SAP CEO: the AI race is being fought in the wrong place

The enterprise AI race is quickly becoming a contest over interfaces. Every week brings another announcement about smarter copilots, more capable agents, or new orchestration layers designed to automate work across the enterprise. The progress is undeniable. But much of the market is not optimizing for how businesses operate. That distinction is more important than many realize. Because enterprises do not run on prompts. They run on execution. A global manufacturer deciding how to reroute inventory during a supply chain disruption needs more than simply an answer. It must evaluate supplier alternatives, inventory availability, customer commitments, and financial tradeoffs simultaneously. A CFO forecasting liquidity exposure during market volatility needs context that a simple chatbot interaction can’t provide. These are interconnected operational decisions shaped by dependencies, preferences, approvals, financial consequences, and tradeoffs that ripple across the business in real time. In countless conversations I’ve had with executives over the past year, the discussion inevitably shifts from AI capability to operational reality. The models are improving quickly. The harder question is whether AI understands the business environments it is operating within. Today, too much of the AI conversation still assumes that better models alone will produce better business outcomes. They will not. Enterprises are discovering that intelligence disconnected from operational context – the processes, the data, the rules and policies that govern and protect your organization – can generate activity without creating much progress. In some cases, it can create more fragmentation and risk. A generated recommendation may sound convincing while missing critical dependencies elsewhere in the system. An AI agent may automate one workflow efficiently while disrupting planning assumptions in another. Enterprises do not suffer from a shortage of AI outputs. They suffer from a shortage of AI