One of the standout topics at this year’s SAP TechEd event was the integration of SAP’s innovative AI tools, specifically the SAP Foundation Model and SAP Knowledge Graph [^]. These advancements represent SAP’s focused effort to harness structured business data in ways that traditional AI models can’t. Built specifically for enterprise needs, SAP’s Foundation Model and Knowledge Graph offer solutions designed to align with SAP’s vast data ecosystem, bridging the gap between AI and complex business requirements.

This blog takes a closer look at SAP’s Foundation Model for structured data, the supporting role of the Knowledge Graph, and how these tools interact with Large Language Models (LLMs) to provide reliable, contextually grounded insights. Finally, there’s a look at key applications and future possibilities for SAP’s Foundation Model, transforming how AI can be applied within SAP business applications.

SAP Foundation Model: Structured Data Focused AI + The Role of the SAP Knowledge Graph

The SAP Foundation Model is designed to handle structured data, making it uniquely suited for tasks in SAP’s ecosystem where data is tabular and linked to defined business processes. This model addresses limitations that typical Large Language Models (LLMs) face with structured data, such as limited context and inadequate understanding of complex relationships. Here’s a breakdown of its key components and how it operates:

  • Purpose-Built for Structured Data: While LLMs excel at unstructured text processing, SAP’s Foundation Model is optimized for the tabular, structured data often found in ERP systems. It includes tailored embeddings for various data types—numerical, categorical, and textual—enabling it to handle business data efficiently and with accuracy.
  • Efficient Predictive Capabilities: Acting as a generic prediction engine, the SAP Foundation Model performs a range of predictive tasks like demand forecasting and anomaly detection. Its focus on structured data helps it achieve high accuracy in SAP-specific use cases, even outperforming some traditional AI methods on complex enterprise data.
The Role of the SAP Knowledge Graph

The SAP Knowledge Graph serves as a critical foundation, enhancing the SAP Foundation Model’s predictions by providing a structured, semantic layer that captures relationships across business objects. By creating a network of entities and their interconnections (e.g., linking CDS views, ABAP tables, and business objects), the Knowledge Graph offers several benefits:

  • Contextual Layering: The Knowledge Graph structures data as entities, properties, and relationships, making business data accessible in a way that’s both meaningful and flexible.
  • Data Integration Across Domains: It integrates both structured and unstructured data, capturing unique SAP business relationships and enabling consistent, accurate insights across SAP modules.
  • Accurate Retrieval through RDF and SPARQL: Using RDF standards and SPARQL for querying, the Knowledge Graph ensures precise data access and verification, grounding predictions and reducing the risk of inaccuracies.

Together, the SAP Foundation Model and Knowledge Graph offer a framework that combines the power of predictive AI with the structured data found in enterprise environments, facilitating robust and context-driven insights.

How SAP Knowledge Graph and LLMs Work Together

An essential innovation presented at TechEd is how SAP Knowledge Graph and LLMs interact to enhance AI functionality within SAP applications. This interaction uses SAP’s Vector Database Engine to streamline data retrieval and ensure accurate, grounded responses to user queries. Here’s a look at the process:

  1. User Query Input: A user submits a natural language query (e.g., “What is the base table of CDS view I_PURCHASECONTRACT?”). The LLM identifies key entities within the question, processing it with support from the SAP Knowledge Graph.
  2. Vector Matching with SAP Vector Database Engine: The identified entities are matched in the SAP Vector Database Engine, which stores the Knowledge Graph’s vector embeddings. This vector-based search process links the user query with the relevant SAP data entity, ensuring accurate retrieval.
  3. SPARQL Query Generation: Once the entity match is identified, the LLM generates a SPARQL query to access specific data points within the Knowledge Graph. This query references metadata (like base table relationships) to ensure the result is exact and relevant.
  4. Retrieving Accurate Results: The SAP Knowledge Graph provides a response based on verified SAP metadata, avoiding “hallucinations” or fabrications that can occur in traditional LLMs. This grounding allows the LLM to deliver precise, fact-based responses.
  5. User-Friendly Output Generation: The retrieved data is then converted into a user-friendly format by the LLM, allowing the user to receive accurate, contextually relevant information in natural language.

This interaction between SAP Knowledge Graph and LLMs empowers business users to interact with SAP data in natural language, bringing greater accuracy and reliability to AI-driven insights.

Future Potential and Interactive Use Cases of the SAP Foundation Model

The SAP Foundation Model offers extensive applications within SAP’s ecosystem, particularly for structured data and predictive analytics. Below are some of its key use cases and future potential:

Key Applications of the SAP Foundation Model
  1. Demand Forecasting: The SAP Foundation Model can predict future product demand by analyzing both historical and real-time data, enabling businesses to manage inventory more efficiently.
  2. Financial Planning: It aids in generating accurate financial forecasts, identifying financial risks, and optimizing budget allocation—all essential for effective business planning.
  3. Risk Management: By analyzing patterns in structured data, the model can identify potential risks, such as fraud or supply chain disruptions, allowing businesses to take proactive measures.
  4. Inventory Optimization: Leveraging data-driven insights, the model helps businesses balance supply and demand, keeping inventory levels at optimal points for operational efficiency.
  5. Customer Insights: The SAP Foundation Model analyzes customer behavior data, providing insights that support personalized marketing strategies and improve customer experience.
Future Potential

The SAP Foundation Model and Knowledge Graph are positioned to expand their impact beyond predictive tasks, particularly through integration with SAP Joule, SAP’s generative AI copilot. This integration could enable conversational interactions with SAP data, allowing business users to make queries and receive insights in real time. Over time, the Foundation Model may support more complex decision-making tasks, becoming an invaluable tool in SAP’s suite of AI-driven business applications.

Conclusion

SAP’s Foundation Model and Knowledge Graph mark a transformative step in bringing AI capabilities to structured business data, addressing limitations of traditional LLMs and narrow AI models. Together, these tools enable SAP to deliver contextually grounded, accurate predictions across complex enterprise environments, while the Knowledge Graph ensures that AI outputs are verifiable and connected to real SAP data.

The integration with SAP’s Vector Database Engine allows for seamless data interaction, supporting reliable AI-driven insights in natural language. SAP’s vision of combining the Foundation Model, Knowledge Graph, and generative AI tools like Joule points to a future where AI is deeply embedded in SAP workflows, providing business users with powerful insights and decision support in a simple, conversational format.