SAP has introduced SAP-RPT-1, a tabular foundation model (TFM) designed specifically for structured business data rather than natural language text. The model was first announced in November 2025 and recently highlighted as part of a SAP Business AI report.
The SAP-RPT-1 is a purpose-built alternative to LLMs for enterprise use cases that rely on tabular data. It supports predictive and analytical tasks across core business functions. Accordingly, the SAP-RPT-1 supports the company’s broader effort to tailor AI models to the data structures that underpin enterprise systems.
What SAP-RPT-1 Is Designed to Do
SAP-RPT-1 is a foundation model built to operate on structured, tabular business data.
SAP describes the model as a relational pretrained transformer, meaning it is designed to understand relationships across rows, columns, and fields commonly found in enterprise datasets. Instead of predicting the next word in a sentence, SAP-RPT-1 predicts values within tables, such as financial records, operational data, or HR information.
The model is intended to support tasks such as forecasting, classification, and pattern detection across core ERP datasets. SAP has positioned SAP-RPT-1 for use cases where business data is highly structured and semantically rich, including finance, supply chain, and workforce management scenarios.
SAP-RPT-1 also supports in-context learning, allowing users to provide a small number of labeled examples directly within a dataset. The model can then generalize from those examples without requiring retraining, reducing the need for custom data science pipelines or lengthy model development cycles.
SAP has said the model is designed to operate efficiently on enterprise data, emphasizing lower computational requirements compared with general-purpose language models when applied to structured data tasks.
Why Tabular Models Matter for Enterprise AI
Most enterprise data is structured rather than conversational, and that reality shapes how AI can be applied inside operational systems.
ERP platforms generate data in tables with defined schemas, consistent fields, and embedded business rules. Financial transactions, inventory records, payroll data, and supplier information are designed for accuracy, auditability, and repeatability. That structure limits the effectiveness of large language models, which are optimized to predict text rather than reason across numerical relationships and field-level dependencies.
Tabular foundation models are designed to address that gap.
They aim to reduce the complexity and cost associated with traditional machine learning pipelines, which often require extensive feature engineering, model training, and ongoing maintenance. Meanwhile, in-context learning allows predictive capabilities to be applied and adjusted without lengthy development cycles.
At the same time, this approach is intentionally narrow.
SAP-RPT-1 is best suited to small to medium structured datasets where business semantics matter, such as finance, supply chain, and workforce data inside ERP systems. It is not designed to replace language models for unstructured text, conversational interfaces, or document-heavy workloads. Nor does it eliminate the role of traditional statistical methods for large-scale numerical analysis.
SAP is positioning SAP-RPT-1 as complementary infrastructure rather than a universal AI layer. Language models handle interaction and explanation, while tabular models support prediction and pattern detection inside transactional systems.
What This Means for ERP Insiders
Enterprise AI is shifting away from language-first models. SAP introduced SAP-RPT-1 to address structured business data, highlighting how many enterprise AI use cases depend more on tabular prediction than on natural language generation.
Tabular models trade breadth for operational fit. SAP-RPT-1 is optimized for structured ERP data and in-context learning, enabling faster deployment and lower overhead, but it is not designed to replace language models or traditional analytics across all scenarios.
Specialization is reshaping enterprise AI architecture. By pairing tabular models for prediction with language models for interaction, enterprises are moving toward layered AI stacks that reflect how operational systems store, govern, and use data.
