Trade-offs in Financial AI: Explainability in a Trilemma with Accuracy and Compliance

As Artificial Intelligence (AI) becomes increasingly embedded in financial decision-making, the opacity of complex models presents significant challenges for professionals and regulators. While the field of Explainable AI (XAI) attempts to bridge this gap, current research often reduces the implementation challenge to a binary trade-off between model accuracy and explainability. This paper argues that such a view is insufficient for the financial domain, where algorithmic choices must navigate a complex sociotechnical web of strict regulatory bounds, budget constraints, and latency requirements. Through semi-structured interviews with twenty finance professionals, ranging from C-suite executives and developers to regulators across multiple regions, this study empirically investigates how practitioners prioritize explainability relative to four competing factors: accuracy, compliance, cost, and speed. Our findings reveal that these priorities are structured not as a simple trade-off, but as a system of distinct prerequisites and constraints. Accuracy and compliance emerge as non-negotiable “hygiene factors”: without them, an AI system is viewed as a liability regardless of its transparency. Operational levers (speed and cost) serve as secondary constraints that determine practical feasibility, while ease of understanding functions as a gateway to adoption, shaping whether AI tools are trusted, used, and defensible in practice.

💡 Research Summary

The paper investigates how financial professionals balance five competing dimensions—accuracy, regulatory compliance, cost, speed, and explainability—when adopting AI systems. While the Explainable AI (XAI) literature traditionally frames a binary trade‑off between accuracy and explainability, the authors argue that this view is insufficient for the financial sector, where strict regulatory mandates, budgetary limits, and latency requirements impose additional constraints.

Using semi‑structured interviews with twenty stakeholders—including C‑suite executives, data scientists, risk officers, and regulators—from multiple regions, the study adopts a qualitative coding approach (open, axial, and thematic coding) validated by inter‑coder reliability (κ = 0.82). The analysis reveals a hierarchical “hygiene‑factor” model: accuracy and compliance are non‑negotiable prerequisites; without them, any level of transparency is irrelevant because the system is deemed a liability. Cost and speed function as secondary operational levers that determine whether a model is feasible to deploy at scale. Explainability, reconceptualized as “ease of understanding,” acts as a gateway to adoption—only models whose outputs can be readily interpreted by business users, documented for auditors, and communicated to regulators are actually used.

The authors further propose a four‑layer systems view of AI in finance: (1) the model layer (where classic accuracy‑explainability tension resides), (2) the decision‑support system (DSS) layer (where user‑centric, context‑specific explanations are needed), (3) the enterprise architecture layer (where latency and implementation cost are managed), and (4) the broader sociotechnical system (including institutions, norms, and external oversight). They argue that explainability must be delivered consistently across all layers—e.g., SHAP values must be transformed into intuitive dashboards, automated reports, and regulatory‑compliant documentation.

Key contributions include: (i) empirical evidence that financial AI decision‑making cannot be reduced to a simple accuracy‑explainability curve; (ii) a nuanced taxonomy of “hygiene factors” versus “operational constraints” versus “adoption gateways”; (iii) practical guidance for practitioners to design multi‑layered XAI pipelines that satisfy both internal efficiency and external regulatory scrutiny; and (iv) policy recommendations urging regulators to specify not only that explanations be provided, but also who the audience is, the format, and the timing of delivery.

Limitations are acknowledged—sample size, regional concentration, and reliance on self‑reported priorities—and future work is suggested, such as large‑scale surveys, longitudinal case studies of deployed AI systems, and rigorous fidelity assessments of LLM‑generated explanations. Overall, the paper reframes explainability in financial AI as part of a complex trilemma involving accuracy, compliance, cost, and speed, offering both scholarly insight and actionable roadmaps for industry and regulators.