“Groom to grow” has been the mantra for corporations for many years. This paper describes a data-driven approach by IBM to use predictive modeling for internal talent management. In the face of rapid innovation and needed customer-centric delivery of results, knowing who is best suited in an organization to tackle certain projects is tantamount to the future competitiveness and even survival of an enterprise.

In contrast to LinkedIn’s free-form descriptions of skills, the authors advocate a semi-structured approach, which includes IBM’s expertise taxonomy, plus additional non-job-related data sources such as social media to make predictions of an employee’s proficiencies. IBM’s expertise taxonomy has five levels: “primary job category, secondary job category, job role, job role specialty, and skill” (p. 1730). For the study at hand, the authors focus on investigating the employee’s primary job role and job role specialty, plus data from IBM’s internal social network.

Based on vast aggregation of the employees’ job roles and job role specialties, predictive modeling in this investigation is cast as a classification problem. Employees are inventoried by job role and job role specialty. As a complementary endeavor, inventorying by expertise allows companies to draw on their employees’ expertise to answer questions. Expertise or skills are more granular than job role and job role specialty and would be harder to use for talent management predictions.

The paper discusses in more detail IBM’s expertise taxonomy, couches job role predictions as a supervised classification problem, describes the empirical study and various classification algorithms, and discusses final deployment of the system within IBM’s global sales force. The purpose of the research was to “develop a classification methodology to predict the expertise of employees based on features derived from [their] digital footprints” (p. 1737). Applied to the 40,000-member global sales force at IBM, the developed classification was able to complete and update job role records with a savings of one man-year plus the added benefit that the accuracy of employee records will also support business processes that depend on that very accuracy. While the return on investment will be even more impressive once the methodology is applied to the enterprise at large to complete and update job role records, there is an even greater gain in the qualitative improvement of advanced talent management when it comes to actually using the complete records in resource planning and career development.

As companies realize the value of data-driven talent management classification schemes and other predictive modeling, it will enable career planning and eventually depict career trajectories that could help formulate the employee’s options for fully realizing his or her potential within a company. This paper is a telling, well-researched piece of evidence that predictive modeling is fundamentally changing the face of talent management.