organization to estimate the likelihood of future states. Doing that helps decision makers anticipate potential gaps in the workforce and other issues of supply and demand.
For example, applying the tools of predictive analytics to the last five years of turnover data would let HR professionals project (i.e., estimate) with some degree of confidence (i.e., likelihood) how many people will join and leave the organization over the next three years. A deeper analysis of the key traits of people who left the organization, such as education, tenure, location and job type, can be used to estimate the likelihood of separation for current employees.
The next step in leveraging predictive analytics is understanding what it is not. Findings are no guarantees that likely events will indeed occur. In fact, all results from a predictive analysis must be interpreted with the expectation that changing a single variable (e.g., workforce skills, staff count, personnel mix) would void the current prediction. This caveat must be kept in mind even when one wants to assume that “all else has been held equal.”
In many cases, the point of using predictive analytics is to target change efforts to achieve a desired future state. When that is the goal, results of predictive analyses must be treated as moving targets, and the analyses should be rerun regularly. It is also essential to remember that predictive analytics indicate likelihood. Complement analytics with qualitative information such as stakeholder interviews to understand potential future states and needs of the organization. For example, the data may show a growing loss of sales professionals. But business owners have been intentionally shrinking the sales staff since software now streamlines the sales function. Therefore, the smaller salesforce is intentional and not necessarily a concern that needs action.
Once you have a baseline understanding of predictive analytics, you can start to identify ways to apply them in your organization’s WFP strategies. Here are some considerations to take when doing that.
Above all, do not be afraid of statistics. Rather, do your research and use statistics strategically. If you have the opportunity, befriend a data scientist or stats colleague to get recommendations on how to use predictive analytics for WFP. If no one in your office has the requisite expertise, consider joining a networking group, reading relevant articles or taking a basic course to build some knowledge.
You do not need to know fancy statistics to gain meaningful insight from predictive analytics. In many cases, you can use basic historical trends to identify potential future states. Time series analysis and simple linear regressions are great starting points. Standard software like Microsoft Excel can be used to perform many of these types of analyses, and several quick tutorials on how to start are just an internet search away.
Data are never perfect. This is especially true for data on human behavior. It is important, therefore, to look at the predictive power (R-squared value) of your models and to remember that the product from each model is only an estimate of what may happen.
R-squared is a statistical measure of how close data are to the fitted model (or, as statisticians would say, “regression line”). Values for R-squared range from 0 to 1, and larger values generally indicate that models have greater predictive ability. If you have a weak pattern in your trends, you will usually get a low R-squared value. That means it is time to rely on a more-robust statistical model such as multiple regression or relative weights analysis.
Also remember that any predictive analysis remains accurate if none of the variables plugged into the statistical model change going forward. The model also cannot predict the influence of a variable no one plugs into the model.
Say, for example, a simple trend analysis predicts that the sales workforce will decrease 35 percent over the next three years. That would be alarming. However, the R-squared value is very low, indicating a low chance of occurrence. While you still may want to consider the risk of this projection, you would do well to run other statistical analyses. A multiple regression that employs several demographic variables will probably produce a more-accurate prediction of salesperson attrition. Experimenting with variables can lead to greater confidence in the projections.
Right now, big data is all the rave. We have more systems, data points and variables than ever before. However, large amounts of data can generate what statisticians call “noise,” or false or overreported correlations, that make it difficult to identify meaningful connections.
Choose the most-relevant data points when doing analyses. For example, you could find that height and weight have a strong predictive relationship with performance, but that does not necessarily mean you should start hiring based on those physical characteristics.
Predictive analytics provide the most valuable insights when you combine them with qualitative research like interviews with key stakeholders and staff. In fact, results s from your predictive model can be used to frame specific questions to obtain more-targeted feedback. In all events, questions worth asking when doing SWP include
These qualitative insights are critical to the WFP process and should be integrated with predictive analysis of hard data to optimize SWP.
Lillian Thomas is an analytics manager with the National Institutes of Health. She has more than 10 years of professional experience in human capital consulting, survey design, workforce planning, statistics and data analysis. Thomas can be reached at email@example.com.