A Control-Theoretic Approach to Dynamic Payment Routing for Success Rate Optimization
This work addresses payment system reliability for companies like JUSPAY, offering an incremental improvement through a hybrid approach.
This paper tackled the problem of dynamic payment routing to maximize transaction success rates by introducing a control-theoretic framework, achieving an improvement of up to 1.15% in success rate over traditional rule-based routing in live production.
This paper introduces a control-theoretic framework for dynamic payment routing, implemented within JUSPAY's Payment Orchestrator to maximize transaction success rate. The routing system is modeled as a closed-loop feedback controller continuously sensing gateway performance, computing corrective actions, and dynamically routes transactions across gateway to ensure operational resilience. The system leverages concepts from control theory, reinforcement learning, and multi-armed bandit optimization to achieve both short-term responsiveness and long-term stability. Rather than relying on explicit PID regulation, the framework applies generalized feedback-based adaptation, ensuring that corrective actions remain proportional to observed performance deviations and the computed gateway score gradually converges toward the success rate. This hybrid approach unifies control theory and adaptive decision systems, enabling self-regulating transaction routing that dampens instability, and improves reliability. Live production results show an improvement of up to 1.15% in success rate over traditional rule-based routing, demonstrating the effectiveness of feedback-based control in payment systems.