Assessment centres are widely used for selecting and developing employees, especially in high-stakes environments like law enforcement, government services, and corporate leadership. They typically involve candidates being assessed across multiple exercises, such as interviews or role plays, on competencies like communication or leadership. However, new research challenges a long-held assumption that these centres reliably measure competencies. Reliability is the most crucial element of any data point on which decision-making is based. Without reliability, there is no measurement, no useful information, and no foundation for guiding decisions.

To explore this, Professor Duncan Jackson and colleagues conducted one of the most extensive studies of its kind, involving 2,917 participants undergoing ACs within the UK police force. Their focus was simple yet crucial: how can assessment centre scores be made reliable?

For decades, many practitioners believed that assessment centres could measure competencies such as communication skills, teamwork, and problem-solving across different exercises. But this approach has repeatedly come under scrutiny as analyses repeatedly reveal that assessment centres don’t measure competencies. Instead, they are found to measure performance on each of the exercises included in the assessment centre (e.g., performance on a role play and performance on a group exercise). This is a very different basis for measurement than what is usually intended by the architects of assessment centres.

Professor Jackson’s team applied a sophisticated statistical method called Bayesian Generalizability Theory to evaluate reliability relating to three possible ways of scoring assessment centre data: by competency, exercise or by a combination of both called the mixed-model.

When competency scores were applied, average reliability was unacceptably low (at .38 on a scale ranging from 0 – 1, where .70 is often considered the minimum benchmark).

In contrast, when exercise scores were applied, reliability was much higher at .91, well above the .70 benchmark.

Interestingly, combining both types of data didn’t help. The mix added no real benefit and only made the results harder to interpret.

This evidence supports a shift toward a simpler model for assessment centres. Instead of trying to identify and score competencies, the suggested alternative approach focuses on how candidates perform in simulations of real job challenges. In this model, each exercise reflects a situation candidates are likely to encounter in their actual roles. Strong performance in one task is taken as a good indication of future success in similar situations.

These findings have important practical implications. If employers rely only on competency scores for hiring or promotion, they risk basing decisions on unreliable data. Unreliable data means that scores cannot be interpreted or compared meaningfully. A better approach is to evaluate how candidates perform on specific tasks, which provides more consistent and accurate results.

This shift would lead to a simplification of the way assessors are trained and how exercises are designed. Instead of rating abstract qualities such as general leadership, assessors would remain focused on observable behaviours. For example, they might assess how a candidate manages conflict in a simulated team meeting.

Professor Jackson’s research also challenges long-standing practices in the governance and design of assessment centres. It highlights the value of rethinking established systems when evidence points to more effective alternatives and will help organisations to redesign their assessment processes to make fairer, more accurate decisions. They also open the door to new research exploring how exercise-based assessments relate to deeper psychological traits, like cognitive ability and personality, offering fresh insight into what underpins strong workplace performance.

Ultimately, this research contributes to a growing body of evidence that urges practitioners to re-evaluate traditional tools and assumptions. In doing so, it brings companies closer to selection processes that are both scientifically grounded and operationally effective.