SAP bets on AI governance to secure its profit margins

SAP highlights AI governance as a crucial lever for securing companies' profit margins. According to the German giant, replacing statistical estimates with deterministic control can reduce the accuracy gaps often observed in consumption models. Manos Raptopoulos, Global President of Customer Success for Europe, APAC, the Middle East, and Africa at SAP, emphasizes the importance of this precision, stating that the gap between 90% and 100% accuracy is critical for operations.

As companies integrate large language models into their production environments, Raptopoulos notes that evaluation criteria have expanded to include governance and tangible business impact. He underscores that the transition from passive tools to active digital agents is a key moment for governance, a topic that will be addressed at the AI & Big Data Expo North America.

Agentic AI systems, capable of planning, reasoning, orchestrating with other agents, and autonomously executing workflows, pose governance challenges. Raptopoulos warns that these systems, if not managed like a human workforce, can expose organizations to serious operational risks. He compares this situation to the shadow IT crisis of the last decade, but with higher stakes. Establishing agent lifecycle management, defining autonomy limits, enforcing policies, and instituting continuous performance monitoring are mandatory requirements according to Raptopoulos.

Integrating modern vector databases with legacy architectures requires significant engineering investments. Teams must limit the agent's inference loop to avoid costly errors, which increases computing costs. Raptopoulos identifies three major issues to resolve before deploying agentic models: accountability for errors, decision verification, and human escalation thresholds.

Geopolitical fragmentation complicates these challenges, with varying regulatory requirements in key markets like New York, Frankfurt, Riyadh, and Singapore. Sovereign cloud infrastructures, AI models, and data localization mandates are regulatory realities in these regions. Raptopoulos believes that integrating deterministic control into probabilistic AI is imperative for leadership.

AI systems depend on the quality of data and processes, which Raptopoulos refers to as the "data foundation moment." Fragmented master data, siloed business systems, and overly customized ERP environments can lead to erroneous recommendations, severely impacting operations. To extract tangible value, models must be anchored in proprietary enterprise data, thus surpassing generic models.

Finally, interaction with enterprise applications is evolving toward generative user experiences. Employees will express their intentions to the system, which will orchestrate the necessary workflows. Raptopoulos gives the example of a user asking the software to prepare a briefing for their highest-revenue client visit this week. However, employee trust in these systems will depend on adherence to governance limits and demonstrable productivity gains.

The financial return on AI manifests most quickly during customer interactions. Training models on proprietary records and internal rules creates client-specific intelligence that is difficult for competitors to replicate. Deploying enterprise intelligence requires leadership to orchestrate three distinct layers in parallel, which Raptopoulos defines as the "strategic moment." The first layer involves integrated features for quick returns, the second requires agentic orchestration, and the last focuses on industry-specific intelligence. Poor sequencing can leave immense financial value uncaptured, while jumping to sector-specific applications without data maturity multiplies risks.