Project Study | SAP SE | Graph-Grounded AI versus AI-Only ApproachesBeta

Project Study: Auditable by Construction? An Empirical Study of Knowledge-Graph-Grounded AI versus AI-Only Approaches in Financial Accounting About SAP SAP SE, a global leader in the field of enterprise software solutions, specializes in providing sophisticated Enterprise Resource Planning (ERP) systems. With presence in over 180 countries, SAP ERP systems assist businesses in streamlining their diverse operational processes, from finance and HR to supply chain and sales. The SAP ERP system is distinguished by its ability to integrate different business functions into a single realm, enabling organizations to operate with improved efficiency and process standardization. This system's capability to coalesce information management across various business functions makes it a preferred choice for businesses of all scales and scopes. SAP's emphasis on process integration and optimization not only simplifies day-to-day business operations but also provides strategic insights to support decision-making and foster growth. As such, SAP continues to be at the forefront of facilitating businesses in navigating their processes towards operational efficiency. Introduction Artificial intelligence is entering the finance function - but can AI-generated accounting be trusted? For CFOs, auditors and regulators, the decisive question is not whether AI can produce a plausible journal entry, but whether the resulting record is reliable, reproducible and auditable. Large language models are probabilistic: first exploratory runs in our research setting show that even a frontier model, at temperature zero, returns different results for the same operating-profit computation across identical calls. For an audited financial statement this is disqualifying - unless the architecture around the model changes. SAP and TUM (Chair of Financial Accounting, Prof. Dr. Jürgen Ernstberger; Chair of Management Accounting, Prof. Dr. Max Blaschke) are running a joint research collaboration on “auditability by construction”: an architecture in which a knowledge graph (ontology) holds the authoritative accounting rules and a deterministic engine derives every posting, while the language model is confined to interpretation and explanation - it never makes the accounting determination. Its centerpiece is a preregistered empirical study - the Differential Determinism Study - testing whether the posted accounting record is invariant to model choice, model version, prompt phrasing, temperature and adversarial user pressure when it is grounded in a knowledge graph with deterministic execution - and whether AI-only approaches, even with best-practice retrieval, fail this test - executed on a realistic model company across fixed assets, foreign-currency valuation and revenue recognition under IFRS. Objectives The experiments are owned and led by Prof. Ernstberger and Prof. Blaschke; your team supports the study end-to-end and develops its interpretation: 1. Planning and preparation. Understand the preregistered study design (four experimental arms, a 45-item test bank, systematic perturbations of prompt, temperature, model version and adversarial pressure) and help finalise run plans and data-collection templates. 2. Execution support. Support the systematic execution of the experiment arms, structured logging of results (JSON result sets, run metadata), and the version and freeze discipline that makes the study itself auditable. 3. Documentation and evaluation. Prepare the results systematically: variance and drift metrics across models and versions, escalation versus confabulation behaviour, and observations and anomalies worth reporting. 4. Interpretation. Assuming the results confirm the hypotheses — and equally if they do not — what do they mean? For auditors and audit standards, for CFOs deciding whether to let AI near the ledger, and for the design of enterprise accounting systems. 5. Vision task. Put yourselves in the shoes of a CFO or a Chief Accounting/Controlling Officer and articulate your wish towards SAP: given the latest technological advancements, what should the future finance and accounting ERP system look like? What is its role and function, how much autonomy should it have, how much control do you retain, what does it do — and deliberately not do — and how do you interact with it? Your profile • Team of 2 to 5 students in Bachelor or Master studies enrolled in Management & Technology • Excellent analytical and problem-solving skills • Agile and flexible mindset with strong communication skills and German and English • Motivated, independent, and target-oriented work approach Your benefits • Work independently in a project-based setting • Receive continuous support by SAP experts • Get first-hand relevant contacts for a career entry