References and tools

Bonelli, M. (2026). ILS-Bench: Investor Language-to-Suitability Benchmark. Mendeley Data, V1. Licence CC BY 4.0. doi.org/10.17632/w48mh2dtg5.1

Training data for both advisors, the shared label vocabulary, the Human review outcome and the case browser.

Pratama, R., and co-authors (2026). Systematic literature review of trust and algorithm aversion in the choice between human and AI financial advisors. Presented at SSRAAI 2026.

The review that identified miscalibrated trust as the problem. AdviceIT is its design-side follow-up.

Nahidi, N., and Zarifis, A. (Eds.) (2026). AI, FinTech, and the Future of Robo-Advisory: Risk Management and Ethical Considerations. Contributions to Finance and Accounting. Springer Nature Switzerland. doi.org/10.1007/978-3-032-18109-1

Domain background on AI-driven robo-advisory: the framing of suitability, escalation to a human adviser, and the ethics note and debrief in the study design.

Lee, J. D., and See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50 to 80.

The notion of appropriate reliance and calibrated trust that the dependent variables operationalise.

Dietvorst, B. J., Simmons, J. P., and Massey, C. (2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General, 144(1), 114 to 126.

Under-reliance, and the reason the flawed trials are debriefed.

Bansal, G., Wu, T., Zhou, J., Fok, R., Nushi, B., Kamar, E., Ribeiro, M. T., and Weld, D. S. (2021). Does the whole exceed its parts? The effect of AI explanations on complementary team performance. CHI 2021.

Explanations can increase reliance on wrong advice, the reason sound and flawed trials are both needed.

Rudin, C. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1, 206 to 215.

The interpretable-by-design position behind the interpretable rule-based advisor and the fidelity factor.

Lundberg, S. M., and Lee, S.-I. (2017). A unified approach to interpreting model predictions. NeurIPS 2017.

SHAP: Shapley values as feature attributions relative to a baseline, computed exactly for the neural network.

Wachter, S., Mittelstadt, B., and Russell, C. (2017). Counterfactual explanations without opening the black box. Harvard Journal of Law and Technology, 31(2).

The counterfactual content and the contrastive why-not panel: smallest change that alters the outcome, found by search.

Guo, C., Pleiss, G., Sun, Y., and Weinberger, K. Q. (2017). On calibration of modern neural networks. ICML 2017.

Temperature scaling and expected calibration error for the probabilities shown in the confidence content.

De Croon, R., and colleagues, Augment research group, KU Leuven (2025). Designing and personalising hybrid health explanations for lay users. ACM Transactions on Interactive Intelligent Systems.

The hybrid content and the adaptive delivery, and personalisation to a measured user characteristic.

Lusardi, A., and Mitchell, O. S. (2011). Financial literacy around the world: An overview. Journal of Pension Economics and Finance, 10(4), 497 to 508.

The Big Three financial literacy questions used as the moderator and by the adaptive delivery.