SAP Releases STS Framework: A New Paradigm for Enterprise Data Generation via Simulation

Summary:SAP has introduced the Synthesis Through Simulation (STS) framework, a novel approach to enterprise data generation that addresses the limitations of traditional data synthesis methods. By leveraging a large language model (LLM) within simulated enterprise environments, STS generates data that adheres to business rules and constraints without relying on database schemas. This ensures structural validity and high fidelity, achieving an average marginal fidelity of 0.88 and 100% constraint satisfa

---

Core Breakthrough

SAP's Synthesis Through Simulation (STS) framework offers a novel solution for enterprise AI training and evaluation by generating data through simulated environments. Traditional data synthesis methods struggle with structural validity and distribution fidelity, but STS addresses these challenges through the following innovations:

1. Schema-free Data Generation: STS generates data without relying on database schemas, eliminating the limitations imposed by schema dependency.
2. Data Generation in Simulated Environments: STS executes operations within simulated enterprise environments to ensure data validity and adherence to business logic.
3. High Fidelity and Constraint Satisfaction: STS achieves an average marginal fidelity of 0.88 and 100% constraint satisfaction across ten diverse environments, demonstrating its versatility in complex scenarios.

Technical Mechanism Analysis

The core component of STS is the Generalist Populator (GP), a domain-agnostic agent that generates data by executing operations within simulated environments. GP achieves high-fidelity data generation through:

• Operation Execution and API Interaction: GP interacts with APIs in the simulated environment to generate data that adheres to business rules.
• Distribution Fidelity and Constraint Satisfaction: GP achieves high fidelity and constraint satisfaction without accessing database schemas, showcasing its adaptability across different environments.

Engineering Trade-offs and Performance

While STS excels in high-fidelity data generation, it also involves some trade-offs:

• Computational Resource Requirements: Data generation in simulated environments demands significant computational resources, especially in complex enterprise settings.
• Data Diversity: Although STS performs well across multiple environments, the diversity of the generated data needs further validation.

Experimental results show that STS significantly outperforms traditional methods in handling complex enterprise data generation tasks, particularly in terms of data fidelity and constraint satisfaction.

Developer Adoption and Deployment Recommendations

For developers looking to adopt STS, the following recommendations are provided:

• Environment Simulation: Developers need to build accurate simulation models of enterprise environments to ensure the generated data meets actual business needs.
• API Design and Implementation: STS relies on APIs for data generation, so developers must design and implement APIs that align with business logic.
• Performance Optimization: In resource-constrained environments, developers should optimize STS's performance, for example, through distributed computing or model compression techniques.

Conclusion

STS provides an innovative approach to enterprise data generation for AI training and evaluation, showcasing its potential in high-fidelity data generation and complex scenario applications.

---

Source: ArXiv AI (cs.AI) (2026-10-10)

— END —

Tags: #SAP Follow #Data Generation Follow #Simulation Environment Follow #Enterprise AI Follow #Large Language Model Follow