Episode Summary
Executive Summary: SAP CTO Philip Herzig argues that SAP’s durability comes from solving end-to-end enterprise outcomes at scale, not just selling software. He frames AI as a business-model and operating-model transition spanning UI, processes, and data, while stressing that enterprise adoption depends on verifiable outcomes, secure integration, and harmonized data—not raw model capability alone.
Main Topics: SAP’s core role as enterprise operating system (Priority: 5/5): Herzig explains SAP as the backbone for finance, HR, supply chain, manufacturing, logistics, procurement, sales, and service—essentially the operating system of large companies. Why SAP has endured through technology cycles (Priority: 5/5): He says SAP survives because customer needs for outcomes, scale, and standardization persist across mainframe, client-server, internet, mobile, cloud, and now AI transitions. AI re-architecting SAP across UI, processes, and data (Priority: 5/5): Herzig describes AI as changing the user interface (generative UI), business processes (agentic workflows), and data layer (harmonized semantic data models). Scale, evals, and enterprise reliability (Priority: 5/5): A central challenge is making AI behave correctly across massive, heterogeneous enterprise environments; he emphasizes evals, boundary conditions, and agent mining to capture decision traces. Predictive analytics needs more than LLMs (Priority: 4/5): He argues LLMs excel on unstructured tasks, but enterprise forecasting and classification still require specialized predictive models; SAP’s RPT-1 is positioned to bring transformer-like benefits to structured data. Enterprise adoption barriers: data fragmentation, security, and trust (Priority: 4/5): Herzig highlights disaggregated data, legacy integrations, security risks, and the need to safely operationalize AI as the main blockers to adoption. Business model shift toward consumption and outcomes (Priority: 4/5): He says AI pushes SAP away from pure seat-based licensing toward hybrid consumption and eventually more outcome-based pricing, depending on customer readiness and verifiability.
Key Arguments: SAP is durable because it standardizes and runs critical end-to-end business processes that customers still need regardless of technology cycle. AI should be judged by business outcomes, not novelty; SAP’s job is to make technology disappear and deliver ROI. Enterprise AI is harder than demos because scale, personalization, compliance, and integration make simple prototypes insufficient. Evals and explicit success criteria are becoming as important for agents as tests are for code generation. Most enterprise AI value will first come from unstructured workflows, but high-value operations require blending structured and unstructured data. LLMs are not sufficient for forecasting, demand planning, cash prediction, and other tabular problems; specialized predictive systems remain necessary. Capturing human decisions and agent traces creates a feedback flywheel that improves future AI behavior and standard operating procedures. SAP expects AI to change pricing from seat-based software toward hybrid consumption and, later, outcome-based models where possible.
Data Points: Enterprise customers: approximately 400,000 - SAP’s customer base described by Herzig Consulting effort reduction with Joule for Consultants: 30% - He says the product can reduce consulting effort in complex SAP projects Countries in one pharma example: 90 countries - Illustrating the scale of predictive-model deployment complexity Models in that pharma example: 180 models - 90 countries times two models for payment delay prediction Forecast improvement example: 3%–4% - He says improving demand forecast accuracy by this amount can create multi-million-dollar value AI transition layers: 3 - Herzig breaks AI change into UI, business processes, and data layers SAP founding year: 1972 - He references SAP’s origin as a response to non-scalable custom implementations Time horizon mentioned for AI adoption gap: the gap between innovation and outcomes is increasing - He cites a Gartner-style framing of AI innovation race vs outcome race
Pivotal Quotes: "Our job at SAP is to make the technology disappear." — Philip Herzig: On what differentiates winners in enterprise software: outcomes over technology for its own sake "AI is only as powerful as the data is." — Philip Herzig: On why data harmonization and semantic models are foundational to enterprise AI "This is actually the problem of scale." — Philip Herzig: On why simple AI demos break down when applied to real SAP customers with many systems, countries, and policies
Implications: Enterprise AI winners will be those who combine trusted data, secure integration, and measurable outcomes. SAP’s future depends on turning AI into reliable workflows and predictions, not just chatbots, and may accelerate a broader shift to consumption and outcome pricing.