For too long, assessment has been stuck between two worlds. The first is a world of mediocrity, characterised by the pervasive use of interviews and CVs with poor rigour and even poorer outcomes. The second is burrowing deeper into the rabbit hole of psychometric tools, which are highly generic in nature and miss the nuances of what actually drives performance in a role. In this article, we look at why talent assessment needs a new paradigm: one that measures what people can do, not just what they say they can do.
The interview has long been the go-to easy option in talent assessment, alongside CVs for shortlisting. This approach is rife with challenges, particularly high levels of bias, limited connection to actual performance, and troubling inconsistency across hiring managers and recruiting teams. These risks are well evidenced across both interviewing and the use of CVs in sifting candidates, and in sum point to an approach that is high on subjective noise but frequently light on genuine signal.
We know that well-conducted structured interviews can achieve good prediction. But they are highly dependent on the skill and consistency of assessors, and are labour intensive. Ensuring quality is hard, especially at volume.
The bigger challenge is that interviews are highly coachable and focused on past experience, not future performance. They strongly favour candidates who can tell a good story, and can be effective at deselecting those who find it harder to articulate their successes. But interviews are much less effective at differentiating the truly talented from the talented storytellers.
It has become commonplace in early careers hiring to see very similar interview answers, articulate but with an uncanny resemblance to the stock answer from ChatGPT or Claude. Candidates under pressure to get through hundreds of application processes are increasingly using AI tools to coach themselves on giving better answers. This is making it harder to get meaningful differentiation from interviews, and creates a sea of sameness that poses a real risk to the integrity of hiring and internal selection decisions.
Automation of the process, whether video interviews or AI interview agents, compounds the problem by rewarding storytelling at the expense of what candidates can actually do. Simply automating failure is not a solution.
Put simply, the traditional interview is giving an increasingly poor signal and getting drowned out by the noise. Rather than measuring what people can actually do, we are resorting to measuring how well they can tell a story about it, leaving organisations exposed if they can no longer consistently differentiate the best talent.
Over recent decades, the adoption of psychometric assessments has grown at pace. Aptitude tests, then personality questionnaires, have become widely used across key areas of talent assessment such as early careers and leadership development.
This has brought real benefits in consistency relative to relying on interviews alone. But we appear to have maxed out on the benefits these methods can deliver, and now face considerable limitations.
Psychometric tools tend to be highly generic by design. They measure broad constructs but are less effective at reflecting the specific needs of a given role or situation, meaning a huge amount of nuance is missed in the assessment process. This might suit test publishers selling the same product repeatedly, but it doesn't match the needs of employers wanting to capture the specific requirements of each role.
Crucially, they also offer an unconvincing candidate experience. Research by Thrivemap / OnePoll (2026)¹ found that across a balanced sample of candidates, the majority perceived job-relevant work sample assessments as valid (54%) and fair (51%), while only one in ten viewed psychometric tools as valid (10%) or fair (10%). Psychometric assessments are unloved by candidates because they are so abstracted from real-world context.
The self-report nature of personality tools means we are still measuring what people say they are like, not what they actually do. More structured than the interview, yes, but many of the same challenges remain.
Psychometric tools also typically use fixed multiple-choice responses, making them susceptible to AI use by candidates. Employer concern here is high, and with good reason: research suggests as many as 28% of candidates admit to using AI to help complete psychometric assessments remotely (Capterra, 2024)². The result is an outsized proportion of this group leapfrogging the rest. Increasingly invasive proctoring can mitigate this, but many recruiters have real concerns about how it affects the candidate experience.
The net result: psychometric assessments, whilst playing an established role in talent assessment, typically fail to capture the vital requirements of specific roles, and face significant risk as candidates weaponise AI to help land a role. Again, the noise has been amplified while the signal grows fainter. The key now is to escape this dead end and take a fresh approach.
Given all this, it's clear that talent assessment needs to overcome the limitations of traditional, generic approaches that lack relevance and only infer performance indirectly.
To gain a much stronger signal on talent and capability, we need a more contextualised, task-based approach: one that cuts through the noise to separate narrative and storytelling from real capability, especially now that storytelling itself has become AI-enabled.
Two existing approaches point in this direction. Situational judgement tests suffer from the same challenges as generic psychometric tests and can be easily gamed by candidates. Assessment and development centres put candidates in much more task-based, relevant settings, but because they are so labour-intensive and expensive to run, requiring many assessors, they simply aren't scalable at volume.
The opportunity to make task-based simulation truly scalable, and to introduce a new paradigm, depends on using the latest AI approaches intelligently to enable immersive role plays and assessment experiences. Combining language models with the latest voice recognition and streaming Live Personas, it's possible to bring real work tasks to life and directly reflect the needs of each role. Through this, we can unlock the next generation of assessment and change both the experience and effectiveness of assessment for good.
Much like human assessment centres, many of the same principles apply. The gamechanger from a measurement perspective is the capacity to use AI-driven technologies to deliver assessment at volume, and to accurately review and assess open responses, much as an assessor would review candidate performance in an observed exercise.
This requires demonstrating convergence between evaluations from AI-enabled simulations and ratings by expert human assessors, to underpin the validity of the approach. Recent research by Dempster (2026)³, for example, demonstrated high levels of agreement between AI analysis of role play transcripts, using careful prompting and evaluation against clear behavioural indicators, and independent ratings by expert psychologist assessors who observed the same role plays. The quality of the assessment criteria matters too: the clearer and more observable the behaviour, the stronger the convergence.
This evidence is crucial to understanding how we can robustly validate simulation assessments, and is fundamental to testing reliability and fine-tuning validation over time. Alongside this, simulations need to be assessed for bias across demographic groups, just like any other type of assessment, with links to on-the-job performance tested over time.
Alongside this emerging field of research, the existing evidence base for assessment centres gives strong grounds for optimism. It's also advisable that human-in-the-loop scoring remains in place, to ensure final decision-making is verifiable, accountable, and in line with data protection legislation such as GDPR.
By finally creating highly immersive assessment experiences that are also scalable, we can unlock benefits that have, until now, eluded talent assessment:
Talent assessment needs fresh thinking: a new paradigm, based on simulation at scale. Helpfully, we already have decades of experience with task-based assessment in centre settings to build on. By applying those same principles at scale, we can escape both the mediocrity of interviews and the dead end of generic tools.
This gives talent assessment a scalable way to become far more grounded and relevant. At Symulate, we directly measure what people can do, not just what they say they can do, delivering the stronger signal on talent that organisations need to succeed and grow.
1. Thrivemap / OnePoll, 2026
2. Capterra, 2024
3. Dempster, 2026
