Real-World Cloud AI Applications in Open Banking and SAP: A Gradient-Boosting and LLM-Driven Approach to Scalable Machine Learning and Software Testing Automation

Abstract

In the evolving landscape of financial services and enterprise software, the convergence of cloud infrastructure, large-language-models (LLMs), and ensemble machine-learning techniques offers new avenues for scalable automation in both incoming banking (open banking) and enterprise-resource planning (ERP) systems, particularly those centred on SAP S/4HANA or the broader SAP ecosystem. This paper presents a unified framework that combines gradient-boosting algorithms for structured transaction and operational data with LLM-driven modules for unstructured text analytics and software-testing automation in a cloud-native environment. In the open-banking domain, the framework processes large volumes of API-mediated banking events to detect fraud, personalise services, and orchestrate workflow automation. In the ERP domain, the same scalable infrastructure addresses predictive maintenance of SAP modules, automated test-case generation, and change impact analysis. The hybrid architecture leverages cloud elasticity, containerised micro-services, and continuous-integration/continuous-delivery (CI/CD) pipelines to deploy models rapidly and manage model drift. Empirical results from two pilot settings—a European challenger bank and a mid‐size manufacturing firm using SAP—show that the gradient-boosting component attained a 12-15 % uplift in detection accuracy compared with baseline logistic models, while the LLM-driven test automation reduced manual test-cycle effort by 40 %. Key advantages include cross‐domain reusability, improved scalability and accelerated development cycles; disadvantages stem from governance overhead, data-labelling cost and explainability constraints inherent in LLM-based modules. The paper concludes by outlining implementation guidelines, discussing results and proposing future research directions including federated learning in open banking and self-adaptive testing frameworks.