Due to the rapidly changing, dynamic nature of today's economic landscape, organizations are often engaged in a continuous exercise of matching their workforce with the changing needs of the marketplace. Re-skilling offers these enterprises the ability to effectively manage and retain talent, while also satisfying business requirements. We describe an analytics-based propensity scoring model for re-skilling by combining historical employee job-role/skill records, relationships between different job-roles/skills, employee resumes, and job postings. This is used to determine the source features that are the closest to a required target skill and hence identify employees that can be easily trained for the target skill. We evaluate this approach for a representative set of target skills at a multinational with a large services/consulting arm. We show that the propensity model learnt from the combined data sources has a high accuracy that is also substantially better than that achieved by using features from job-roles or resumes alone. The performance is improved further by using an ensemble model to evaluate the propensity scores.