I was in the room for the press Q&A straight after the keynote, and I got to ask SAP’s leadership team where this all goes from here. More on that further down. First, my recording of the keynote so you can watch it yourself, after which this article continues, please read on!<\/p>\n<\/p>\n
The three-eared unicorn problem<\/p>\n
Klein opened the keynote with a story about his daughter. He uses AI to draw her pictures, and she’s perfectly happy with a unicorn that’s roughly 80% accurate, even if it has three ears.<\/p>\n
“At least it doesn’t have two horns,” Klein joked from the stage.<\/p>\n
But then the punchline.<\/p>\n
“In business, it’s different. If AI agents run your payroll, your financial close, or do the demand and supply chain planning, 80% accuracy is just not good enough. When you run the world’s most mission-critical businesses, your AI agents they should not guess. They should deliver accurate, reliable and compliant outcomes,” said\u00a0SAP CEO Christian Klein<\/a>.<\/p>\n That framing did great work in the keynote. Klein cited a recent Stanford AI survey suggesting almost every company is now using AI, but most see little value. His diagnosis: the public LLMs everyone is excited about aren’t trained on your business data, your processes, your governance rules or your authorisation model. They guess. Which is fine for unicorns, less fine for a financial close.<\/p>\n His fix: glue the ERP brain to the AI.<\/p>\n The SAP Business AI Platform, explained without the marketing fog<\/p>\n SAP is merging\u00a0SAP Business Technology Platform<\/a>,\u00a0SAP Business Data Cloud<\/a>\u00a0and SAP Business AI into a single thing called the SAP Business AI Platform. Three layers sit on top of each other.<\/p>\n The context layer is the bit that matters most. SAP is bringing in a knowledge graph that maps thousands of business processes and over 7 million data fields across a customer’s ERP estate, plus new SAP domain models trained on SAP’s own code. The idea is that when you ask Joule for a year-end forecast, the LLM doesn’t hallucinate; it looks up the right process, pulls the right data, and checks your authorisation rules before answering.<\/p>\n The build layer is Joule Studio 2.0, which SAP, CTO, Dr. Philipp Herzig demoed on stage. It’s intent-based, deeply contextualised in SAP data and process know-how, and open enough to let you bring your own LLM (Claude, GPT, Gemini, Mistral, take your pick).<\/p>\n “Joule Studio is open by principle. You can choose from any third-party AI model. We have them all on the platform already, of course also SAP’s own AI models, and you can build in the environment of your choice. No lock-in,” said SAP CTO Dr.\u00a0Philipp Herzig<\/a>.<\/p>\n The governance layer is built on\u00a0SAP LeanIX<\/a>\u00a0and adds an AI Agent Hub that catalogues every agent (SAP and non-SAP), manages risk, enforces compliance and tracks behaviour through Cloud ALM,\u00a0Signavio<\/a>\u00a0and\u00a0SuccessFactors<\/a>. Governing non-SAP agents will be free of charge, which is a sharp competitive jab if you’ve been pricing rival platforms.<\/p>\n This infographic independently created by iTWire using Gemini Nano Banana 2. The article continues below, please read on!<\/p>\n Joule Studio 2.0 and why it might actually make it into production<\/p>\n Herzig brought a KPMG demo on stage to show what building an agent looks like now. KPMG, Global Head of Advisory, Rob Fisher took the mic to talk about the firm’s own SAP journey: 270,000 users on S\/4HANA public cloud, Joule deployed across 34 countries, 3,000 consultants already using it, and 20 client agents already built on the new platform.<\/p>\n “For one of the clients we’re working with, we’re targeting one hundred twenty million dollars in reduced contract leakage. That is real tangible value,” said KPMG Global Head of Advisory\u00a0Rob Fisher<\/a>.<\/p>\n The demo itself walked through a pricing validation agent for a pharma distributor flagged with $24 million of annual margin leakage. The agent was scoped in plain English, generated a product requirements document in minutes (Herzig said this used to take weeks), and then assembled itself, including unit tests, with n8n orchestrating the underlying multi-agent flow.<\/p>\n Herzig claimed Joule Studio gets developers to outcome an order of magnitude faster than comparable platforms on accuracy, thanks to baked-in SAP domain knowledge. SAP is also giving Joule Studio away for free until the end of the year, which is a clear move to grab developer mindshare while the agent platform wars are still warm.<\/p>\n In the Q&A, Isaac Sokolik from StarCIO pressed Herzig on what differentiates Joule Studio 2.0 from the wall of MCP servers and fabric platforms CIOs are seeing everywhere. Herzig’s answer was useful, and slightly contrarian. The LLMs are commodity. The coding assistants are commodity. Use whatever Claude Code or Codex or coding tool you like.<\/p>\n “What we are really focused on is everything that comes on top that is kind of the defining factor as to what stays in PoC or is very likely to stay in PoC versus what hits production really adds value,” said SAP CTO Dr. Philipp Herzig.<\/p>\n Translation: the knowledge graph context, the connection to your real S\/4 and Business Data Cloud data, the security and trust plumbing, and the evals built from your actual invoices and supplier docs (not synthetic ones) are what separates a demo from something a CFO will sign off on.<\/p>\n The autonomous suite: 224 agents, 51 assistants<\/p>\n SAP, Product and Engineering Lead, Muhammad Alam then took the stage to walk through the autonomous suite. Alam is on the SAP Executive Board and runs all of SAP’s business software applications. Find him on\u00a0LinkedIn here<\/a>.<\/p>\n SAP is launching:<\/p>\n 224 agents<\/p>\n 51 Joule assistants<\/p>\n 5 autonomous domains (finance, spend, supply chain, HCM, CX)<\/p>\n All built on a SOX-audit-compatible, ISO-certified development process<\/p>\n Each assistant maps to a real role inside a business. So instead of giving someone a tool to use, you’re giving them a coworker that already knows the company’s processes, controls and history. Alam called the layer that captures the tribal knowledge “company memory”, and it’s built on SAP Signavio. It ingests process models, policy docs, Slack and Teams content, and even messy email approval chains, then extracts what SAP calls process atoms: small structured rules an agent can follow (and, crucially, exceptions it can learn from).<\/p>\n The standout demo was the financial closing assistant, presented by SAP, Engineer, Sophia Levins. It walked through a quarterly close across 12 legal entities, 4 ERPs and 6 currencies, with intercompany reconciliations resolved with a single click and accruals approved against confidence thresholds set by the company. Levins put it well.<\/p>\n “Each close will teach the next one. The system gets sharper, the exceptions get rarer. The close stops being something you survive and becomes something that can run itself,” said SAP demo lead Sophia Levins.<\/p>\n Twelve customers are already deploying the close assistant in early design partnerships.\u00a0Bayer<\/a>,\u00a0Novartis<\/a>,\u00a0Kaeser Compressors<\/a>,\u00a0Ericsson<\/a>\u00a0and\u00a0Petco<\/a>\u00a0all got name-checked for other domains.<\/p>\n Importantly for the long tail of customers stuck mid-modernisation, SAP confirmed a significant chunk of these assistants and agents will work in hybrid setups connecting to S\/4HANA on-premises and ECC. That’s a real concession to reality. As Alam said on stage, “We’ve done this, so you can start generating value from AI today on the SAP Business AI platform. While you’re modernising your estate.”<\/p>\n SAP, Chief Operating Officer,\u00a0Sebastian Steinhaeuser<\/a>\u00a0elaborated in the press Q&A.<\/p>\n “Our digital assistants and agents are designed for the cloud. This is the destiny and the destiny remains absolutely unchanged. However, we also want to meet our customers where they are on their transformation journey,” said Steinhaeuser.<\/p>\n The RPA 2.0 trap<\/p>\n Steinhaeuser also dropped one of the sharpest lines of the entire press conference, and it’s worth reading twice if you’re planning an AI program.<\/p>\n “We are in the tech push era that we’ve always seen with new technology, and everyone is trying. And then many people have actually called with light speed into an RPA 2.0 disaster where they realise plastering AI over a broken architecture without real governance,” said Steinhaeuser.<\/p>\n His counter-example: H&M, where SAP didn’t just drop five agents on top of an existing process. They worked through the industry logic, the underlying architecture, the process redesign, and the user interaction model. That’s the change management gap most enterprises will trip on.<\/p>\n Klein echoed the same point. He pulled out an economic-history analogy: the shift from steam to electric engines took industry 30 years to actually produce productivity gains, because retooling factories and reskilling labour took longer than the technology itself.<\/p>\n “At the end, never forget: the change management is so important. Our HR officer Gina, I mean she is running a huge reskilling program within SAP. When we want to simplify and automate pricing with AI, our sales people need to understand how to code in the future,” said SAP CEO Christian Klein.<\/p>\n For Australian and global iTWire readers running SAP shops: this is the line you need to take to your board. The agents are real and they work. But if you drop them onto a fragmented process and call it transformation, you’ll be in an RPA 2.0 mess by Q3.<\/p>\n JPMorgan Chase: the trust test<\/p>\n If you want to know whether enterprise customers actually buy any of this, watch the JPMorgan Chase cameo. JPMorgan Chase, CFO, Jeremy Barnum took the stage to confirm the bank is upgrading its general ledger to the latest SAP version under RISE with SAP. That’s roughly 86 million US customers, 7 million small businesses and around $12 trillion in daily payments across 120 currencies sitting on top of this.<\/p>\n Barnum, whose\u00a0JPMorgan bio is here<\/a>, wasn’t shy about the bar.<\/p>\n “JPMorgan Chase, we like our unicorns with only two ears, please. Thank you very much. It’s very important,” said Barnum.<\/p>\n He also pinpointed the trap most finance organisations are about to walk into.<\/p>\n “AI is only as good as the data and processes underneath it. It can’t reach its full potential in a fragmented legacy environment. So we’re very focused on end-to-end process reengineering, and through Rise with SAP, we’re embedding intelligence directly into the workflows,” Barnum added.<\/p>\n He summed up the partnership in three words: scale, speed and trust.<\/p>\n That’s a useful frame for anyone in Australia trying to sell an AI transformation business case to a sceptical board.<\/p>\n The API policy elephant<\/p>\n The Q&A also surfaced the SAP API policy controversy, with a sharp question to Herzig: are the new policies just a way to lock customers into SAP-controlled integration?<\/p>\n Herzig pushed back hard. Rate limits, deprecating old protocols, moving traffic to A2A and Apache Iceberg, that’s industry-standard stuff, he argued, and necessary given the scale of agentic workloads now hitting SAP systems. But the more interesting argument came from Alam.<\/p>\n “Not all SaaS applications are created equal. If you just go access SAP, you don’t really understand sort of the engine of what it’s doing behind the scenes. The better way for you to do that is when I go invoke that piece through APIs, including orchestrations, because what we’re bringing is that logic, that reasoning in a safe, certified compliant way that you can rely on,” said SAP Product and Engineering Lead Muhammad Alam.<\/p>\n Read that how you will. If you’re a partner who’s been raw-API-scraping SAP for years, this is going to require a rethink. If you’re a CFO worried about agentic AI hitting your statutory reporting through an unmanaged integration, you’ll probably like the new direction.<\/p>\n Anthropic and NVIDIA: the partners doing the heavy lifting<\/p>\n Two of the bigger ecosystem moments came via video.<\/p>\n Anthropic<\/a>\u00a0confirmed Claude is now available to SAP customers through the SAP Business AI Platform, with Claude powering Joule agents across finance, HR, procurement and supply chain.<\/p>\n “We build Claude, the AI that enterprises trust most. At Anthropic we build frontier AI systems, and I’ve learned that when the stakes are high, enterprises reach for the AI they trust,” said Anthropic, in a recorded keynote message.<\/p>\n (For the\u00a0iTWire\u00a0crowd: this is the same Claude that powers the desktop and Chrome agent tooling many of you are already trying out.)<\/p>\n NVIDIA, Founder and CEO, Jensen Huang appeared by video too, talking up SAP’s deployment of\u00a0NVIDIA OpenShell<\/a> as the secure runtime for Joule Studio. OpenShell is an open-source project that defines what an agent is allowed to do, see and execute, with everything traceable.<\/p>\n “Agents can take action, but only what they’re allowed to do. They can access data, but only what they’re permitted to see. They can execute workflows, but every action is traceable. This is how we make AI safe and useful,” said NVIDIA Founder and CEO Jensen Huang.<\/p>\n The takeaway: SAP wants to be the trusted execution layer that sits on top of every model, every cloud and every agent framework, with the ERP brain as the system of truth underneath. Microsoft and Google Cloud are in on agent-to-agent interoperability. Mistral and Cohere are providing sovereign model options (handy for regulated sectors and Australian government customers eyeing data residency).\u00a0Reltio<\/a>\u00a0brings master data unification.\u00a0Dremio<\/a>, which SAP intends to acquire, brings open-table-format federation via Apache Iceberg. And\u00a0Praalabs<\/a>, another stated acquisition target, brings tabular foundation models for non-SAP data.<\/p>\n That’s a lot of moving parts, but the architecture is coherent. SAP is building the plumbing for an agent economy where its ERP brain is the source of truth.<\/p>\n Industry AI and the H&M moment<\/p>\n The industry AI announcement might be the most underrated part of the keynote. SAP’s seven autonomous industry solutions cover regulated manufacturing, project delivery, revenue growth management, unified commerce and more, with deep verticals baked in for pharma, retail, energy and consumer goods.<\/p>\n The retail story came from H&M Group, Chief Digital and Information Officer, Ellen Svanstr\u00f6m, who joined SAP, Chief Operating Officer, Sebastian Steinhaeuser on stage. Svanstr\u00f6m is on\u00a0LinkedIn here<\/a>\u00a0(she’s reportedly leaving the company later in 2026 as Diego Teijeiro Ruiz takes over as CIO).<\/p>\n H&M is running over 4,100 stores in 81 markets on a foundation that now includes RISE with SAP, SAP Business Data Cloud, SAP Commerce Cloud and SAP SuccessFactors. On top of that, the retailer is deploying a Store Intelligence app and a Store Intelligence Agent that recommends next-best actions for store managers.<\/p>\n “This isn’t just a single agent, this is a system of agents working end to end across our value chain. So when the store receives a recommendation, it is based on what’s happening real time across different business processes,” said H&M Group Chief Digital and Information Officer Ellen Svanstr\u00f6m.<\/p>\n The bit that made the audience laugh was the store concierge demo. Svanstr\u00f6m had it suggest an outfit for Steinhaeuser’s keynote based on real-time inventory and personalisation signals, and the agent obliged. (His daughter, apparently, said he looked like he always looks. (Make of that what you will.)<\/p>\n Other industry wins worth noting:<\/p>\n Takeda<\/a>\u00a0targeting up to 25% reduction in revenue lost to stockouts and 5% reduction in safety stock with autonomous regulated manufacturing<\/p>\n Coke One North America (CONA)<\/a>\u00a0automating direct-to-store delivery exception handling<\/p>\n Salling Group<\/a>, Denmark’s largest retailer, automating product introduction across categories, suppliers and channels<\/p>\n RWE<\/a>\u00a0using autonomous asset management to reduce unplanned downtime across offshore wind turbines<\/p>\n That last one is particularly relevant for the renewables and infrastructure crowd in Australia. The same playbook applies to offshore wind, large-scale solar and grid-scale battery operators.<\/p>\n CX is the heart, ERP is the brain<\/p>\n A question from Constellation Research analyst\u00a0Liz Miller<\/a>\u00a0in the media and analyst Q&A following the keynote teed up one of Alam’s most interesting answers. If ERP is the brain of the enterprise, CX is the heart. What happens when SAP turns autonomous CX loose on the front office?<\/p>\n Alam’s view: the chasm between front and back office was created by the tech industry decades ago, and the agentic layer is what finally closes it. He pointed to H&M as the proof point, with agents on the front office talking directly to inventory, pricing and fulfilment agents on the back.<\/p>\n “For us, the opportunity we’re going against very aggressively is in the front office, in the CRM space, to say you can build like sales, service, marketing experiences now deeply aware and connected to the data in the back office to be able to now create value that wasn’t possible before,” said SAP Product and Engineering Lead Muhammad Alam.<\/p>\n If you’re in the camp that thinks SAP can’t win in CX, this is the strongest argument I’ve heard for why it might. Last-mile distribution, revenue growth management, retail merchandising. Categories where back-office data is the actual moat.<\/p>\n The \u20ac100 million migration sweetener (and how to claim it)<\/p>\n Klein closed the keynote with a fairly aggressive commercial pitch. SAP is investing over \u20ac100 million into its partner ecosystem to drive agent development and AI adoption. RISE with SAP customers get three Joule assistants activated in their first year as a contractual commitment. SAP GROW customers (typically smaller and mid-market) get the full portfolio at onboarding.<\/p>\n In the press Q&A, Steinhaeuser broke down how the fund actually works. Partners can apply for three things on behalf of customers: activate an assistant, extend an assistant (because a chemicals company runs cash collection differently from a retailer), and build entirely new agents to complete the autonomous suite for a customer.<\/p>\n SAP is also tackling the elephant in the ECC room. The vendor’s AI-powered migration toolchain is being augmented with assistants for data migration, test automation and business process reengineering. SAP is targeting up to 50% reduction in migration effort (the press release pegs it at 35%, but Klein went bigger on stage), and the company has named\u00a0Palantir<\/a>\u00a0and\u00a0Accenture<\/a>\u00a0as partners for the gnarliest scenarios, plus\u00a0Conduct<\/a>\u00a0for cloud ERP migrations.<\/p>\n S\/4HANA on-premises and ECC customers committing to move the majority of their estate to SAP Cloud ERP get access to select AI scenarios in the meantime. That’s SAP throwing a rope to the customers who’ve been dragging their feet on cloud, and it’s smart. Lots of Australian mid-market and government SAP customers are still on ECC. They’ve been told for years that AI is coming. Now SAP is saying: start now, and we’ll meet you halfway.<\/p>\n Prediction on the fly: RPT 1.5 and the tabular play<\/p>\n One of the most technically interesting reveals was around tabular foundation models. SAP’s\u00a0Rapid 1.5<\/a>\u00a0(now with a chat API, what-if simulation and full explainability) handles tabular predictions on SAP data without an army of data scientists training a hundred narrow models. The pending acquisition of\u00a0Prior Labs<\/a>\u00a0adds TabPFN model strength on the non-SAP side, including emerging use cases like financial trading.<\/p>\n In the Q&A, Klein recounted seeing Rapid in action with a Swiss retailer.<\/p>\n “For the Swiss retailer we put the data in, and I saw that the accuracy was actually better. The data was not even curated, and there was no data scientist. I said, ‘My God, how cool!’ And if our agents now can perform these predictions on the fly. I mean, this is huge,” said SAP CEO Christian Klein.<\/p>\n Herzig’s caveat is worth keeping in mind: garbage in, garbage out still applies. The tabular foundation models work on arbitrary tables (someone’s apparently testing one for stock-price forecasting already), but the master-data quality story Reltio and Dremio are filling in is the other half of the equation.<\/p>\n This infographic independently created by iTWire using Gemini Nano Banana 2. The article continues below, please read on!<\/p>\n “Captain’s log, Sapphire 2027”: my Q&A question to SAP<\/p>\n When the press Q&A opened up, I had a chance to ask the panel where this all goes from here. Early access starts in June, but in this age of monthly cycle releases, by Sapphire 2027 the autonomous enterprise will have been humming along for a year in lots of places. I asked: what does that next stage look like? On the Starship Enterprise, space was the final frontier, but with the autonomous enterprise, is there an equivalent final frontier? Or does AI eliminate the concept of “final” altogether? (You can’t spell “final” without AI, after all.)<\/p>\n Alam went first.<\/p>\n “In settings like this, I tend to resist the urge to look into a nonexistent crystal ball and predict the future like some of the other tech pundits do. It sounds really good; you can get a headline out of it that ‘the world changes,'” said SAP Product and Engineering Lead Muhammad Alam.<\/p>\n He then made a point worth quoting at length, it\u2019s a strategy statement:<\/p>\n “We’re excited that the things that we talked about are goods that are on the truck now. That our customers today even as we sit here are touching and playing with. It was broader access beginning June. But I am also excited about the things I hope and expect Christian is going to launch via LinkedIn over the next month and the month after, and the month after, because I think the world around us is changing in a monthly and a quarterly way,” Alam noted.<\/p>\n The signal: SAP’s roadmap will drop as a steady stream of LinkedIn posts from Klein, with announcements rolling out month by month rather than waiting for the next big-bang keynote, although there will obviously be massive things to unveil at future SAP Connect, Sapphire and other events, as you\u2019d expect. Follow him\u00a0here<\/a>\u00a0if you want the cadence.<\/p>\n Klein then jumped in with what another great answer:<\/p>\n “One of the founders of SAP once told me. The moment, when you know you have a platform is when all of your partners and ecosystem is building something. And next year at Sapphire, I want to see all of our partners on stage and said, ‘Hey, look. This is the best trending AI platform in the industry,” Klein added.<\/p>\n That’s the success metric. Sapphire 2027 will be measured by how many partners show up on stage with agents built on this platform. If that happens, SAP wins big \u2013 as do its customers. If it doesn’t, this whole thing is a very expensive marketing exercise.<\/p>\n For what it’s worth: based on what I saw on the floor in Orlando and the partner roster already activated, my money is on it happening.<\/p>\n So, is SAP still a software company?<\/p>\n Klein circled back to his opening question at the end. He asked Joule. The agent answered.<\/p>\n “SAP is becoming a business AI company. That’s 100% accurate, unlike your unicorn drawing,” said Joule, SAP’s AI assistant (yes, Klein got roasted by his own agent).<\/p>\n It was a clever line, and a pretty honest read on where SAP is heading. The company has spent five decades building the system of record for the world’s biggest businesses. It now wants to be the system of execution, with agents doing the work and humans staying in the loop only where it matters.<\/p>\n Will it work? The architecture is right. The ERP context advantage is real. The partner list is serious. And the customer roster, JPMorgan Chase, Bayer, Novartis, H&M, RWE, Takeda, makes it hard to wave away as marketing.<\/p>\n The challenge is execution at customer sites. Change management, process redesign, data hygiene and developer enablement are still the hardest parts of any AI rollout. SAP knows it. That’s why the RISE and GROW programmes now include onsite consultants, activation commitments and audit-ready agents out of the box.<\/p>\n Klein’s closing line will probably echo for the rest of 2026: “It’s the end of long negotiations, supply chain disruptions, timesheets, financial blind spots. And hopefully the end of being overworked. It’s definitely the beginning of better.”<\/p>\n Or as the keynote’s opening sketch put it, when leeches were being replaced by inoculation and rawhide salesmen were being told they were now selling cars: every great shift in business technology has been met with the same nervous shrug.<\/p>\n The autonomous enterprise is the next one. And if you run SAP, you’re already on the runway, whether you’ve noticed or not.<\/p>\n SAP \u2013 it truly is a superlative autonomous platform!<\/p>\n![\"Sapphire](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2026\/05\/sap-says-the-autonomous-enterprise-has-arrived-and-80-accuracy-isnt-going-to-cut-it-inline-5-1a87375.png\")
<\/p>\n![\"SAP](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2026\/05\/sap-says-the-autonomous-enterprise-has-arrived-and-80-accuracy-isnt-going-to-cut-it-inline-6-0f7ba7c.png\")
<\/p>\n