Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study investigates the impact of transitioning to agentic AI methodologies on the workflow-level return on investment (ROI) in portfolio risk analysis at Navy Federal Credit Union, comparing a pre-agentic baseline period with a subsequent transition period. By employing an evidence-constrained quasi-experimental design, the authors triangulate data from public financial indicators, digital-investment disclosures, and a newly constructed Agentic Portfolio Risk Capability Index (APRCI). The findings indicate significant increases in assets, members' equity, and net income, alongside a rise in the delinquency ratio, suggesting that enhanced analytics operated in a more challenging risk environment, while the APRCI reflects a substantial capability gain.
Transitioning to agentic AI methodologies led to a notable increase in portfolio assets and equity, but also highlighted the risks of operating in a more volatile environment.
This study examines whether publicly available evidence is consistent with an improvement in the workflow-level return on investment (ROI) of portfolio risk analysis at Navy Federal Credit Union after the institution moved from bounded digital-investment automation toward a broader generative and agentic artificial intelligence (AI) posture. The paper deliberately avoids claiming observed client alpha, portfolio outperformance, or causal financial effects because Navy Federal does not publicly disclose member-level portfolio returns, advisory assets by workflow type, workflow-cycle times, analyst labor costs, or a standalone AI ROI figure. Instead, the study uses an evidence-constrained quasi-experimental comparison of a pre-agentic baseline period (2023-2024) and a 2025 transition period. The analysis triangulates three evidence layers: public institution-level financial indicators, observable digital-investment capability disclosures, and an Author-Constructed Agentic Portfolio Risk Capability Index (APRCI). To make the APRCI more defensible, the revised methodology provides a five-dimension codebook, score anchors, source-to-score crosswalk, equal-weighting rationale, and sensitivity checks under alternative assumptions. Results show that assets increased from USD 170.8 billion in 2023 to USD 197.2 billion in 2025, members’ equity increased from USD 14.2 billion to USD 18.9 billion, and net income increased from USD 1.37 billion to USD 1.91 billion. At the same time, the delinquency ratio rose from 1.59% to 1.96%, indicating that stronger analytics would have operated in a more risk-intensive environment rather than an artificially benign one. The APRCI increases from 48 in 2023 to 64 in 2024 and 92 in 2025, with conservative sensitivity analysis still supporting a large directional capability gain. The bounded conclusion is that the public record is consistent with improved workflow-level ROI potential through faster review preparation, richer monitoring, improved documentation, stronger explanation capacity, and more scalable human-in-the-loop governance. The magnitude of ROI remains unobserved and should be tested in future work using de-identified workflow telemetry or synthetic portfolio-review datasets.