SAP Says ERP Context Will Make AI Agents Reliable for Business

Christian Klein described SAP’s “autonomous enterprise” push as an attempt to solve a specific weakness in generic large language models: they may be improving quickly, but “they don't know anything about your business data and your business processes.” SAP’s answer is to put enterprise context around AI agents by connecting them to ERP data, which Klein called “the brain of every company.”

The platform claim is not that SAP has a proprietary large language model that customers must use. Klein said customers can use “any commodity LLM” or bring their own preferred models. SAP’s proposed differentiation sits above that layer: agents get access to SAP’s context about how a company runs, including business processes and ERP data fields. That context is what SAP says can make agent outputs “accurate, compliant, and reliable” inside a company.

Klein repeatedly returned to the same point when pressed on competition: the platform’s “heart” is its context layer. SAP’s ERP knowledge is the asset he presented as the company’s “right to win.” He quantified the scope of that context as “over 7.5 million data fields” and “thousands of business processes” inside ERP.

That makes the autonomous enterprise less a claim about replacing business systems than about coordinating them through agents. Klein’s examples described agents that work across functions: a demand agent signaling an inventory agent, which then helps optimize procurement. The larger claim is that SAP can connect front-office activity to fulfillment functions because its platform carries the context of company operations.

When Caroline Hyde asked what data SAP could show customers, Klein cited three operating examples rather than broad productivity estimates.

With JPMorgan Chase, he said SAP showed that the company can close its books “faster now by 30%.” With H&M, he said a personalized agent improved turnover in the company’s commerce shop. He also said inventory was reduced by 10% through agents coordinating with each other: demand signals flowed to inventory optimization and procurement decisions.

SAP’s pitch depends on more than a general AI-productivity story. Klein argued that value comes when agents understand business context and can operate across functions. In the inventory case, the productivity gain was not described as a single assistant making an employee faster; it was described as multiple agents coordinating across demand, inventory, and procurement.

That cross-functional framing is also how Klein positioned SAP against AI tools that sit outside the enterprise process layer. The company is not merely selling an interface to a model. It is selling the ability to use model output in business processes that, in Klein’s examples, connect front-office demand signals to inventory, procurement, and fulfillment.

Bloomberg’s on-screen market panels, separate from Klein’s remarks, placed the discussion against a stronger SAP share-price backdrop: the graphics showed SAP up about 1.9% intraday, up about 43% over one year, and up about 31% year to date.

Ed Ludlow nevertheless put the competitive problem directly: AWS is a partner of SAP’s, but also moving into the same agentic-platform domain. His premise was that SAP is under pressure to offer an all-encompassing agentic platform because “if you don't, someone else will.”

Klein did not reject the premise that the market is crowded. He described AWS as a “great partner” and said he speaks regularly with AWS CEO Matt Garman. His explanation was practical: “not all data of the world sits in SAP systems.” For agents to work reliably, SAP needs to incorporate SAP and non-SAP data into a harmonized layer.

Klein stated the constraint plainly: “agents, they can't compensate for broken data models.” SAP’s Business AI platform is therefore also a data-harmonization effort. Klein said SAP is working with AWS, Microsoft, Nvidia, Databricks, and Snowflake to harmonize business data and feed it into agents.

SAP’s partnership strategy and its competitive claim sit in different layers of the stack. The company is relying on hyperscalers, chip companies, and data platforms to help supply or harmonize the broader data environment. Klein’s differentiation claim is that SAP contributes the enterprise operating context: ERP data, business-process knowledge, and the relationships among functions that make agents accurate enough for business use.

Hyde sharpened the same tension by calling the ecosystem “frenemy” territory and asking whether accuracy is what wins customers over amid “software anxiety.” Klein’s response stayed consistent: accuracy comes from context, and context comes from ERP.

Ludlow invoked Nvidia CEO Jensen Huang’s formulation — “More compute, more tokens, more revenues” — and asked whether SAP had customer cases showing real revenue from agentic work.

Klein answered with Nvidia and supply-chain automation. He said Huang appeared with him during the keynote and wished SAP luck, first as an SAP customer and then because higher compute and chip consumption require “consumption at the top where the real value creation happens.”

The example Klein gave was an “autonomous supply chain for Nvidia.” If SAP can build that, he said, Nvidia sees it immediately in hardware consumption. In Klein’s framing, SAP’s agentic business applications sit where enterprise AI use is meant to become operational consumption: the business-process layer that can create demand for compute.

Nvidia’s role in the answer was both customer and partner. Klein said Nvidia and SAP are partnering to build the autonomous enterprise, and he placed SAP’s role at the application and business-process level, where AI is meant to affect closing books, commerce turnover, inventory levels, procurement, and supply-chain execution.