Self-Evolving Virtual Agents for Enterprise Financial Helpdesks: A Simulation-Based Longitudinal Study
American Journal of Technology · 18.05.2026 · 5 (5) · 59-75
DOI: 10.58425/ajt.v5i5.540 ↗Kurz gefasst
Fachartikel in der Zeitschrift American Journal of Technology, online erschienen am 18.05.2026. Verfasst von Swaraj Singh. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Background
Static LLM-based virtual agents in financial helpdesks face declining ticket deflection because their knowledge bases become outdated faster than teams can manually update them. To address this, this paper proposes a self-evolving architecture that autonomously expands and refines its knowledge using reinforcement learning from human feedback (RLHF) and retrieval-augmented generation (RAG).
Methods
We evaluated this architecture against a static LLM baseline using a discrete-event simulation. The simulation replayed approximately 270,000 historical tickets from a Fortune 50 investment bank over a 12-month operational horizon. We also used Monte Carlo analyses to test the system’s robustness against variations in ticket volume and difficulty.
Results
The self-evolving agent increased ticket deflection from 38% to 61% (t(18) = 7.45, p < 0.001, Cohen’s d = 3.51). It achieved 340% knowledge base growth at human-comparable quality (ANOVA: F(1,58) = 2.15, p = 0.15) while reducing manual curation effort by 67%. Additionally, user satisfaction improved from 4.8 to 5.9 on a 7-point scale (p < 0.001).
Conclusion
Continuous, feedback-driven knowledge self-evolution offers a viable and operationally superior alternative to static LLM deployments in regulated enterprise environments.
Zitieren
APA
Singh, S. (2026). Self-Evolving Virtual Agents for Enterprise Financial Helpdesks: A Simulation-Based Longitudinal Study. American Journal of Technology, 5(5), 59-75. https://doi.org/10.58425/ajt.v5i5.540
Vancouver
Singh S. Self-Evolving Virtual Agents for Enterprise Financial Helpdesks: A Simulation-Based Longitudinal Study. American Journal of Technology. 2026;5(5):59-75. doi:10.58425/ajt.v5i5.540
BibTeX
@article{Singh2026,
author = {Singh, Swaraj},
title = {Self-Evolving Virtual Agents for Enterprise Financial Helpdesks: A Simulation-Based Longitudinal Study},
journal = {American Journal of Technology},
year = {2026},
volume = {5},
number = {5},
pages = {59--75},
doi = {10.58425/ajt.v5i5.540}
}
EUVIMED