Citation :

Aman Sardana, Pradeep Manivannan, Manish Tomar, 2025. "Synthetic-Persona Reinforcement Agents for Adaptive Authorization Policies" ESP International Journal of Advancements in Computational Technology (ESP-IJACT)  Volume 3, Issue 2: 25-35.

Abstract :

Authorisation systems in financial services are under growing pressure to strike the right balance between strong fraud control and minimal user friction. Rule-based authorisation models typically rely on a set of threshold-based static rules, which are often stiff and slow to react to shifts in fraud patterns or a variety of customer behaviour. It introduces a new framework which combines generative AI to forge synthetic customer personas that span a broad spectrum of spend behaviours and risk aversions.Without sacrificing privacy or requiring a significant amount of sensitive real data, these artificial personas allow for the extensive simulation of transaction behaviours. Using this synthetic data, reinforcement learning (RL) agents are trained to create thresholdless, adaptive authorisation policies that dynamically optimise the trade-off between reducing the risk of fraud and causing friction for users. The efficacy of the framework is at the same time tested through large-scale Monte Carlo simulations over realistic issuer–acquirer network topologies for millions of transactions under different hypotheses. The results imply that there is a large scope for fraud loss reduction, revenue lift, and friction reduction, especially for users with low risk. This research provides a scalable, privacy- sensitive solution to the problem of authorisation that can adjust in real time to changing risk environments and offers an exciting path to the next generation of financial authorisation systems that are able to learn and continually optimise policy.

References :

[1] Y. M. Akatkin, et al., “Digital economy: Conceptual architecture of a digital economic sector ecosystem,” *Business Informatics*, no. 4 (42) eng, pp. 17–28, 2017

[2] T. Thi, L. D. T. Ngoc, and S. N. Thanh, “Financial Data Management in the Digital Economy,” in *Proc. Int. Conf. Res. Manage. & Technovation*, Singapore: Springer Nature Singapore, 2023.

[3] Wronka, “Financial crime in the decentralized finance ecosystem: new challenges for compliance,” *J. Financial Crime*, vol. 30, no. 1, pp. 97–113, 2023.

[4] W. Hilal, S. A. Gadsden, and J. Yawney, “Financial fraud: a review of anomaly detection techniques and recent advances,” *Expert Syst. Appl.*, vol. 193, p. 116429, 2022.

[5] Ashun, A. Adaan, and B. Elly, “Business Analytics and Financial Modeling: A Review of the Literature,” 2024.

[6] Han, et al., “A review on financial robot process auto-mining based on reinforcement learning,” in *Proc. Int. Forum Digital TV Wireless Multimedia Commun.*, Singapore: Springer Singapore, 2021.

[7] O. Bello, A. B. Ige, and M. N. Ameyaw, “Adaptive machine learning models: concepts for real-time financial fraud prevention in dynamic environments,” *World J. Adv. Eng. Technol. Sci.*, vol. 12, no. 2, pp. 021–034, 2024.

[8] P. S. R. Soundarapandiyan, “AI-Driven Synthetic Data Generation for Financial Product Development: Accelerating Innovation in Banking and Fintech through Realistic Data Simulation,” 2022.

[9] O. Orogun, et al., “Strategies for Combating Synthetic Identity Fraud: The Role of Machine Learning and Behavioral Analysis in Enhancing Financial Ecosystem Security.”

[10] Balog and C. Zhai, “User Simulation in the Era of Generative AI: User Modeling, Synthetic Data Generation, and System Evaluation,” *arXiv preprint* arXiv:2501.04410, 2025.

[11] R. D. Camino, “Machine Learning Techniques for Suspicious Transaction Detection and Analysis,” 2020.

[12] M. Greco, S. D. Charlier, and K. G. Brown, “Trading off learning and performance: Exploration and exploitation at work,” *Hum. Resour. Manage. Rev.*, vol. 29, no. 2, pp. 179–195, 2019.

[13] J. Xiong, et al., “Parametrized deep q-networks learning: Reinforcement learning with discrete-continuous hybrid action space,” *arXiv preprint* arXiv:1810.06394, 2018.

[14] G. Leuenberger and M. A. Wiering, “Actor-critic reinforcement learning with neural networks in continuous games,” in *Proc. 10th Int. Conf. Agents Artif. Intell. (ICAART)*, SciTePress, 2018.

[15] B. Mzoughia, S. Borle, and M. Limam, “A MCMC approach for modeling customer lifetime behavior using the COM‐Poisson distribution,” *Appl. Stoch. Models Bus. Ind.*, vol. 34, no. 2, pp. 113–127, 2018.

Keywords :

Synthetic Personas, Reinforcement Learning, Adaptive; Fraud Detection, Generative AI, Transaction Simulation, Risk Modeling, Financial Security.