Bottom-up: utilising artificial intelligence to identify what successful leaders do, using an abductive approach

Purpose Identifying exceptional leaders remains a top priority for organisations. Traditionally, leaders were selected using a battery of assessments measuring key personal attributes believed to make leaders successful. However, there has been a shift away from selecting leaders based on an a priori list of virtues, towards a focus on the actual behaviours linked to high performance. This abductive approach in competency framework development for leaders starts with the behaviours demonstrated in assessments and then constructs the competency framework in a bottom-up approach. Design/methodology/approach In the current study, a total of 15 411 leaders from 283 different companies and 38 countries completed a 3.5-hour online assessment centre simulation exercise. Natural language processing and machine learning were then utilised to identify 38 competencies by screening more than 30 million words. We then mapped this artificial intelligence-derived framework to prominent leadership models. Findings Despite strong congruence between frameworks, the abductive approach revealed three critical findings: 1) Not all competencies are equally distributed across the leader population; 2) Competencies are distributed across job levels, with some commonly observed in junior leadership roles and others primarily amongst senior leaders; and 3) the artefacts collected by the machine-learning algorithm enable a more granular view of the behaviours that constitute a competency compared to human assessors. Originality/value Using abductive methods to supplement deductive models capitalising on the strengths of both approaches, yield more nuanced leadership competency frameworks.