Research
Does Explainable AI Improve Financial Integrity? The Evidence
It is easy to assume that adding AI to a finance function automatically makes it more trustworthy. New empirical evidence suggests otherwise: AI improves financial integrity mainly when it is wrapped in governance-explainability and human oversight-rather than on the strength of the models alone.
How the study was built
The analysis uses a fixed-effects panel regression across five advanced economies-the United States, Germany, France, Japan, and the United Kingdom-over the 2020-2024 period. The dependent variable is a composite Financial Integrity Index capturing risk management, internal controls, and transparency outcomes. The explanatory factors are AI Adoption, Explainable AI (XAI), Human-Centered Oversight (HCO), and Embedded Compliance Capacity (COMP), with country and time fixed effects and robust standard errors.
What the numbers show
Across all five economies, the relationship between AI-related factors and financial-integrity outcomes is stable and statistically significant. Crucially, the interaction terms matter: the effect of AI adoption grows substantially when it is paired with explainability (AI × XAI) and with human oversight (AI × HCO). Explainability alone is associated with a meaningfully higher integrity effect-on the order of 35-55 percent-while embedded compliance and human oversight each contribute their own significant, positive gains.
Financial integrity depends less on AI's raw predictive power than on the governance architecture around it.
Why it matters for practitioners
The practical takeaway is direct: deploying a more powerful model is not a substitute for transparency, auditability, and human-in-the-loop control. AI can manage risk sustainably only when it is embedded in institutional structures that are explainable, audited, and human-governed. For leaders, that reframes the build decision-invest in the governance layer with the same seriousness as the model itself.
Originally published on ResearchGate. Read the original article.