SAP CFO Urges Enterprise AI Shift Beyond Chatbots to Core Business Processes for Real ROI

SAP CFO Dominik Asam recently stated that the enterprise adoption of artificial intelligence, including Large Language Models (LLMs), needs to evolve beyond current "low-hanging fruit" applications like chatbots and coding assistants. Speaking after SAP's second-quarter results, Asam highlighted that while these applications consume a significant portion of AI tokens today, their impact on core business processes is limited. He argued that the real productivity gains and returns on investment will come from embedding AI into more complex, critical functions such as finance and supply chain management. This perspective from a major enterprise software vendor's finance chief is crucial for practitioners. It underscores a growing realization that the initial hype around general-purpose LLMs needs to be tempered with practical considerations for enterprise-grade deployment. For DevOps teams, cloud architects, and AI engineers, this means a shift from rapid prototyping of conversational agents to a more rigorous approach focused on data quality, model reliability, and stringent cost management within existing business workflows. The risks associated with AI hallucinations, which are tolerable in low-stakes applications, become unacceptable in financial or supply chain systems where errors can compound across multiple steps and lead to significant compliance and operational issues. The broader trend in enterprise AI has been a rapid experimentation phase, driven by the accessibility of powerful foundation models. However, as organizations move from proof-of-concept to production, the challenges of integration, data governance, and achieving measurable ROI become paramount. This sentiment from SAP aligns with a growing industry consensus that generic LLMs, while powerful, are not a panacea. Instead, specialized, fine-tuned, or domain-adapted models, coupled with robust data strategies, are proving more effective for specific business problems. The emphasis on "governed systems embedded in specific business processes" reflects the maturity curve of AI adoption, moving from broad exploration to targeted, value-driven implementation. This also echoes the ongoing discussions around responsible AI and the need for explainability and control in high-impact scenarios. Practitioners should prioritize building strong data governance frameworks and pipelines to ensure clean, reliable data for AI models. The focus should be on identifying specific, high-value business processes where AI can deliver clear, quantifiable benefits, rather than deploying LLMs indiscriminately. This may involve exploring smaller, more specialized models or fine-tuning larger models with proprietary data, rather than relying solely on the most advanced, and often most expensive, frontier models. Furthermore, a critical evaluation of the cost-effectiveness of AI solutions, considering token consumption and infrastructure, will be essential. The "cheapest reliable tool that can deliver the required outcome safely" will often be preferred over the most powerful, general-purpose model. Organizations should also invest in developing internal expertise to manage and monitor AI systems, ensuring they meet the "excruciating assurance levels" required for critical enterprise functions.