Our Self-Improvement

The vision of the Autonomous Enterprise marks a shift in the ERP perspective. At its core is the promise of new enterprise software that uses the interplay of AI, autonomous agents, and integrated feedback loops to self-manage, adapt, and optimize itself.

I discussed self-optimization with my regulars at the bar: From our perspective, however, autonomous self-optimization must be subject to scrutiny. At its core, an ERP system functions as a deterministic general ledger and a reliable accounting system (system of record), where absolute precision, audit compliance, and legal conformity are non-negotiable prerequisites. In contrast, modern language models and generative AI approaches are based on probabilistic stochastic logic. When autonomous agents are expected to optimize business processes on their own, mathematical uncertainty clashes with the exactness of business transactions; consequently, the claim of self-sufficient system optimization without human oversight remains, for the time being, a theoretical pipe dream.

I assigned two interns to work on the SAP Knowledge Graph this summer: SAP positions the Knowledge Graph as the semantic foundation for all data and process optimization. The HANA Graph Engine converts relational data, APIs, process chains, and customer-specific extensions into a machine-readable ontology and a network of relationships consisting of nodes and edges. This Knowledge Graph is a crucial foundation for targeted system improvements: It provides the AI agents and the Copilot Joule with a compass-like business context.

However, our DSAG rightly points out that the practical data translation of this knowledge graph has yet to prove its long-term viability in highly complex customer environments. Our interns are in the middle of their computer science studies and are familiar with RDF (Resource Description Framework) and OWL (Web Ontology Language). The Resource Description Framework refers to a syntactic approach for formulating logical statements about any entity. RDF was originally conceived by the World Wide Web Consortium as a standard for defining metadata.

What does our SAP Basis team do? Together with the BDC, the BTP provides the technical runtime environment, the AI Core Framework, and the Governance Cockpit (such as the LeanIX-based AI Agent Hub) for developing AI agents. By linking process mining tools such as Signavio with the Joule Studio development environment, a closed-loop optimization cycle is created: Inefficiencies in business processes are identified, after which AI agents generate targeted suggestions for adjustments or realign workflows.

Despite this potential, we face business, organizational, and licensing barriers that put the dream of autonomous self-optimization into perspective. On the legal front, the European EU AI Act mandates continuous human oversight and transparency for autonomous systems, which is why true „self-optimization“ in the core ERP system would be legally prohibited without a final human approval signal (“human-in-the-loop”).

On the business side, SAP uses the autonomous vision as a lever to drive BTP and BDC adoption. While the availability of AI features in the cloud editions (Rise and Grow) increases the pressure to migrate, the use of the Generative AI Hub and the BDC platform subjects us to usage-based billing models with opaque capacity units and token fees, where unused credits expire at the end of the year and peak loads result in unpredictable costs.

The SAP Autonomous Enterprise provides tools for process automation; however, reliable optimization of the ERP system can only be achieved if we maintain control over data sovereignty, consistently streamline the ECC (Clean Core), and manage the AI agents under strict human governance.

noname@e3mag.com