This article is written by Pritesh Gupta, a serial entrepreneur, advisor, and investor.
The Science of Behavioural Prediction in Talent Assessment — And Why Traditional Tools Are Failing
What the gap between who someone presents in an interview and who they actually are at work is costing your organisation — and what behavioural science can do about it.
I’ve spent a long time thinking about the gap between who someone presents themselves to be during a hiring process, and who they actually are when the pressure is on, the novelty has worn off, and the real work begins. That gap — and the cost of not being able to see across it — is one of the most expensive, most persistent, and most underappreciated problems in modern organisations.
I remember sitting with a head of talent at a large BPO operation who told me they’d just lost their third cohort in a row to early attrition. They had good processes, decent interviewers, standard psychometric tools. They couldn’t understand why the people who looked right on paper kept leaving. The answer, as I’ve come to understand it, is that the tools they were using could only see the surface. And the surface is exactly what candidates learn to manage.
Here’s the uncomfortable truth: despite decades of investment in applicant tracking systems, structured interviews, psychometric questionnaires, and CV-screening algorithms, the outcomes from hiring have barely moved. The tools we use to assess people are fundamentally limited in what they can see. And what they can’t see is the part that actually matters.
The Scale of the Problem
Let me start with the numbers, because they’re worse than most people realise.
Gallup’s research on global workforce engagement makes for sobering reading. Around 77% of employees globally are not engaged or actively disengaged at work — costing the world economy an estimated $8.8 trillion per year, equivalent to roughly 9% of global GDP. That’s not disengagement as a soft concept. That’s an enormous, quantifiable, structural drag on the value organisations are trying to create.
And then there’s early attrition — the specific problem of employees who leave within their first 90 days. This is where the financial damage is most acute and most visible. For entry and mid-level roles, the replacement cost typically runs at 50% to 150% of annual salary once you factor in recruitment fees, onboarding time, lost productivity, and team disruption. For senior roles, that figure climbs considerably higher.
The question worth sitting with is this: if a company’s best people are engaged and its worst attrition is in the first 90 days, then the hiring decision itself — specifically the ability to accurately predict fit, commitment, and performance — is one of the highest-leverage decisions the organisation makes. And yet it’s routinely made with tools that were designed decades ago and have fundamental scientific limitations.
One of our clients, Strongbond Philippines Inc., was experiencing 3-month churn rates of 25–30% when they came to us. Every quarter, a quarter of their new hires were cycling out. The cost wasn’t just financial — it was operational. Projects were disrupted. Teams were destabilised. The cycle was compounding.
Within three months of deploying InsightGenie’s behavioural AI platform, their 3-month churn rate dropped to 0%. I’ll come back to how and why — but I want to first explain what makes the current approach so systematically limited, because the Strongbond outcome only makes sense once you understand the science.
What Current Tools Are Actually Measuring (And Why That’s The Problem)
Most HR assessment tools share a common failure mode. It helps to state it clearly:
Traditional assessment: measures what a candidate is willing to say about themselves InsightGenie: measures how a candidate’s biology actually behaves
That distinction sounds simple. Its implications are profound.
CV screening and ATS systems are retrospective by design. They look at what someone has done, not who they are. They filter for proxies — degrees from particular institutions, specific company names, keyword matches — that have weak predictive validity for performance in new contexts. They also amplify existing biases, because they use past hiring decisions as their reference point.
Self-reported psychometric assessments have a specific problem that anyone with a background in behavioural science will immediately recognise: they can be gamed. Not necessarily deliberately — though that happens too — but because when people know they’re being assessed for a hiring decision, they present a filtered, socially desirable version of themselves. The result is that the assessment measures the candidate’s understanding of what the assessor wants to hear, rather than their actual dispositional profile. The signal collapses precisely when you need it most.
Structured interviews are better than unstructured ones, but they’re still dependent on a candidate’s ability to perform under interview conditions — a skill that correlates weakly with the skills needed for most jobs, and which can be practised and gamed through interview coaching.
What none of these approaches can access is the layer beneath the candidate’s conscious self-presentation: the stable, physiologically grounded personality traits and behavioural tendencies that actually determine how someone will perform, engage, and stay when the job gets hard.
That’s the layer InsightGenie was built to read.
The Science: What Voice and Behaviour Actually Reveal
This is the part I find genuinely fascinating, because the science has been developing for decades but has only recently become practically deployable.
The core insight is this: the way a person speaks encodes reliable, measurable information about their personality and psychological state — information that exists independently of the words they choose, and that cannot be consciously controlled or fabricated.
This isn’t sentiment analysis. It’s not about detecting whether someone sounds “positive” or “nervous.” It’s about the prosodic architecture of speech — the rhythm, intonation patterns, stress distribution, pitch variability, and temporal structure — that is generated by the larynx under the influence of the autonomic nervous system. These are physiological signals, not linguistic ones.
Brigitte Zellner Keller’s foundational 2005 research in Logopedics Phoniatrics Vocology demonstrated that individual speech prosody is a stable and distinctive marker of personality style — that even when two speakers perform identical tasks, their prosodic profiles remain consistently different in ways that encode psychological characteristics. Subsequent work by Lee, Park, and Um (2021, Applied Sciences) confirmed specific relationships between acoustic speech features and Myers-Briggs personality dimensions: speech rate, response time, loudness, and discourse marker frequency all discriminate reliably between personality types.
The personality framework InsightGenie’s platform is anchored in is the Big Five — Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism (OCEAN) — which is the most extensively validated model in the history of personality psychology, with decades of research linking it to real-world outcomes.
Two findings from that literature matter most for hiring:
Conscientiousness is the single strongest predictor of job performance across all role types. A 2023 study by Kang, Guzman, and Malvaso in Frontiers in Psychology, analysing over 19,000 participants across employee, supervisor, manager, and entrepreneur categories, confirmed that Conscientiousness distinguishes between employment levels with statistical significance. A meta-analysis by Fukuzaki and Iwata (2022, Industrial Health) found that Conscientiousness was the strongest personality predictor of work engagement (ρ=0.41), followed by Extraversion and Openness (0.38), with Neuroticism negatively correlated (−0.36). Together, the Big Five explain approximately 30% of the variance in work engagement across populations.
Neuroticism is the strongest predictor of burnout and early attrition. This is the direct scientific link to the Strongbond outcome. High Neuroticism — emotional instability, difficulty managing stress, susceptibility to anxiety under pressure — predicts with statistical regularity who will disengage and leave when the going gets difficult. And it’s encoded in voice prosody. It’s measurable without a questionnaire, without an interview, and without any deliberate cooperation from the candidate.
InsightGenie’s voice analytics engine extracts over 40 prosodic biomarkers from a 30-second natural speech sample — no special hardware, no structured questionnaire, no preparation advantage — and maps them to a validated psychometric profile using machine learning models trained on over 500,000 voice samples.
Beyond voice, the platform triangulates with digital footprint analysis — a structured reading of behavioural patterns across digital platforms that reveals habitual traits without self-report — and video-based physiological measurement using remote photoplethysmography (rPPG). The scientific basis for rPPG was established in a landmark 2013 paper by de Haan and Jeanne in IEEE Transactions on Biomedical Engineering, which demonstrated that camera-based methods achieve 92% agreement with contact pulse oximetry across diverse populations, with error rates a factor of two better than prior approaches. From a 30-second video clip, taken on any standard smartphone or webcam, the platform can extract cardiovascular signals — including heart rate variability — that serve as objective markers of stress resilience, emotional regulation, and autonomic balance.
I want to be honest about what this means and doesn’t mean. These are probabilistic assessments, not certainties. Behavioural science gives us strong predictive signals, not deterministic outcomes. What InsightGenie does is substantially narrow the uncertainty that hiring managers currently have to navigate with tools that are nowhere near as informative.
Because I’ve sat across the table from enough HR leaders to know the questions that come immediately, let me address them directly.
It is not a lie detector. Polygraphs measure physiological arousal on the assumption that deception produces stress — a model that is scientifically discredited and legally inadmissible in most jurisdictions. InsightGenie measures stable personality traits encoded in the prosodic structure of speech. These are not arousal states; they are dispositional characteristics that are consistent across contexts and time. Very different science, very different purpose.
It is not surveillance. The assessment requires active, informed participation. A candidate provides a voice sample as part of the hiring process, with full knowledge that it is being analysed for psychometric insight. All processing is GDPR-compliant, operates within each country’s data protection framework, and is hosted on AES-256 encrypted infrastructure.
It does not replace human judgement. It provides the information that human judgement has been operating without. Hiring managers still make the call. The difference is that the information they make it with is an order of magnitude more reliable than a 15-minute phone screen conducted on intuition and a CV.
It does not work by reading the content of what people say. This is perhaps the most important clarification. The biomarkers InsightGenie analyses are properties of how people speak — their prosodic patterns — not what they say. A candidate could discuss anything. The signal is in the physiology of speech, not the semantics.
The outputs InsightGenie generates are designed to be immediately actionable alongside existing hiring workflows:
All of this is generated within minutes of a 30-second voice sample. There is no questionnaire to game, no interview format to prepare for. The candidate simply speaks.
When Strongbond came to us, they weren’t looking for a science project. They were losing people. The seasonal nature of construction project cycles meant their workforce demand fluctuated, and every hiring cycle was producing the same outcome: 25–30% of new hires were gone within three months.
The root cause, from a behavioural science perspective, was selection error. The tools they were using — and that most organisations use — couldn’t distinguish between candidates who had the psychological makeup to commit and persist through difficult working conditions, and those who didn’t. Both groups looked similar on a CV and in a brief interview. The difference only became visible under pressure.
InsightGenie’s voice-based assessment identified candidates whose Conscientiousness and Emotional Stability profiles predicted commitment under exactly those conditions. The selection process changed. Three months later, churn was zero.
In the words of the President of Strongbond Philippines Inc.: “InsightGenie brought unprecedented stability to our workforce, dramatically reducing employee churn, minimizing hiring costs, and enhancing project efficiency. All at a very reasonable price point.”
That’s not a soft outcome. That’s a structural change in the quality of every hiring decision the organisation makes.
Across InsightGenie’s broader deployments in credit risk — where the same voice and behavioural assessment predicts loan repayment behaviour — the models achieve 82–93% predictive accuracy, with a Gini coefficient improvement of 65–77% over traditional methods and a 30–50% reduction in non-performing loans. The HR application draws on the same validated science.
CHROs and HR directors at organisations experiencing attrition problems, engagement challenges, or volume hiring pressures — particularly in sectors like BFSI, healthcare, BPO, construction, and retail — will find that InsightGenie addresses the root cause of those problems rather than the symptoms.
Talent acquisition teams can shortlist faster, with greater confidence, and with a scientifically grounded basis for decisions they’re currently making intuitively. A voice-based Big Five score in 30 seconds replaces a 45-minute structured assessment that candidates drop off from — and does so with higher predictive validity.
Recruitment agencies and RPOs can differentiate on candidate quality, reduce replacement placements, and offer clients something genuinely novel: a placement backed by behavioural science, not just a screened CV.
Healthcare and specialist organisations — where psychological fit and resilience are as important as technical qualification — have a particular need for exactly what InsightGenie provides.
If you’re evaluating HR assessment tools — whether you’re considering InsightGenie or anything else — here are the questions that will separate genuinely differentiated science from repackaged self-report:
1. Does the assessment require the candidate to answer questions about themselves? If yes, it’s self-report. Self-report instruments are gameable in a hiring context. Ask specifically how the vendor addresses socially desirable responding.
2. What peer-reviewed science underpins the link between the assessment and actual job performance or retention? Ask for the specific citations. Not white papers — peer-reviewed journal publications. If the vendor cannot provide them, the scientific basis is weaker than their marketing suggests.
3. What is the assessment’s predictive validity for the specific outcome you care about — job performance, retention, culture fit? Predictive validity is a precise statistical concept. Ask for it. A number like “our candidates perform better” is not predictive validity. A correlation coefficient against a labelled outcome dataset is.
4. How has the assessment been validated to prevent demographic bias? Any psychometric tool that has not been explicitly tested for differential impact across gender, age, ethnicity, and language background should not be used in a hiring context. Ask for the bias analysis data.
5. What does the assessment measure when the candidate is trying to game it? This is the question that matters most. The honest answer from most self-report vendors is: it measures what the candidate wants you to think. A voice-based assessment has a different answer: it measures what the candidate’s physiology is doing, regardless of what they want you to think.
Hiring is hard because people are complex, and the tools most organisations use to assess that complexity are thin. A CV tells you what someone has done. An interview tells you what they’re willing to say under social pressure. A questionnaire tells you what they think you want to hear.
Voice prosody tells you who they actually are. And after years spent studying the science of human behaviour, I find that genuinely remarkable — and genuinely useful.
If you’re losing people in the first 90 days, the problem isn’t your onboarding process. It’s your selection signal. InsightGenie fixes the signal.
Want to see how it works with your own voice?
Book a 30-minute demo — we’ll run a live assessment during the call so you can experience the output firsthand. No slides. No pitch deck. Just the science, live.
InsightGenie is a Behavioural AI platform headquartered in Singapore, active across Bangladesh, Vietnam, Indonesia, Philippines, and India.
How does voice analysis predict job performance? The prosodic features of speech — rhythm, pitch variability, stress patterns, speech rate — are generated by the autonomic nervous system and encode stable personality traits including Conscientiousness and Emotional Stability. These are the two strongest known predictors of job performance and retention respectively. InsightGenie’s engine extracts over 40 of these biomarkers from a 30-second voice sample and maps them to a validated Big Five psychometric profile without requiring any self-report from the candidate.
What is the difference between voice-based psychometric assessment and a traditional personality test? Traditional personality tests rely on self-report — candidates answer questions about themselves. In a hiring context, candidates present a socially desirable version of themselves, which reduces the assessment’s discriminative power. Voice-based assessment reads physiological signals in speech that are not under conscious control and cannot be fabricated. The assessment measures who someone is, not what they choose to present.
Can candidates prepare for or game a voice-based assessment? No. The biomarkers InsightGenie analyses are prosodic properties of speech — rhythm, pitch variability, stress patterns — that are generated by the autonomic nervous system and are not under conscious control. Unlike interview performance or questionnaire responses, these signals cannot be rehearsed or strategically manipulated.
How accurate is the assessment? InsightGenie’s behavioural models achieve 82–93% predictive accuracy across credit risk deployments — where repayment behaviour is a direct, measurable outcome against which the model can be validated. In HR applications, the Strongbond Philippines outcome (3-month churn from 25–30% to 0%) reflects the same underlying science applied to retention prediction.
Is the assessment GDPR-compliant? Yes. All data collection is with candidate consent and full knowledge of the purpose. InsightGenie operates within each country’s data protection framework, hosted on AWS private cloud infrastructure with AES-256 encryption.
What languages does it support? InsightGenie’s platform is language-agnostic. The prosodic biomarkers it analyses are physiological properties of speech that exist across all languages. The model has been validated across multiple Asian and European language populations.
How long does the assessment take for a candidate? 30 seconds of natural speech. The entire candidate-facing step adds less than a minute to any existing hiring workflow.
Does it replace interviews or other assessments? No — it complements them. InsightGenie is designed to sit alongside existing workflows, not replace them. It provides a physiologically grounded psychometric signal that traditional methods cannot generate, which hiring managers can use in combination with their existing process.
