This informal CPD article, ‘What Business Leaders Need to Understand About Emotional AI’, was provided by Jonathan Prescott, Strategy Director at AI Wales, who help people understand artificial intelligence theories, techniques and tools.
Introduction
Artificial intelligence has transformed how businesses process data, automate tasks, and generate insight. Yet one dimension of human experience has, until recently, remained largely outside the reach of machine analysis: emotion. Emotional AI - the branch of artificial intelligence concerned with recognising, interpreting, and responding to human emotional states - is changing that. For business leaders, understanding both its possibilities and its responsibilities is no longer optional.
This article provides an educational overview of emotional AI: what it is, how it works, where it is being applied, what ethical questions it raises, and how it is likely to reshape leadership and decision-making in the years ahead.
What Is Emotional AI?
Emotional AI, sometimes called affective computing, refers to systems that can detect and process signals associated with human emotional states. These signals may be drawn from multiple channels, including:
- Facial expressions and micro-expressions
- Voice tone, pitch, pace, and prosody
- Body language and gesture
- Written text and linguistic patterns
- Physiological indicators such as heart rate or galvanic skin response
The field has its roots in academic research from the 1990s, particularly the work of Rosalind Picard at MIT, whose 1997 book Affective Computing laid much of the theoretical groundwork. Since then, advances in computer vision, deep learning, and large-scale data processing have brought emotional AI from the laboratory into commercial applications.
A key scientific underpinning is the Facial Action Coding System (FACS), developed by psychologists Paul Ekman and Wallace Friesen in 1978. FACS provides a taxonomy of facial muscle movements — called Action Units (AUs) — that correspond to identifiable expressions. Modern emotional AI platforms map combinations of these AUs to emotional states such as confidence, surprise, engagement, or discomfort, providing a structured and replicable basis for analysis.
How Emotional AI Works in Practice
In applied settings, emotional AI typically follows a pipeline. Raw input — such as a video frame, an audio clip, or a paragraph of text - is captured and passed through a machine learning model trained to identify relevant features. Those features are then mapped to emotional or psychological constructs, often represented along dimensions such as Valence (positive to negative), Arousal (calm to excited), and Dominance (submissive to confident) — the VAD model, widely used in affective science.
The output might be a real-time confidence score during a sales conversation, an engagement heatmap across a training session, or a risk signal in a financial advisory interaction. The data can inform decision-making at the point of experience or be aggregated for broader pattern analysis.
Importantly, emotional AI does not read minds. It detects observable signals and applies probabilistic inference. Outputs are indicators, not certainties. Responsible implementations make this distinction clear and build uncertainty into how results are communicated.
Practical Business Applications
Emotional AI is being deployed across a growing range of sectors, with applications that span customer experience, human resources, health, education, and security.
In customer-facing environments, organisations are using emotional analysis to assess the quality of conversations in real time, identifying moments where customers show confusion, frustration, or disengagement. This allows for coaching, escalation, or process improvement based on actual behavioural evidence rather than survey data collected hours or days later.
In learning and development, emotional AI can measure whether training content is producing genuine understanding and engagement, not just completion. This is particularly relevant in high-stakes settings such as defence, healthcare, and aviation, where procedural knowledge must be reliably retained.
In recruitment, some organisations have explored using emotional analysis during interview processes, though this application area carries significant ethical complexity (discussed below). Financial services regulators, including the UK's Financial Conduct Authority, have begun to examine how vulnerability - a concept that includes emotional and cognitive states - should be detected and responded to in consumer interactions.
GDPR and EU AI Act
Ethical Considerations
Emotional AI raises important ethical questions that leaders must engage with seriously. Four areas deserve particular attention.
Consent and transparency. In most jurisdictions, the analysis of biometric or emotional data requires explicit informed consent, particularly under frameworks such as the UK GDPR and the EU AI Act. Individuals should understand what is being measured, why, and how the results will be used. Covert emotional monitoring is not only legally risky but corrosive to trust.
Accuracy and bias. Emotional AI systems are trained on datasets that may not reflect the full diversity of human expression. Research has documented variation in how facial expressions are produced and interpreted across cultures, ages, and neurological profiles. Leaders should require evidence of how any system they consider has been tested for bias and what its stated accuracy boundaries are.
Appropriate use. There is a meaningful difference between using emotional data to improve a product or service and using it to make consequential decisions about individuals - employment, creditworthiness, or security clearance. The latter demands a much higher threshold of justification, governance, and human oversight.
Data governance. Emotional and biometric data is classified as sensitive personal data in most regulatory regimes. Robust data handling, storage limitations, access controls, and audit trails are not optional features; they are baseline requirements.
The Impact on Leadership and Decision-Making
For leaders, the relevance of emotional AI extends beyond its operational applications. It signals a broader shift in how organisations will understand the humans within and around them.
Communication is increasingly a measurable performance dimension. As emotional AI tools become more accessible, leaders will face environments where their own communication patterns - the signals of confidence, uncertainty, or empathy they project — can be analysed. This is not a cause for alarm, but it does invite greater self-awareness and preparation.
Evidence-based empathy is becoming possible. Historically, decisions about culture, engagement, and change management relied heavily on intuition or delayed survey data. Emotional AI offers the possibility of real-time, evidence-based insight into how people are experiencing organisational life - a meaningful shift for leaders serious about wellbeing and performance.
Governance frameworks must evolve. The boards and executive teams that fail to develop a coherent position on emotional AI - its appropriate uses, its limits, and its oversight mechanisms - leave their organisations exposed, both reputationally and legally. Proactive engagement is preferable to reactive compliance.
Conclusion
Emotional AI represents a significant and expanding capability with genuine value for organisations willing to apply it responsibly. Business leaders who take time to understand how it works, where it is most appropriately deployed, and what protections individuals require will be better positioned to capture its benefits without incurring its risks.
The technology itself is neither inherently beneficial nor harmful. The quality of its deployment depends, as always, on the quality of human judgement surrounding it.
We hope this article was helpful. For more information from AI Wales, please visit their CPD Member Directory page. Alternatively, you can go to the CPD Industry Hubs for more articles, courses and events relevant to your Continuing Professional Development requirements.
References
- Picard, R.W. (1997). Affective Computing. MIT Press.
- Ekman, P. & Friesen, W.V. (1978). Facial Action Coding System: A Technique for the Measurement of Facial Movement. Consulting Psychologists Press.
- Mehrabian, A. (1996). Pleasure-arousal-dominance: A general framework for describing and measuring individual differences in Temperament. Current Psychology, 14(4), 261-292.
- Financial Conduct Authority. (2021). FG21/1: Guidance for firms on the fair treatment of vulnerable customers. FCA.
- European Parliament. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (AI Act).
- UK Information Commissioner's Office. (2023). Biometric data guidance. ICO.
- Barrett, L.F., Adolphs, R., Marsella, S., Martinez, A.M. & Pollak, S.D. (2019). Emotional Expressions Reconsidered: Challenges to Inferring Emotion From Human Facial Movements. Psychological Science in the Public Interest, 20(1), 1-68.
