SAP’s Christian Klein and Marielle Ehrmann

SAP Sapphire is the splashiest SAP show – and few Sapphires were splashier than Orlando 2026. This year, SAP staked its claim to the so-called “autonomous enterprise” – with much exuberance. 

But here at the line-of-business event known as SAP Connect, things got real.

The SAP Connect keynote was vastly superior to Sapphire’s (hear our live review if you want). It was mercifully concise, showcasing a confident SAP that didn’t need long-winded explanations of why they aren’t a throwback system of record anymore. 

The onstage customers radiated industry authority, and shared their successes on their SAP’s AI pursuits to date. 

Though the news perhaps paled against Sapphire, SAP Connect’s typically elaborate news guide underscores that SAP’s major Joule-related announcements – including Joule Work – are now generally available to customers. Then there was SAP’s intent to acquire TechWolf, which opens up new angles in SAP’s context-graph-for-work ambitions. 

On day one, SAP came out swinging against “generic models,” and rightly so. For SAP to succeed in enterprise AI, it must convince customers that its contextual approach to AI, with embedded governance and “zero trust” agentic security, is superior to what frontier models (and their Forward Deployed Engineers) are pitching. But that was just one of the hot topics that SAP addressed at SAP Connect. Two full days into the show, I’ve now had the chance to press executives one-on-one, and attend open sessions as well. 

Here are the notable answers to what I consider the hottest topics at the show. My picks are inspired what I hear from customers, as well as from user group dialogue with ASUG, UKISUG and DSAG. 

How will SAP differentiate in AI? SAP CEO Christian Klein pushes back against generic LLMs

How will SAP differentiate in enterprise AI? I have my own take – and we should keep in mind that differentiation is both what you do, and what you choose to emphasize on stage. To open the SAP Connect press Q/A, Klein acknowledged that SAP is hammering on context, as well as AI governance. 

SAP’s agentic AI moves are all about embedding AI in mission-critical/industry processes – a higher stakes game where accuracy and auditability are non-negotiable. The context required to pull this off will be the focus of my next piece, but it definitely includes SAP’s knowledge graph, drawn from generalized SAP system tables and core metadata views across SAP systems. Klein alluded to that knowledge graph as a vivid contrast to other vendors:

Coding features is different than building an ontology. When you talk about 7 million fields and half a million tables, that needs to find its way into an ontology. That’s what we are pushing. That’s why now Joule Work can solve things. We can also get tasks done which others can’t do.

Post-keynote, Klein doubled down on context and governance: 

Then Marielle [Ehrmann] came and said, ‘Hey, but it’s also about important about trust. Where does your data reside? How do we also manage and guard AI agents? We talk to many IT leaders these days. Of course, everyone experimented with AI, but every custom agent you need to do afterwards, you need to also run them and govern them. When you’re doing business in 60 countries in the world, have fun with all the tax regulations. Have fun with all the governance requirements, which is sitting in SAP. So it’s the context, it’s the governance. 

SAP’s big job? Convince customers that their industry AI path is with SAP, not with a slew of startups and AI behemoths. Whatever customers decide, SAP’s talking points are getting very clear now.

The TechWolf acquisition – more context graphs for SAP. “Workforce intelligence” is the goal

Speaking of context, SAP’s plans to acquire TechWolf are another proof point. Note the mention of TechWolf’s context graph, as well as the use of external data. (Often, it’s the external data that nudges an AI agent from useful to a different level of effectiveness). 

During the press Q/A, Rebecca Carr, SAP’s President of HCM Products (and CEO SmartRecruiters), explained how TechWolf’s context graph will impact the “skills journeys” of SAP customers: 

TechWolf has actually built a very strong context graph for work that decomposes every single job into a set of tasks. They then map that with skills that people have, and they can infer a level of potential as someone moves across an organization. So you can imagine how powerful that could be in the context of SAP. 

We have a lot of business signals. We have a lot of people signals, and so pulling those together into that context graph for work would, in theory, help our customers answer very practical questions: Where can I redeploy talent? How can I reskill this person within my business? And we want to be doing that right in the flow of work for our users.

During my sit down with Manoj Swaminathan, President and Chief Product Officer of the SAP Autonomous Suite he added more. After citing the disclaimer that this is still formally an intent to acquire, he said:

That is the core foundation of that context graph, and that is the model that they have… We will bring that as a first-class context model into SAP SuccessFactors, so that we will provide that workforce intelligence right out of the box for all of our customers.

Some clarity on AI monetization and pricing – but questions remain 

One press conference standout? Hearing Klein discuss SAP’s paths to SAP monetization. Sometimes, we get too hung up on token costs, and the implications of consumption models. Klein brings the debate on AI pricing back to value. He contends that if SAP’s AI solutions have real industry impact, customers will have no problem paying for the value delivered: 

 Use cases like Jan [Gilg] showed, where you can compare customer demand and optimize your supply chain, look into your inventory. When agents are doing that, then of course this has value, and then customers are willing to pay for that – because that is done today by a lot of people. AI already has realized all the signals. So AI can do it even better, more intelligently. 

Klein’s sees opportunity to monetization across platform services, migrations, and Signavio/tool chain/transformation services. He also sees that monetization potential in Industry AI. He cited the hypothetical examples of helping a customer build an AI layer for manufacturing, last mile delivery, or return claims management. One thing is certain: if you want to charge for agentic AI, customers must see the value in it: 

But always, of course, the customer sees the value and then pays based on the consumption, and how well they adopt the the agents we are delivering with Joule Work, obviously. 

Like virtually all vendors, SAP talks frequently about outcomes. Why then, asked an an analyst, does SAP price around consumption? As Klein explained, customers can withdraw from consumption agreements if they aren’t seeing results. So in a sense, there is a clear outcome there. (Looking further ahead, SAP has said that core AI services are likely to be included in the software license, while customer will pay in “AI Units” for premium SAP AI services, which will typically be the services where agentic actions are invoked – but this is not SAP’s official policy statement, just my understanding as of today). 

Bringing customers along – not just “the keynote luminaries”

The SAP Connect opening keynote was strong from a customer interview standpoint, but one analyst asked SAP: how do you bring all of your customers along, not just the “keynote luminaries” that spoke effectively on stage? Klein responded: 

We made a big commitment at Sapphire. We said, ‘Hey, we don’t leave our on-premise customers behind, so there will be APIs and MCP servers delivered.’ But still, of course, we need to connect. You need to [join] with us on this journey to the cloud, because everything that we deliver just in the last three months is enormous. 

Klein does issue a challenge to on-premise customers: AI point solutions for customized on-premise systems aren’t going to scale. Customization could be an option, but I would ask: is that going to be sustainable for customers, financially and otherwise? Klein: 

When it comes to ontology and the governance of AI, we can’t, of course, do this point-by-point into every on-premise installation, which is also probably where we customize. 

Ergo, SAP is hitting on interoperability more than I’ve heard in the past: 

But we have solutions for those customers, and also on the data layer. Manoj showed really well that, especially in the manufacturing area, you probably will need 40% of the data that sits in SAP, but 60% sits in non-SAP. So extensibility is key. 

Can SAP help by making data context management easier for companies, especially in the midmarket? 

With Reltio, you can do zero copy. Then we build the ontology out of the box. We don’t need to rebuild the ontology. A mid-sized company doesn’t have the IT budget to do this over and over again. It’s good for predictive asset management when there is an ontology layer already available.

Zero trust security is in, Doomsday is out

For enterprises, AI safety and Doomsday hyberbole is a distraction – but getting security right is job one. Even within that narrower/practical AI security frame, there are misconceptions. But SAP went further that most, devoting a key section of the opening keynote to agentic security hype – and reality. 

At the press Q/A, Marielle Ehrmann, SVP and Chief Security Officer, SAP SE hit on the crucial point:

Agents themselves are not the problem. It’s always about the unmanaged, the ungoverned agents that are creating the the problems.

During the keynote, Ehrmann said that boardrooms went from: ‘How can we deploy AI’ to ‘How do we make sure we’re not in tomorrow’s headlines?’ Ehrmann was asked: Did she see a stall of some projects? What is she hearing from customers? Ehrmann: 

We are seeing agents going rogue across the industry where governance is missing. And I want to be very clear: Agents themselves are not the problem. It’s always about the unmanaged, the ungoverned agents that are creating the the problems. This is exactly where our customers then come to us and say, ‘What are you putting in place to create that kind of safety layer for us? Why can we trust SAP, and why should we trust SAP that this is not happening on on your side?’

Ehrmann doesn’t just emphasize security – it’s about an overall risk management framework. Aligned with regulations like the EU AI Act, Ehrmann’s team put three baseline frameworks in place:

I mentioned that in the keynote also, with the three layers of risk definition, so that we have human in the loop, the human on the loop, but also the human then to pull the plug if anything specific happens, which is not meant to be used in high risk scenarios. We are looking at the NIST AI risk management framework, which is exclusively popular here in the U.S. And we are one of the large enterprises certified by the ISO 42001.

My take – lingering questions and on-site debates, an open discussion

Many vendors have lost touch with the art and necessity of a on-site press conference. SAP facilitates these with aplomb, including online questions from global media. Why does that matter? Because what executives tell press/analysts collectively carries far more weight, in my view, that sidebar conversations. Other vendors should take note. If you can’t handle open questions, how you can cultivate Ai trust with customers? 

This does not mean SAP effectively answered all questions, but: there was much more clarity here than last spring. SAP did its homework since then, and it shows – particularly in how customers articulate their own evolution. I’m out of time now, but I could cite numerous examples where customers told us that their ERP modernization efforts were not a distraction from SAP Business AI, but a necessary precursor. If SAP’s own customers embrace the data and process discipline of ERP modernization, along with LOB components like HCM or CX, then SAP’s AI play is looking strong. 

I mean that in two ways: first, that’s how Klein’s message on context and governance really comes to fruition: driven by a modern data/process foundation. Second: that would mean customers are resisting the generic LLM approach Klein was pushing back against. 

That does not mean SAP has great answers to every question. Example: Klein has asserted “the end of three year migrations,” and singled out SAP’s own AI Migration tooling as well. But while customers have better implementation stories overall, I hear more about “the end of three year migrations” from SAP than I do from customers. SAP would do well to reframe migrations not just in terms of speed, but deriving impact all along the way, as customer John Boos & Co. did, by obtaining process improvement benefits with Signavio – even in the early migration stages. 

Given this is a show about end-to-end processes, we could stand to hear more about CX. Building AI agents on an end-to-end platform is simply a better way – for now – than elaborate agent-to-agent scenarios. SAP should consider framing SAP Connect keynotes to start with customer demand signals, all the way through the supply chain and logistics aspects. 

SAP has many of these capabilities, but why not start with cross-channel engagement as the spark for process change? That would be a refreshing way for SAP to lead off this event. (Analysts had hot debates on how much SAP’s CX emphasis has evolved, or not – but I prefer to think more about this from the customer/attendee perspective. Why not start with the demand signal and go from there?).

As readers know, I’ve been grilling SAP pretty hard for not bringing AI innovation to SAP customers on older releases. Here, SAP now has much more aggressive and clear answers, involving giving ECC customers access to API/MCP services, as well as the BTP options for AI integrations/app building. There is also the option for customers to get in deeper now, if they commit to a Rise transformation in the future. Not all customers will consider these perfect answers, but they are much clearer (and openly discussed) than before. 

The same goes for the on-premise versus cloud issue in general. Though SAP is firm that cloud is the go-to-platform for SAP AI services, on-premise customers can utilize hosted environments, with all the Joule bells and whistles, on the condition that SAP manages those hosted environments via Rise. Again, not necessarily a perfect answer, or one that will work for all on-premise customers, but definitely a big improvement over the last couple years when these questions were asked. It’s a reflection of a more diverse approach, one that takes into account the need to help customers get started on SAP AI with their data sovereignty needs in mind, etc. (I get into these topics with a diginomica reader in the comments here).

It’s also a push into an AI co-innovation mindset, which is what vendors should cultivate with customers. Don’t make customers sign a big migration contract – just start building AI with them now. As trust builds, and the need for reliable context and scale deepens, the migrations will come. And when they come, they won’t just be obligatory lift-and-shifts, but strategic moves. 

We are also seeing that play out in SAP’s Industry AI co-innovations, an important topic I’m out of time for here, but check SAP and PwC’s co-innovation on an agentic Billing Assistant for one example. 

Next up: I’ll have more from a deeper dive I did on-site into SAP company memory, and SAP’s progress on AI context, including the SAP Knowledge Graph and other customer-specific context as well.