In conjunction with SALT, SAP is also developing the SAP Foundation Model, which is designed to facilitate predictive tasks on tabular data with minimal training requirements, as detailed in the accompanying PORTAL paper.

SAP has launched the 'Sales Autocompletion Linked Business Tables' (SALT), marking a significant step in providing the first real ERP dataset designed to enhance enterprise AI research.

This innovative dataset offers anonymized data from customers' ERP systems, addressing the critical shortage of realistic tabular data needed for training AI models in business environments.

Looking ahead, SAP plans to expand the SALT dataset to encompass a wider range of customers and use cases, thereby broadening the data available for AI research.

By supplying researchers with authentic enterprise data, SALT aims to bridge the gap between academia and industry, enabling the tackling of real-world problems.

Knowledge graphs will be instrumental in this initiative, as they provide structured representations of data that help AI models understand the relationships within the data.

The ultimate goal of the SALT dataset is to develop better foundation models for linked business data, which will advance enterprise automation.

However, challenges persist in obtaining enterprise data due to issues related to data privacy, confidentiality, and the inherent complexity of business data, which is often stored across multiple interconnected tables.

To address these challenges, SAP aims to foster collaboration with academic institutions, encouraging exploration and validation of new methods for managing tabular enterprise data.

Furthermore, business data is heterogeneous and can exhibit significant column imbalances, complicating the training of AI models.