Ridderprint

Discussion 155 not learn this and will keep recommending job candidates that were originally successful, but may no longer be. Some exploration may be introduced in the form of feedback loop, where the model is updated as candidates are processed and succeed or fail, and thus continuously learns by example. However, more rigorous learning through exploration could be incorporated. For instance, we could allow the system to recommend historically less successful candidates on purpose, every now and then: candidates that would normally have been rejected automatically by the system. As these unorthodox candidates flow into the later stages of the selection process (e.g., interview with a human recruiter), they are labelled as either success (hire) or failure (reject) and the system can be updated accordingly. This would allow the system to learn at a considerably faster rate, either updating its assumptions about the patterns that reflect successful hires or reaffirming its basis of knowledge in case of failure. Such exploration is widely deployable, also outside of decision-support systems, and used in many other domains. For instance, large tech companies optimize their services continuously through real-time experimentation. Similarly, marketing and pharmaceutical companies simultaneously explore and exploit what works best during marketing and clinical pilots. This exploration does not have to be random, but may also occur through organized, theory-based, experimental design. Translated to an HRM context, exploration could imply more consciously recruiting, hiring, training, rewarding, or promoting in ways that are different from the current procedures – different from what we currently believe to be effective. On the bright side, incorporating exploration in HRM processes helps to identify patterns and other insights that would otherwise not have been uncovered. This holds great potential for the optimization of our HRM policies and practices, allowing us to base HRM decisions on observations and facts instead of on legacy assumptions. Although disadvantages exist on the short term (e.g., time, costs, bad hires), organizations would learn and benefit from the newly generated knowledge on the long term. On the dark side, there are implications of exploration in an employment situation. A decision-support system that recommends random candidates for interviews every now and then will already be difficult to sell, let alone one that consciously recommends predicted misfits for learning purposes. Such exploration is likely in conflict with short-termbusiness goals, employer branding, candidate interests, and potentially even legal boundaries. Moreover, governments, societies, organizations, managers, HRM departments, employee councils, and employees themselves are probably not yet ready for such exploration in terms of their mindset. However, in my eyes, a better balance between exploitation and exploration is the only way for predictive analytics to achieve its full potential in the functional HRM domain. Therefore, scholars, practitioners, and governments need to consider what it means to (not) introduce machine learning to the HRM domain, with or without explorative elements. 7.3.5 Unclear Value and Integration On conferences, in newspapers, and at universities, the prominent message is that people analytics has the potential to generate incredible value for organizations. However, from an evidence-based perspective, there is little foundation for such claims.