Automated debt-collection messages are seen as more efficient and less stigmatizing without reducing trust, but human contact boosts perceived fairness, reciprocity and empathy — suggesting AI can improve efficiency but should be avoided where empathy or fairness sensitivity is crucial.

Main Finding

Across a large (n = 3,514), cross-national experiment in 11 European countries, AI-mediated debt-collection communication was perceived as more efficient and less stigmatizing than human-mediated communication, while human contact was perceived as fairer, more likely to elicit reciprocity, and more empathic. There was no meaningful difference in perceived trust between AI and human assistants.

Key Points

• Design: Between-subjects randomized experiment comparing a scripted telephone debt-collection interaction attributed to either a human assistant or an AI assistant.
• Sample: 3,514 participants (mean age 35, 48% female) recruited on Clickworker and Prolific across 11 European countries (e.g., Germany n=1036, Netherlands n=647, France n=421, etc.).
• Primary outcomes (5‑point Likert): perceived fairness, trust in information, reciprocity inclination, efficiency, stigma, perceived empathy.
• Main treatment effects (ordered logit; controls: age, gender, country fixed effects):
  • Fairness: higher for human vs AI (ordered logit coeff for human indicator = +0.278, p < .001). Predicted probability process rated as fair: human ≈ 90% vs AI ≈ 87%.
  • Reciprocity: higher for human (coeff = +0.155, p < .05). Predicted probability reporting intentions to reciprocate: human ≈ 72% vs AI ≈ 69%.
  • Efficiency: higher for AI (human coeff = −0.204, p < .01 → AI > human). Predicted probability rating high efficiency: AI ≈ 91% vs human ≈ 89%.
  • Trust: no significant difference (coeff ≈ +0.005, ns; predicted probabilities ~84–85%).
  • Stigma: higher for human (coeff = +0.606, p < .001). Predicted probability feeling stigmatized: human ≈ 19% vs AI ≈ 11%.
  • Empathy: higher for human (coeff = +0.234, p < .001). Predicted probability high empathy: human ≈ 79% vs AI ≈ 75%.
• Demographics: older participants and women tended to rate interactions as fairer, more trustworthy, more reciprocal, and more empathic; men reported higher stigma. Exploratory analyses indicated cross-country and subgroup variation (details in appendix).
• Robustness / analysis choices: outcomes analyzed with ordered logistic regression, age and gender included as covariates, country fixed effects used, robust standard errors reported. Standard errors were not clustered by country due to the small number of countries.

Data & Methods

• Stimuli: Two nearly identical telephone scripts (human vs AI); controlled for content (debt amount, installment offer, resolution). Authors note the AI script differed in features consistent with AI-mediated interactions (e.g., described shorter waiting/automatic processing vs human with pauses).
• Measures: Multi-item scales for fairness, trust, reciprocity, efficiency, stigma, empathy (5-point Likert). Also collected BFI-10 personality, demographics.
• Recruitment & timing: Online data collection (Unipark) Aug–Oct 2025; participants compensated (~€1).
• Statistical approach: Separate ordered logit regressions per outcome; main predictor = assistant type (human-coded 1), controls = age (continuous), gender (binary), country fixed effects; interaction terms with age and gender explored; α = 0.05.
• Limitations noted by authors: hypothetical scripted scenario (not real collectors/debtors), self-reported perceptions rather than behavioral repayment outcomes, potential stimulus differences (e.g., described waiting times/pauses), and limited ability to cluster SEs by country.

Implications for AI Economics

• Trade-off between efficiency and relational outcomes: AI reduces perceived stigma and raises efficiency—potentially lowering barriers to early contact and administrative cost—but human contact increases perceived fairness, reciprocity, and empathy, which may matter for long-term cooperation and sensitive cases.
• Trust resilience: Trust in the information did not decline with AI attribution, suggesting firms can deploy AI for informational tasks without necessarily reducing perceived trust—important for cost-reduction strategies that rely on automated communications.
• Targeting & segmentation: Heterogeneous responses by age, gender, and country imply value in segmenting communication strategies (e.g., automated, low-stigma messages for large-scale early outreach; human or hybrid handling for fairness-sensitive or high-distress cases).
• Policy and regulatory considerations: Regulators and firms should weigh efficiency gains against fairness and emotional costs. Because human agents were seen as more capable of moral judgement (and elicited more stigma), transparency, appeals processes, and fairness safeguards are important when automating consumer-facing financial interactions.
• Design recommendations for practitioners:
  • Use AI for routine, administrative, or early-stage outreach to capitalize on efficiency and lower perceived stigma.
  • Reserve human or hybrid (human-in-the-loop) interventions for negotiations, contested cases, or situations requiring empathy and fairness judgments.
  • Monitor downstream behavioral outcomes (actual repayments, dispute rates) and reputational metrics, not only perceptions.
  • Invest in empathetic design for AI messages (tone, personalization) and explicit fairness checks to mitigate perceived unfairness.
• Research agenda for AI economists:
  • Field experiments linking AI vs human contact to actual repayment behavior, default dynamics, and long-term customer value.
  • Cost–benefit analyses that incorporate psychological externalities (stigma, reciprocity) alongside direct cost savings.
  • Cross-country studies to quantify regulatory and cultural moderators, and experiments varying transparency (explicitly labeling messages as AI vs human-mediated) to test signaling effects.
  • Examination of heterogeneous effects among actual debtors and across different delinquency severities.

Short summary takeaway: AI can deliver efficiency gains and reduce stigma in debt-collection communications without eroding trust, but human contact still matters for perceived fairness, reciprocity, and empathy. Optimal deployment is likely hybrid and targeted, with evaluation focused on behavioral outcomes and equity safeguards.