Publications

This paper examines how AI-driven systems can simulate empathy-like responses in service settings and help reduce the emotional burden placed on human workers. It argues that while AI cannot truly “feel” emotions, it can still provide consistent affective support in contexts such as customer service, healthcare, and care for older adults. At the same time, the paper emphasizes important risks, including data privacy concerns, manipulation of vulnerable users, labor displacement, and the need for transparency when consumers interact with emotionally responsive AI. It also discusses practical and policy-oriented solutions, such as disclosure requirements, algorithmic accountability, hybrid human–AI service models, and ethical safeguards for vulnerable populations.

Why this highlights FinE Lab’s expertise

FinE Lab’s expertise is reflected in its interdisciplinary approach to AI, consumer psychology, and consumer well-being. This work shows how AI can shape consumer experiences not only through efficiency and automation, but also through emotional influence, trust, and perceived support. By combining perspectives from behavioral science, data-driven technology, and consumer policy, FinE Lab helps advance research on how AI systems should be designed and governed to protect both consumer autonomy and worker well-being.

Paper is accessible: https://onlinelibrary.wiley.com/doi/10.1111/joca.70050

At FinE Lab, we believe that consumer voices are not just background noise in the financial marketplace—they are valuable behavioral signals that can reveal emerging risks, unmet needs, and changing market dynamics. Our recent research demonstrates exactly that. By analyzing more than 1 million consumer complaint narratives submitted to the CFPB and combining natural language processing, time-series panel modeling, and gradient boosting, we found that complaint volume, complaint concentration, and emotional tone contain meaningful information for understanding stock price movements in the U.S. financial sector.

This study speaks directly to a central mission of FinE Lab: using advanced analytics to better understand the intersection of consumer experience, financial behavior, and institutional decision-making. In our view, consumer complaints should never be ignored or treated as routine administrative data. Complaints often reflect real frustration, confusion, distrust, or barriers in the consumer’s financial life. When these signals accumulate, they can indicate deeper operational, reputational, or relational problems that financial institutions cannot afford to overlook.

At the same time, the rapid advancement of AI creates a major opportunity for banks and other financial institutions. Instead of reacting to complaints after damage has already spread, firms can now use AI-driven tools to detect patterns in consumer narratives earlier, identify common pain points faster, and respond more strategically. In other words, AI offers financial institutions a practical way to narrow the gap between themselves and the consumers they serve. This is especially important in an era when trust, responsiveness, and customer understanding are increasingly tied to both institutional reputation and market performance.

This is exactly where FinE Lab’s work is positioned. We focus on transforming behavioral and consumer-centered data into actionable financial intelligence. By integrating consumer complaints, sentiment, and other real-world behavioral signals into financial research, FinE Lab aims to help scholars, practitioners, and institutions better understand how consumer well-being and financial system performance are connected. Our research shows that listening carefully to consumers is not only good service—it is also smart finance.

Paper is accessible: More than words: valuation of words for stock price by using the combination of natural language processing, time-series panel and gradient boosting | International Journal of Bank Marketing | Emerald Publishing

This paper explains how recommendation systems (e.g., shopping or social media feeds) can narrow what people see and unintentionally strengthen common thinking shortcuts—especially confirmation bias (seeing more of what you already agree with) and anchoring (being pulled toward the first options shown).

It also outlines practical ways to reduce harm, including greater transparency, forced diversity/serendipity features, consumer education, and consumer-centered tools (user-controlled “re-rankers”) that shift recommendations toward user welfare and autonomy rather than platform profits.

Why this highlights FinELab’s expertise
FinELab’s work sits at the intersection of AI, consumer psychology, and consumer well-being, using an interdisciplinary lens (behavioral economics + data science + psychology + policy) to study algorithmic influence and design solutions that protect consumer autonomy.

Paper is accessible: https://onlinelibrary.wiley.com/doi/10.1111/joca.70032

In this article, we show how survey-based Financial Risk Tolerance (FRT)—captured at scale through repeated assessments—can be treated as a dynamic preference signal that co-evolves with stock-market conditions. Using signal decomposition and regime-switching evidence, the study separates gradual shifts from episodic deviations in aggregate FRT and evaluates when (and under what stress regimes) these components align most strongly with market dynamics.

For FinELab, this work exemplifies our core mission: turning behavioral/psychological measurements into actionable financial intelligence—building a methodological foundation for future AI-driven decision support that is more behaviorally grounded, inclusion-aware, and sensitive to changing market environments.

Paper accessible: https://www.mdpi.com/2227-7390/14/4/667

This early paper helped lay the foundation for what later became the Financial Intelligence for Consumer Well-being Lab (FinELab). Using U.S. household data (PSID), we compared a traditional regression approach with an Artificial Neural Network (ANN) model to better describe and predict key household financial ratios—such as savings-to-income and debt-to-asset. The main takeaway is simple: household finances are shaped by complex, nonlinear patterns, and ANN-based methods can capture those patterns more effectively than standard linear models—improving overall model fit and forecasting performance.

Why it matters

It reflects FinELab’s long-standing expertise in combining consumer well-being + data science to produce practical, decision-relevant insights for educators, practitioners, and policymakers.

It shows how AI/ML can improve understanding of real household financial conditions, not just markets.

Paper accessible: https://www.sciencedirect.com/science/article/abs/pii/S2214635019302230

The study reflects FinE Lab’s broader mission: moving beyond surface-level patterns to identify the underlying psychological and behavioral mechanisms embedded in everyday data. Rather than only using data-driven method, the lab uses theory-guided analytics as well to connect what people say, feel, and experience with what they actually do in real life.

Paper accessible: https://link.springer.com/article/10.1007/s11482-025-10431-3

Turning Online Reviews into Behavioral Intelligence: A Theory-Driven AI Framework

Online reviews are more than opinions—they contain behavioral signals that can help us understand why people return (or don’t) and how real-world decisions form over time. In this study, we introduce a theory-driven AI framework that combines the Theory of Planned Behavior (TPB) with modern machine-learning tools to connect what consumers say in reviews with what consumers do in real life.

What we did

We collected a large set of TripAdvisor hotel reviews and transformed the text into measurable psychological drivers—such as attitudes, social influence (subjective norms), and perceived control—using topic modeling (LDA) and sentiment scoring (VADER). We then tested how these factors shape revisit intention and actual revisit behavior, and evaluated predictive performance using a neural network model.

Key findings

• TPB signals are detectable in real-world text data: attitude, subjective norm, and perceived control extracted from reviews significantly predicted revisit intention.
• Intentions matter—because they connect psychology to behavior: revisit intention served as the key pathway linking TPB constructs to actual revisit behavior.
• AI improves prediction: a neural network achieved the strongest performance at 60 neurons (training precision 75.9%; test precision 59.3%; 65.6% with resampling), showing how ML can complement theory-based models.

Why this matters for FinELab (AI × Consumer Well-Being)

At FinELab, we build financial intelligence for consumer well-being by combining behavioral theory with AI/NLP to extract actionable insights from real-world, unstructured data. This paper exemplifies our approach: we move beyond simple sentiment trends to a theory-grounded, replicable modeling pipeline that translates text into psychological constructs and links them to observable behavior—exactly the kind of human-centered analytics needed for trustworthy, consumer-focused decision systems.

Paper accessible: A novel approach to online review analysis: integrating theory of planned behavior and machine learning techniques | International Journal of Contemporary Hospitality Management | Emerald Publishing

FinE Lab (Financial Intelligence for Consumer Well-being Lab) develops research and practical resources that help individuals and families navigate complex financial decisions. We combine consumer finance, behavioral economics, and AI/ML methods to understand financial attitudes and behaviors—such as financial risk tolerance, stress responses, and advice-seeking—and how these change when people face major shocks or uncertainty. For example, our research shows that extreme events like the COVID-19 pandemic can shift financial decision makers toward greater risk aversion, underscoring the need for timely, human-centered guidance and supportive interventions.

By partnering with researchers and professional communities, we aim to turn rigorous evidence into actionable strategies that improve financial capability, confidence, and long-term well-being.

Paper accessible: An Evaluation of the Effect of the COVID-19 Pandemic on the Risk Tolerance of Financial Decision Makers – ScienceDirect

Reducing Fashion Subscription Hesitation with Theory + Machine Learning

Fashion subscription services can deliver meaningful environmental benefits, but many consumers still hesitate to adopt them. In this study, we identify which specific concerns drive hesitation and test whether clearly communicating environmental benefits can reduce that hesitation.

What we did

Using a nationwide online experiment with 1,000+ U.S. adults, we combined theory-based causal modeling (SEM) with machine-learning predictive modeling to explain and validate the drivers of hesitation.

Key takeaways

• Not all concerns matter equally—financial risk, compatibility (fit/style uncertainty), and maintenance worries are the most important drivers of hesitation.
• Consumers’ ambivalence (mixed feelings) plays a meaningful role in how concerns translate into hesitation.
• Importantly, environmental-benefit messaging can weaken the impact of some concerns on hesitation—showing a practical way for brands to reduce adoption barriers.

Why this matters for FinELab

This paper reflects FinELab’s core strength: building human-centered intelligence by integrating behavioral theory + machine learning to generate actionable insights that improve consumer decision-making and well-being. Methodologically, we use ML to cross-validate theory-driven findings, strengthening both real-world relevance and predictive value.

Paper accessible: How to mitigate fashion subscription hesitation: two-step exploration using theory-based causal modeling and machine learning predictive modeling | Journal of Product & Brand Management | Emerald Publishing

Financial Risk Tolerance (FRT) can be understood in several ways. Beyond treating FRT as a single, fixed individual trait, an important approach is to view FRT as a dynamic signal that changes over time. In real life, people’s willingness to take financial risk can shift with market conditions, uncertainty, news, and social climate—and those shifts can provide meaningful information.

At the Financial Intelligence for Consumer Well-being Lab (FinELab), we study FRT as a time-varying signal by separating noise from informative changes and applying smoothing and machine-learning methods to produce interpretable indicators. We also pay attention not only to the average level of FRT, but to its variability (volatility), which may reflect uncertainty and collective sentiment in financial decision-making.

Paper accessible: Smoothing the Subjective Financial Risk Tolerance: Volatility and Market Implications | MDPI

At FinELab, we study how people’s stated financial preferences translate into real-world financial behaviors. In this publication, we examine a common but important pattern in investing: many investors report a willingness to take risk, yet their actual portfolios often reflect more conservative choices. The study compares elicited (self-reported) portfolio risk with revealed (portfolio-based) risk and explores whether working with a financial advisor is associated with closer alignment between the two.

This work offers practical implications for financial planning and client communication—especially around how risk tolerance is assessed, discussed, and implemented in portfolio construction. We’re sharing it here as part of FinELab’s ongoing effort to connect rigorous behavioral finance research with insights that can inform practice and improve consumer financial well-being.

Paper accessible: An Analysis of the Discrepancy Between Elicited- and Revealed-Portfolio Risk Among Individual Investors: Understanding the Role of Financial Advisors: Journal of Behavioral Finance

Aligned with FinELab’s mission to connect data-driven intelligence with real-world financial well-being, this study develops a practical machine-learning framework to improve prediction of insurers’ loss reserve error—a key indicator of reporting quality and reserving accuracy in insurance risk management.

Using 14 years of National Association of Insurance Commissioners (NAIC) property–liability insurer data, the paper applies a two-stage approach: unsupervised hierarchical clustering to form more homogeneous insurer groups, followed by supervised machine-learning models (ANN, Gradient Boosting, Adaptive Boosting, SVM) to forecast reserve error within clusters, benchmarking against standard linear regression.

Results show that boosting-based models—especially Adaptive Boosting—consistently outperform OLS, achieving lower prediction errors (RMSE/MAE) and stronger overall performance, while SVM performs poorly in this setting.

Impact: The combined unsupervised + supervised ML workflow offers a scalable, data-driven tool for monitoring reserving accuracy and supporting financial stability oversight.

Paper accessible: Improving insurers’ loss reserve error prediction: Adopting combined unsupervised-supervised machine learning techniques in risk management – ScienceDirect