AI-Driven Nudge Optimization: Integrating Two-Tower Networks and Multi-Armed Bandit With Behavioral Economics for Digital Banking Campaign

In the era of digital banking, personalized financial product recommendations play a crucial role in enhancing customer engagement and conversion rates. Traditional recommender systems often fail to account for behavioral economic factors and adapt to real-time user interactions, leading to suboptimal nudge strategies. This study tries to resolve this problem by creating a recommendation model that combines Two-Tower Networks (TWN) and Multi-Armed Bandit (MAB) reinforcement learning. The goal is to make nudge-based financial product suggestions that are better all the time. The proposed model leverages TWN to generate structured user-product rankings, capturing behavioral insights. These rankings serve as input to the MAB model, which continuously refines recommendations through an adaptive exploration-exploitation strategy. In contrast to static suggestion models, MAB changes the weight of different nudge strategies based on how engaged users are in real time. This approach makes sure that financial nudges are personalized and relevant to the user’s situation. The model stays stable even when tried with datasets that are out of date, which indicates that it can adapt to changing financial conditions. This study is unique because it uses both deep learning-based representation learning (TWN) and reinforcement-driven decision optimization (MAB). This lets us create a nudge strategy that can learn on its own and change over time. This study is a big step forward in using AI to make financial recommendations. It closes the gap between basic personalization and real-time adaptation. Future studies will focus on putting the model to use in real-life banking settings and improving reward functions to make users more engaged and help them make better financial decisions over time.