FULL PDF
Abstract
The proliferation of customer relationship management (CRM) systems such as Salesforce has led to the accumulation of vast amounts of sensitive client data across globally distributed servers. However, privacy regulations and organizational policies often restrict centralization of such data. This paper proposes a federated learning (FL) framework tailored for Salesforce environments to enable collaborative model training without direct data exchange. By integrating differential privacy and secure aggregation protocols, the proposed framework maintains data confidentiality while achieving competitive model performance. We evaluate this framework on synthetic CRM datasets designed to simulate Salesforce instances, demonstrating a negligible performance drop (<2%) compared to centralized models. This approach offers a scalable, secure, and regulation-compliant alternative to conventional machine learning workflows.
View more >>