SAP Joule and the traveling salesman problem

Königsberg and the traveling salesman

The Travelling Salesman Problem (TSP) is a generalized version of the Königsberg Bridge Problem. In all cases, the aim is to find an optimal path (edges) through existing locations (nodes), whereby it should naturally be the shortest path and no location may be visited more than once. Unfortunately, there is no solution for Königsberg.

The city of Königsberg in Prussia is crossed by the River Pregel and two islands. The two halves of the city were each connected to the islands by three bridges, which were connected to each other by a further bridge (seven bridges in total). The question to be answered was whether there was a way to cross all seven bridges exactly once, and if so, whether a circular route could be constructed that would enable the return to the starting point. The mathematician Leonhard Euler proved in 1736 that the realization of such a path in Königsberg was not possible, as an odd number of bridges (edges) led to all four shore areas or islands (nodes).

It is assumed that there are a maximum of two banks (nodes) with an odd number of connected bridges (edges). It is possible that these two banks represent the starting or end point. The remaining bank sections must be provided with an even number of bridges in order to enable a new passage on the new path. The bridge problem is not a classic geometric problem, as the exact position of the bridges is not relevant, only the connection of the islands by the bridges. The problem at hand can therefore be classified as a topological problem. Euler succeeded in solving this problem by applying methods that are currently assigned to graph theory.

SAP Hana and the Graph Engine

The Walldorf-based software group SAP is currently presenting its artificial intelligence as a panacea for modern ERP, but a closer look at technical platforms such as SAP Hana and SAP BTP reveals a highly complex interplay of old concepts and new promises of salvation around the Hana database platform. In addition to classic SQL processing, the so-called SAP Graph Engine, which stores and processes data not in rigid tables but as a networked mesh of nodes and edges, has been working in the engine room of this architecture for some time.

Although graph theory is an established branch of computer science, the Hana Graph Engine has led a rather niche existence in the past, with hardly any practical use cases worth mentioning among SAP's existing customers. This is now changing radically with the rise of SAP Business AI, as SAP is reanimating this technology to solve a fundamental problem of generative artificial intelligence: The dangerous hallucination of language models that simply lack a deep understanding of the highly complex, historically grown ERP databases.

SAP Knowledge Graph and SAP Business AI

The SAP Knowledge Graph, which is deeply rooted in the Hana architecture, is now positioned as an IT revolution and semantic link between the Business AI Platform and the actual ERP applications. From a technical perspective, this knowledge graph models the explicit relationships between business data, metadata and business objects - such as customers, orders and invoices - using so-called ontologies, which define the standardized processing of heterogeneous information in a formal model and map the inherent business context.

Vectors versus graphs

In contrast to vector databases, which are primarily based on mathematical similarity searches and whose relationships often remain implicit, the Knowledge Graph provides transparent and interpretable relationships, which is essential for reliable and explainable AI in a business-critical environment. By integrating graph query languages such as OpenCypher or the RDF syntax SPARQL, SAP also enables a high-performance hybrid query logic in which traditional relational SQL data can be seamlessly combined with semantic graph structures without the need for time-consuming data migration or duplication.

The Knowledge Graph is presented by SAP as the semantic link between the new Business AI Platform and the actual SAP applications of the Autonomous Suite. For the critical existing SAP customer, it is essential to understand that this knowledge graph is intended to make the highly complex, historically grown and often fragmented ERP data readable and retrievable in a structured way for artificial intelligence and autonomous AI agents in the first place. The technical theory behind this concept states that the knowledge graph explicitly models relationships between data, metadata and business objects, thus forming an indispensable semantic context layer. Without this deep, structured contextual knowledge, language models and AI agents would be able to analyze bare table data, but never grasp the business-critical logic and relationships behind it, which would inevitably lead to dangerous and unacceptable AI hallucinations in day-to-day business.

Risky AI bet on the future of ERP

However, if you look at the architectural IT construct through the critical lens of an IT decision-maker, SAP's narrative turns out to be a risky bet on the future. The German-speaking SAP User Group (DSAG), in particular Thomas Henzler, a member of the Executive Board, puts his finger precisely in the wound and warns that in the past it was often extremely difficult for the AI assistant Joule to answer specific questions correctly at all due to the massively complex SAP databases.

According to SAP, the Knowledge Graph should drastically reduce manual data modeling in the future and provide AI models with the business context they urgently need for grounding, but the operational reality is currently still completely different. Independent analyses show that the SAP Knowledge Graph is not yet fully available in the much-vaunted SAP Business Data Cloud (BDC), but has only been announced for the second half of 2026. Until then, the highly praised autonomous AI agents are effectively flying blind semantically, as they lack precisely the essential context layer that is supposed to reliably prevent hallucinations in the SAP Autonomous Suite. In addition, the error-free modeling of such graph structures requires immense technical expertise and flawless data quality, as a knowledge graph based on inconsistent master data will only accelerate the AI's logically incorrect decisions instead of preventing them. For existing SAP customers, this realization clearly means that although the technical ERP vision of a graph-based AI architecture on SAP Hana is academically brilliant and architecturally compellingly logical, it still has to prove that it is more than just another cloudy promise on the road to the autonomous enterprise in the tough day-to-day IT environment.