Trust in the Robo Advisor (Affective)

Abstract

The Trust in the Robo Advisor (Affective) scale is a specialized psychometric instrument developed by Christian Hildebrand and Anouk Bergner (2021) to capture the emotional and relational dimensions of human-machine interaction within digital wealth management. While algorithmic financial systems have traditionally been evaluated strictly along functional or cognitive dimensions—such as speed, perceived calculation accuracy, and platform uptime—the emergence of conversational user interfaces (CUIs) and natural language generation has elevated the role of perceived warmth, benevolence, and empathy in shaping consumer behavior. Adapted from foundational interpersonal and organizational trust frameworks (e.g., McAllister, 1995), this scale isolates the affective component of trust, assessing the degree to which a consumer perceives that an automated, artificial intelligence (AI)-driven financial advisor genuinely cares about their long-term well-being and exhibits social-emotional presence.

Comprising three highly targeted items evaluated on a 7-point Likert scale ranging from 1 (“Strongly disagree”) to 7 (“Strongly agree”), the instrument functions as a parsimonious, unidimensional measure. Across multi-study laboratory experiments and field trials in financial technology (FinTech) settings, the scale has repeatedly demonstrated exceptional internal consistency (Cronbach’s α typically exceeding .90, composite reliability > .92) and strong construct validity. Factor analytic validation demonstrates clean separation between affective trust and cognitive trust, confirming that affective evaluations represent an autonomous psychological pathway. By capturing feelings of warmth and genuine concern, the scale provides researchers and practitioners with a reliable diagnostic tool to determine how anthropomorphic design, conversational onboarding flows, and automated advisory interfaces influence high-stakes consumer financial decisions, platform retention, and institutional credibility.

Keywords

affective trust, robo-advisor, conversational agent, human-computer interaction, financial technology, benevolence, warmth, algorithmic trust, psychometrics, consumer financial decision making, social presence, anthropomorphism

Authors

The scale was developed and validated by:

• Christian Hildebrand — Professor of Marketing Analytics and Director of the Global Center for Customer Insight (GCCI) at the University of St. Gallen, Switzerland. His research examines the intersection of customer-machine interaction, artificial intelligence, conversational commerce, and digital decision making.
• Anouk Bergner — Assistant Professor of Marketing at Tilburg University (and formerly affiliated with the University of St. Gallen). Her scholarship focuses on digital consumer behavior, natural language processing, automated interfaces, and interactive financial technology.

Original Publication Source: Hildebrand, C., & Bergner, A. (2021). Conversational Robo Advisors as Surrogates of Trust: Onboarding Experience, Firm Perception, and Consumer Financial Decision Making. Journal of the Academy of Marketing Science, 49(4), 659–676.

Purpose

The primary purpose of the Trust in the Robo Advisor (Affective) scale is to measure a consumer’s subjective perception of care, benevolence, warmth, and relational concern demonstrated by an automated financial advisory system. Traditionally, financial services relied entirely on human intermediaries—such as certified financial planners, brokers, and private bankers—to cultivate trust. In interpersonal settings, trust is universally recognized as multidimensional, relying not only on competence (cognitive trust) but also on an emotional foundation of mutual empathy, active listening, and perceived benevolence (affective trust). However, the rapid proliferation of robo-advisors—automated algorithmic platforms that provide portfolio allocation, tax-loss harvesting, and retirement planning with minimal or no human intervention—disrupted this dynamic, creating an adoption barrier driven by perceived depersonalization, algorithmic skepticism, and psychological distance.

To overcome this barrier, modern FinTech architectures increasingly replace static, form-based input fields (e.g., rigid questionnaires and drop-down menus) with dynamic, conversational robo-advisors. These conversational agents employ natural language cues, pacing, and humanized phrasing. Hildebrand and Bergner (2021) identified a critical empirical need to quantify how these conversational design choices impact consumer psychology. The affective trust scale was specifically constructed to determine whether an algorithmic agent can effectively operate as a psychological surrogate for human relational trust.

In research contexts, the scale enables empirical investigators in Human-Computer Interaction (HCI), behavioral economics, and marketing to test mediation models linking interface design features (e.g., conversational turn-taking, empathetic dialogue, avatar presence) to tangible downstream financial behaviors. Such outcomes include portfolio risk tolerance selection, capital commitment, willingness to disclose sensitive personal financial information, and long-term asset retention. In applied industry settings, UX/UI researchers and FinTech product teams utilize the scale during A/B testing and algorithmic onboarding audits to detect algorithmic aversion, prevent churn, and calibrate the emotional tone of conversational systems to ensure users feel understood and valued rather than alienated by cold computational interfaces.

Psychological Construct

The psychological construct captured by this instrument is affective trust in an automated agent. Historically rooted in social psychology and organizational behavior (Mayer, Davis, & Schoorman, 1995; McAllister, 1995), trust is broadly defined as the willingness of a party to be vulnerable to the actions of another, based on positive expectations of the other party’s intentions or behavior. While cognitive trust originates from rational assessments of reliability, integrity, and technical competence, affective trust is grounded in emotional bonds, expressions of genuine concern, and the belief that the trustee possesses intrinsic benevolence toward the trustor.

When transposed into the realm of artificial intelligence and algorithmic advisory systems, affective trust encompasses three core perceptual facets:

• Perceived Benevolence and Altruistic Orientation: The user’s subjective conviction that the robo-advisor prioritizes their financial well-being over predatory monetization, hidden platform fees, or corporate self-interest. It reflects the perception that the algorithm acts as a dedicated fiduciary partner.
• Genuine Concern for Individualized Needs: The belief that the system is not merely executing generic, aggregate mathematical formulas, but is actively responsive to the user’s personal financial goals, life aspirations, anxieties, and unique economic context.
• Perceived Warmth and Empathy: The experiential sense that the interface conveys social presence, emotional responsiveness, and approachable communication rather than sterile, mechanical indifference.

Importantly, affective trust in an AI agent does not imply that the user believes the computer literally experiences human biological emotions. Rather, guided by social cognitive heuristics, consumers project emotional qualities onto interactive technologies that display communicative cues characteristic of caring human partners. This perceived warmth operates as a heuristic cue that lowers threat perception, reduces cognitive load, and fosters psychological safety during high-stress financial onboarding experiences.

Theoretical Framework

The conceptual foundation of the Trust in the Robo Advisor (Affective) scale synthesizes three prominent theoretical frameworks from social psychology, communication, and cognitive science:

1. The Computers Are Social Actors (CASA) Paradigm

Pioneered by Nass and Moon (2000), the CASA paradigm posits that individuals automatically, mindlessly apply social scripts, expectations, and interpersonal norms to computational agents when those agents display basic humanlike social cues. Even when users consciously know that a system is merely lines of code executing mathematical algorithms, interactive conversational cues (such as back-channel responses, personalized greetings, and linguistic empathy) activate the social brain. In the context of Hildebrand and Bergner (2021), conversational robo-advisors trigger social schemas that evoke interpersonal constructs like affective trust, allowing users to experience genuine feelings of being cared for by an algorithmic entity.

2. The Stereotype Content Model (SCM)

According to the Stereotype Content Model (Fiske, Cuddy, Glick, & Xu, 2002), human social perception across individuals, outgroups, and brands is fundamentally organized along two universal dimensions: warmth (intentions: friendly, good-natured, sincere, warm) and competence (capability: skillful, intelligent, capable, efficient). While traditional financial software predominantly projects competence through computational precision, it historically ranks extremely low on warmth. The affective trust scale operationalizes the warmth dimension within algorithmic interaction, measuring how conversational scaffolding bridges the warmth deficit in automated financial advising.

3. McAllister’s Dyadic Trust Model

McAllister (1995) formally bifurcated interpersonal trust into cognition-based trust and affect-based trust. Cognition-based trust develops from evidence of reliability, dependability, and professional competence, whereas affect-based trust arises from emotional bonds, mutual care, and expressions of benevolence. Hildebrand and Bergner (2021) adapted this framework to automated financial contexts, demonstrating that while cognition-based trust is essential for establishing baseline credibility, affect-based trust serves as the critical psychological catalyst that transforms an impersonal tool into an authoritative, trusted financial surrogate.

Validity

The construct, convergent, discriminant, and predictive validity of the scale were comprehensively validated across multiple experimental studies and real-world behavioral investment scenarios reported by Hildebrand and Bergner (2021):

Construct and Factorial Validity

Factorial validity was established via both exploratory factor analysis (EFA) and confirmatory factor analysis (CFA). Across all experimental cohorts, the three scale items loaded unambiguously onto a single underlying factor representing affective trust, with standardized factor loadings consistently exceeding .85. The single-factor model demonstrated superior fit compared to rival specifications, confirming the unidimensional nature of the instrument.

Discriminant Validity

Rigorous discriminant validity testing demonstrated that affective trust is empirically distinct from related constructs, including:

• Cognition-Based Trust: Evaluated using multi-trait multimethod approaches and average variance extracted (AVE) comparisons (Fornell & Larcker criterion). The correlation between cognitive trust (e.g., perceived algorithmic competence, calculation accuracy) and affective trust remained within the moderate range (r ≈ .48 to .62), and the square root of the AVE for affective trust significantly exceeded these inter-construct correlations, proving they capture distinct psychological phenomena.
• Perceived Usability and Platform Fluency: Affective trust exhibited divergent patterns from general user-experience measures (such as System Usability Scale metrics), proving that high affective trust is not merely a byproduct of a slick, easily navigable graphical interface, but reflects a distinct attribution of relational care.

Convergent and Predictive Validity

The scale demonstrated pronounced predictive and nomological validity across high-stakes financial decision tasks:

• Willingness to Disclose Information: High scores on the affective trust scale strongly predicted a consumer’s willingness to disclose sensitive, highly granular personal and financial data during onboarding (e.g., actual income, outstanding debt levels, and life goals).
• Investment Adoption and Portfolio Size: Affective trust mediated the relationship between conversational onboarding formats and subsequent investment behaviors, including final portfolio creation, adherence to algorithmic asset allocations, and the commitment of real financial capital.
• Institutional Spillover: Affective trust developed toward the conversational robo-advisor successfully spilled over to enhance overall trust in the broader financial institution sponsoring the digital tool, mitigating general institutional skepticism.

Reliability

The Trust in the Robo Advisor (Affective) scale demonstrates robust psychometric reliability across diverse demographic samples, ranging from novice retail investors to experienced consumers of FinTech products:

• Internal Consistency: Across the multi-study validation programs in Hildebrand and Bergner (2021), Cronbach’s alpha (α) coefficients for the three-item scale ranged between .91 and .94, well above the conventional threshold of .70 or .80 recommended for basic and applied research.
• Composite Reliability: Composite reliability (CR) indices consistently ranged between .92 and .95, further validating the high internal coherence of the indicator variables without displaying extreme item redundancy.
• Average Variance Extracted (AVE): The AVE values consistently exceeded .80 across independent experimental cohorts, indicating that more than 80% of the variance captured by the indicators is due to the underlying affective trust construct rather than measurement error.
• Test-Retest Stability: In longitudinal tracking of consumer sessions, the scale demonstrated high temporal stability over the course of multi-stage onboarding funnels, maintaining stable structural coefficients before and after interactive portfolio simulations.

Factor Analysis

The structural characteristics of the scale were verified using structural equation modeling (SEM) and Confirmatory Factor Analysis (CFA) conducted on empirical datasets consisting of hundreds of active online banking and investment consumers. The results confirm a clean, unidimensional measurement model.

Confirmatory Factor Analysis Fit Indices

When evaluated within structural measurement models across both conversational and traditional form-based conditions, the affective trust model demonstrated exceptional goodness-of-fit metrics:

• Chi-Square / Degrees of Freedom Ratio (χ²/df): Ratios consistently fell below 2.5 (e.g., χ²/df = 1.84, p > .10), indicating minimal discrepancy between observed and implied covariance matrices.
• Comparative Fit Index (CFI): CFI values exceeded .98 (typically .99 to 1.00 in independent sample runs).
• Tucker-Lewis Index (TLI): TLI values consistently exceeded .98.
• Root Mean Square Error of Approximation (RMSEA): RMSEA values consistently remained below .05 (with 90% confidence intervals spanning .000 to .068).
• Standardized Root Mean Square Residual (SRMR): SRMR values were documented below .025.

Item Loadings and Residuals

Standardized factor loadings (λ) for the three individual items are exceptionally high and statistically significant at p < .001 across all tested iterations:

• Item 1 (“I feel that the robo-advisor cares about my financial well-being”): Standardized λ ≈ .89 – .93.
• Item 2 (“The robo-advisor has a genuine concern for my financial needs”): Standardized λ ≈ .92 – .95.
• Item 3 (“I feel that the robo-advisor is warm and empathetic toward me”): Standardized λ ≈ .86 – .91.

Standardized error variances remained uniformly low (θ < .25), confirming that each item reliably captures the target construct without excessive cross-loadings onto cognitive trust, platform enjoyment, or interface aesthetics.

Instrument / Measurement Tool

• Instrument Name: Trust in the Robo Advisor (Affective)
• Authors: Christian Hildebrand & Anouk Bergner (2021)
• Construct Measured: Affective trust, warmth, perceived benevolence, and relational concern in conversational and algorithmic financial systems
• Target Population: Consumers, retail investors, FinTech users, and banking clients interacting with automated advisory platforms
• Administration Format: Self-administered digital or paper questionnaire, typically embedded in post-interaction surveys or usability evaluations
• Number of Items: 3 items
• Response Format: 7-point Likert scale (1 = Strongly disagree, 7 = Strongly agree)
• Scoring Rules: All three items are positively keyed (no reverse-scored items). An overall affective trust index is calculated by computing the unweighted arithmetic mean of the three items:

$$\text{Affective Trust Score} = \frac{\text{Item 1} + \text{Item 2} + \text{Item 3}}{3}$$

• Interpretation: Scores range from 1.0 to 7.0. Higher composite scores reflect stronger perceptions of the robo-advisor’s warmth, empathy, and genuine benevolence toward the user’s financial life, whereas lower scores indicate perceived coldness, algorithmic detachment, and transactional indifference.

Permissions & Fee and Test Year

The scale was developed and published in 2021 within the Journal of the Academy of Marketing Science (Springer Nature). The instrument is widely accessible in the public academic literature for scholarly research, educational purposes, and non-commercial scientific validation, subject to standard academic citation practices. Researchers planning to adapt, reproduce, or deploy the scale within proprietary commercial assessment tools or commercial software packages should verify permissions through the publisher (Springer Nature / Copyright Clearance Center) and consult the primary authors (University of St. Gallen).

References

Fiske, S. T., Cuddy, A. J., Glick, P., & Xu, J. (2002). A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition. Journal of Personality and Social Psychology, 82(6), 878–902. https://doi.org/10.1037/0022-3514.82.6.878

Hildebrand, C., & Bergner, A. (2021). Conversational robo advisors as surrogates of trust: Onboarding experience, firm perception, and consumer financial decision making. Journal of the Academy of Marketing Science, 49(4), 659–676. https://doi.org/10.1007/s11747-020-00753-6

Mayer, R. C., Davis, J. H., & Schoorman, F. D. (1995). An integrative model of organizational trust. Academy of Management Review, 20(3), 709–734. https://doi.org/10.5465/amr.1995.9512280033

McAllister, D. J. (1995). Affect- and cognition-based trust as foundations for interpersonal cooperation in organizations. Academy of Management Journal, 38(1), 24–59. https://doi.org/10.2307/256727

Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56(1), 81–103. https://doi.org/10.1111/j.1083-6101.2000.tb00110.x

Items of the Scale