Most S/4HANA migrations end up in the same place. You moved off ECC, met the 2030 deadline on schedule or close enough, and, somewhere downstream, stood up a modern analytics platform. Snowflake. Databricks. Something cloud-native and fast.
The business is asking for AI—and while the platform may be ready, the data is not.
That is the problem nobody wants to name out loud. The migration moved the data, but it did not fix it. Years of custom Z-tables, modified standard fields, and workarounds that made sense in 2009 followed you into the new environment. They are sitting in your hybrid cloud platform right now, waiting for a Joule agent or an AI model to try to reason over them.
Last week at Sapphire Orlando, SAP made the stakes explicit.
What SAP Said at Sapphire 2026
Christian Klein opened Sapphire 2026, describing the Autonomous Enterprise as not a roadmap but capabilities available now. The repositioning was direct: SAP moves from software that records what people do to software that does the work itself.
The centerpiece is the SAP Business AI Platform, which unifies SAP Business Technology Platform, SAP Business Data Cloud, and SAP Business AI into a single governed environment. At the center of that platform sits the SAP Knowledge Graph, a semantic map of all business entities, processes, and relationships within a customer’s SAP landscape. Agents use it to act on real business context rather than generic prompts.
The final sentence is paramount. The Knowledge Graph does not produce clean semantics from dirty source data; it simply mirrors what exists. If your cost center hierarchy has been maintained inconsistently since the original ECC go-live, the Knowledge Graph will reflect that inconsistency. This also influences downstream agent reasoning and will be a source of hallucinations, creating distrust.
The Day 2 keynote put enterprise customers on stage, and the message was consistent across ExxonMobil, Levi Strauss, and Lockheed Martin: AI cannot fix a broken process, and clean core is a prerequisite for AI speed. Constellation Research put it plainly: AI deployments require a clean core and cloud migration as the enterprise operating model.
This is no longer a marketing position. It is an architectural requirement.
What the Data Already Showed
Before anyone flew to Orlando, the evidence was already in.
Horváth surveyed 200 executives across six countries in Q1 2025. Projects ran 30% over schedule on average. Budgets were significantly exceeded in more than 65% of cases.
Nearly two-thirds of completed migrations identified severe-to-very-severe quality deficiencies after go-live. ASUG and Precisely user group data tell the same story: organizations that deferred clean core discipline during migration are carrying those deficiencies into their analytics environments.
The industry normalized this. Budget overruns became a known risk. Quality deficiencies became a post-project cleanup item. The cleanup never quite happened.
Now the bill has come due.
What Most Organizations Get Wrong
This is the assumption that causes companies to run into problems. They believe the hybrid cloud analytics platform solves the data quality issue, but it does not.
This is best demonstrated through a use case. A manufacturing company completes its S/4HANA migration. They bring in Snowflake or Databricks as the analytics layer. Data flows from SAP through a connector, lands in the platform, and the team builds dashboards and data products on top of it. It works well enough for reporting. Then the business asks for AI-driven forecasting or an autonomous agent to handle exception management in procurement.
The agent begins pulling context. Material master data has three unit-of-measure conventions carried over from two legacy systems consolidated in 2017. Vendor records contain duplicate entries that were never resolved because the migration team ran out of time. The cost object assignments that drive margin analysis were partially remapped, but the old structure was kept in parallel for two fiscal years and was never fully retired.
The agent does not fail loudly. It produces output that looks plausible. That is actually worse than a visible error. Plausible but wrong, at scale in mission-critical finance and supply chain processes, is the specific failure mode SAP’s customers described from that Sapphire stage.
No platform fixes these issues upstream. Snowflake does not know your intended cost object hierarchy. Databricks cannot resolve your vendor master duplicates. These are SAP data governance problems that require SAP data governance decisions.
Where the Adoption Numbers Land
The DSAG Investment Survey 2026 reports that only 3% of SAP customers run SAP Business AI in production. 77% of AI-active SAP enterprises use non-SAP tools Snowflake Cortex or Databricks Mosaic AI. Access to Joule requires a RISE or GROW contract, which excludes on-premises installations entirely.
That 3% figure reflects two factors. Some of it is the contract structure. Most of it is the data foundation. The tooling exists. Joule Studio is now generally available. Several Autonomous Finance assistants will reach GA in Q2 2026. The full Joule Work engagement layer and Agent-to-Agent interoperability are planned for Q4 2026. The stack is rolling out on a defined schedule.
The installed base is not ready for it. That is a prediction based on the state of installations today.
SAP introduced a formal Clean Core Certification Programme at Sapphire 2026, certifying BTP extensions as upgrade-compatible across three S/4HANA Cloud release cycles.
Clean Core now has a certification standard attached. SAP is not leaving this as a recommendation. It is an infrastructure policy with a compliance track.
What This Means For Your Timeline
The 2030 ECC deadline still anchors most migration planning. That is the wrong clock to watch.
The AI capability gap between organizations with a clean, governed SAP data foundation and those without it is widening right now. The Joule agent stack does not wait for 2030. Organizations standing up reliable AI-driven finance, close, procurement automation, and supply chain exception management in 2025 and 2026 are doing so because they did the hard data governance work two or three years ago that most organizations deferred.
Deferring your data foundation is not a neutral holding pattern. It is a compounding cost. Every quarter you run on a pre-clean-core data model is another quarter your analytics platform is working around problems it cannot solve, and your AI readiness falls further behind organizations that made different decisions.
If your S/4HANA migration is complete or underway and no one has owned a clean core data governance workstream, that conversation needs to happen now. Not at the next planning cycle. Now.
Not sure where to start? Talk to our AI and SAP data experts today.
