Making AI real for business success

The Wi-Fi password at the SAP TechEd conference in Berlin this week encapsulated the new mission of the global leader in enterprise resource-planning software: GetReal2025. 

The timing could hardly have been more pointed. Across the world, executives are losing patience with AI experiments that overpromise and underdeliver. Boardrooms have heard enough about pilots and proofs of concept, and now want systems that improve margins and forecast outcomes.

TechEd took place in that atmosphere, amid assurances that performance could replace promise. SAP used the event to demonstrate how deeply AI now runs through its own operations before unveiling its next leap forward.

Rather than another round of hype about possibilities, SAP aimed to show a working example of AI at scale, handling everyday complexity inside one of the world’s largest software organisations.

SAP chief technology officer and chief AI officer, Philipp Herzig, told Business Times at TechEd that the clearest evidence of maturity came from within SAP itself. The company’s AI assistant, Joule, acts as a conversational layer across its software stack, connecting data, applications, and agents to automate tasks and surface insights on demand.

“If you look at AI at scale, what is really real and what is working very well, just look at Joule,” he said. “It’s used by more than 30,000 employees every month, about a third of the SAP workforce. We have more than 100,000 policies and documents in different languages, and depending on where you work, in Brazil or South Africa, you get the correct HR or travel policy surfaced to you.”

Herzig said Joule had become the company’s single interface for daily tasks. “You can do your expense reports, indirect procurement, and financial tasks all in one place. It works, and it works at scale. We have a thumbs-down rate of only 1%, which is phenomenal when you think about the size of the company.”

Innovations across SAP’s unique flywheel of applications, data and AI put developers in the driver’s seat — Muhammad Alam, SAP executive board member

That success set the stage for TechEd’s central announcement: an AI model called SAP-RPT-1, short for relational pre-trained transformer. The model introduces a new class of AI: the enterprise relational foundation model. It interprets structured business data and the relationships within it, mapping how orders, invoices, logistics, and payments interact to forecast what comes next.

SAP described it as a model that “can make fast and accurate predictions for common business scenarios like delivery delays, payment risk or sales order completion”. Instead of producing text, it reads how data behaves across systems, turning business logic into predictive insight.

Herzig said SAP-RPT-1 marked the transition from incremental automation to full predictive architecture.

“We see a shift from what I call a cloud-native architecture to an AI-native architecture, as AI becomes an ever-increasing part of the software stack. “What we wanted to solve are the problems where we have a reason to solve them: because we have the data, the relational data, the structured business data and so on. We set out this research project two years ago, talked a little about it, but now it actually becomes a reality.

“It’s a shift that needs several things to come together: the knowledge graph, this predictive model now with RPT-1, and of course the large language models. So there are many elements in the software stack that change. Every day, each little piece adds to this picture and solves a particular challenge in the stack.”

Herzig said the greatest challenge lay in making this intelligence work at enterprise scale.

“Anyone can do a demo. Getting it enterprise-ready at scale — that’s the tough challenge. That’s why we’re solving one problem after another, each for a specific outcome in the overall stack.”

SAP executive board member Muhammad Alam tied this philosophy to the developer community.

“SAP’s announcements give developers the tools they need to deliver at the speed of AI,” he said.

“Innovations across SAP’s unique flywheel of applications, data and AI put developers in the driver’s seat.”

That “flywheel” anchored the narrative of TechEd:

• applications generate data;
• data trains predictive models;
• models return intelligence to the applications.

Each loop strengthens the next, creating a flywheel effect of acceleration of innovation. SAP announced that it would equip 12-million individuals worldwide with AI-ready skills by 2030 through a partnership with Coursera that provides hands-on certification in SAP’s ecosystem. The goal is to align those skills with the AI-native architecture now taking shape inside the company. It also sends the message that people remain at the heart of the AI journey.