What SAP’s TechEd Story Says about Enterprise AI’s Next Phase

SAP has quietly rewritten its AI story. At TechEd Berlin 2025, the company set out its revised AI strategy: a process-first bet built on data fabric, agents and a long-term view of how autonomy should be governed. Instead of chasing the latest large language model headline, SAP believes that the real battleground is where AI, data and business processes live in the enterprise stack.

CCS Insight’s new report, SAP Chooses Its AI Lane: Process-First Apps, Data and Agents, digs into this strategy in depth, examining SAP’s architecture choices, agent strategy and competitive positioning and testing it against our survey findings on AI, automation and everyday work. In this blog we outline several important highlights from the report.

From Theatre to Signal

TechEd Berlin certainly embraced theatrics. Robots patrolling plants and humanoids walking warehouse aisles make for good video content, as do quantum logistics pilots in partnership with IBM. The important point to note was that this was all wired through SAP’s technology stack rather than a separate lab environment.

The quieter stories gave more telling signals about SAP’s AI progress: a significant S/4HANA migration where Joule, SAP’s assistant, supports thousands of users; and a purchase process reworked with SAP Build so that staff spend less time correcting bad requests and more time on supplier strategy. These are precisely the types of high-volume, low-glamour processes where AI and automation should first earn their keep.

The open question – and one we explore in our report – is whether SAP can turn these examples into repeatable patterns that customers can actually adopt, or if they are just isolated showcases.

Ambition Is High, But the Basics Still Matter

In our Senior Leadership IT Investment and Employee Workplace Technology surveys, the same picture keeps emerging: ambition for AI is high, but current support for the basics is patchy.

Most organizations say they are piloting generative AI. Most employees say they use it in some form at work. Yet only a minority of companies have a joined-up automation strategy, and even fewer say their IT and business strategies are fully aligned. AI is often layered onto fragmented processes and inconsistent data.

This is the backdrop for SAP’s process-first message. By pulling AI back into core applications and a shared data fabric, SAP aims to position itself as the place where ambitious AI projects can be grounded in real workflows, governance and data quality. Whether it can deliver this in practice is another matter. However, the narrative reflects what buyers say they need, which is fewer experiments and proof that AI can improve collections, supply chains and asset performance without upending security and compliance.

Data Fabric as the Real AI Play

SAP’s strongest card is not the assistant on the screen; it is the data and semantics underneath. HANA Cloud is evolving into a multimodel engine that can store transactions, relationships, documents and vectors in one place. Business Data Cloud sits above as a “business data fabric”, offering predefined data products for SAP applications and tools to blend SAP and non-SAP sources, with built-in governance.

SAP is also starting to surface these capabilities in external data tools, rather than insisting that everyone comes to its own catalogue. The strategic bet is simple: years of work on structured business data and process models can now be turned into an AI advantage. Instead of saying “we have a big model too”, SAP argues that it has already done the hard work in semantics and is now using that foundation to power prediction and automation.

For a technology player that often sits behind the hyperscale cloud providers in perceived AI innovation, this is a more credible path than trying to outspend them on generic models.

Agents on Duty, Not on the Loose

The other big pillar in the TechEd narrative was agents. Joule is positioned as a family of role-based assistants – for collections, planning, maintenance and more – backed by an agent platform that prioritizes governance as much as cleverness. SAP also talked more than in past years about user experience, casting Joule as a new, more human-centred front door to SAP rather than another awkward overlay on top of old screens. The focus was as much on how work flows – who sees what, when – as it was on fresh user interface components.

Solutions like Agent2Agent contracts, LeanIX Agent Hub and Signavio agent mining all point to the same goal: agents should be catalogued, observable and controlled, not just dropped into production because they make a good demo.

This is much closer to where most enterprises really are. In our report Scaling Agentic AI: Foundations, Partners and Next Steps, we see organizations’ AI strategies clustering around “smart assistants” and “agents with approvals”, with only cautious moves into creating more autonomous co-workers in tightly bounded domains. Open-world autonomy is still essentially a robotics and research topic.

SAP also takes a more critical line than many on popular frameworks such as Model Context Protocol (MCP). Customers are free to experiment, but SAP will not expose MCP directly through its products owing to concerns about security and scalability. The company argues that serious agent orchestration needs knowledge graphs, policies and routing, not just a flat list of tools. This may not be everyone’s view, but it is at least a thoughtful stance in a market full of hand-waving.

Strengths, Gaps and the Next Test

What does this tell us about SAP and enterprise AI more broadly?

On the positive side, SAP is finally playing to its structural advantages. It has deep process coverage, long-honed business semantics and a large developer community. At TechEd Berlin, those assets were front and centre in the AI story, not buried in breakout sessions. The process-first, data-led narrative fits a market where executive boards want integration, automation and guardrails.

Many industries use SAP, but the implications are particularly clear for telecom and communication service providers. Many already run SAP at the heart of finance, partner settlements, logistics and field operations, even if network and service platforms sit elsewhere. A process-first, data-led AI stack that sits across ordering, assurance, workforce and supply-chain flows is highly relevant in a market where margins are tight, partner ecosystems are growing and board members are asking hard questions about automation and service quality, not just new network features.

On the negative side, SAP’s AI stack is heavy and some foundations are still thin. Identity and policy at the action level – which agent is allowed to do what, in which system, under which conditions – remains underexplained. A lot of the value is still only modelled in calculators rather than evidenced at scale. And the role of partners in making this deployable and supportable is not yet as clear as the architecture diagrams.

From an analyst’s point of view, SAP has chosen its AI lane, and this time the story hangs together. The next test is not another round of TechEd demos. It is whether, in the coming year or two, SAP and its ecosystem can turn assistants, data fabrics and agents into measurable, repeatable outcomes across a broader set of customers. That is what budget holders and employees will expect to see before they call this a success.