Why Cloud Excellence Is the Precondition for SAP’s AI Story

Most of the discussion at Sapphire 2026 centered on the Autonomous Enterprise, autonomous agents, and the Business AI Platform supporting the portfolio. Peter Pluim, Head of Global Cloud Operations at SAP, runs the infrastructure that must work underneath all of that. His team manages maintenance windows, upgrade cadences, disaster recovery footprints, and the service-level agreements keeping more than 9,000 RISE with SAP private cloud customers running. He opened our conversation in direct terms.

“We’re now at that point at SAP where my team is essentially providing the infrastructure hygiene. It’s hard to discuss transformation with the customer and AI adoption if an organization lacks the basic hygiene of providing harmonized maintenance windows, cloud excellence, availability, and all the related topics.”

While this may not be the headline used in a keynote, it is the right note to be taking because, for SAP’s AI strategy to work, the underlying platform must be reliable.

Harmonized Maintenance

The scale Pluim operates on is meaningful context. His team manages more than 9,000 private cloud customers and over 200,000 production system IDs, with combined customer revenue representing a significant proportion of global GDP. At that scale, small percentages become operationally significant.

About 75% of upgrade and maintenance schedules across SAP’s cloud line-of-business products are now harmonized, with 100% as the target. The rationale matters more than the metric. In the legacy model, each line of business could celebrate “one outage a quarter” as a win. Pluim sees that posture as incompatible with the cross-suite, AI-driven workflows SAP now sells.

“If every single cloud line of business says I have one outage a quarter and is proud of it with the new AI strategy and the business suite, that could translate into a customer being impacted every month, even every week, of a quarter,” he said. A user asking an AI agent to validate workforce capacity against a revenue forecast may touch SAP S/4HANA, SAP Fieldglass, and SAP SuccessFactors in a single prompt. If any one of those systems is in a maintenance window, the response fails, and the customer does not care which product was down.

AI Operations and Cognitive Automation

The second shift is internal. SAP is injecting AI into cloud operations, not just into the products it sells. Pluim is explicit that this involves more than scripted automation.

“I call this cognitive automation because it’s not automating an existing process. It’s being able to recognize input parameters, and take an action regardless of whether there is a process in place for doing so. We do all this through what we call our agentic AI hub.”

The operational examples are concrete. Pattern recognition algorithms know that a large batch run at quarter-close is normal, but the same workload mid-period is unusual and should launch a query to determine why the spike has occurred. Other use cases include faster root-cause analysis and engineer-facing recommendations that surface relevant best practices, so the team avoids reinventing the wheel. These capabilities route through what SAP calls the agentic AI hub, which starts with discrete use cases and integrates them into standard operational workflows.

Pluim was honest about the economics because, as he emphasized, AI is not free. The business case for AIOps focuses on quality, not cost takeout.

From 99.998% to Absolute Numbers

A notable management shift Pluim described involves moving away from percentage-based SLA targets. On SAP Cloud ERP Private, the company reports 99.998% average availability across its 200,000 production system IDs. While that number sounds high, when multiplied across 200,000 systems there can be a significant impact.

“That means around 400 systems, or 90 customers, might still have an SLA miss,” Pluim explained. “That’s why we are focusing on thinking in absolute numbers instead of percentages. We want to achieve a goal where there are no more than 18, 19, or 20 failures a month, not a 99.9% success rate on change request. Because if I have 4 million change requests, that still means 4,000 might be adversely impacted.”

Because each failure represents a customer ticket, an angry CIO, or a missed cutover, focusing on absolute numbers forces the organization to treat every miss as a problem rather than a rounding error.

Sequential to Parallel Innovation

Pluim also observes customers operating differently. The classic sequence of migrating from SAP ECC to SAP S/4HANA, moving to Cloud ERP, executing a custom-code cleanup, and then launching an AI pilot is collapsing into parallel workstreams.

“Customers three, four years ago would say, ‘Okay, let’s go from SAP ECC to SAP S/4HANA and then from SAP S/4HANA to RISE with SAP’,” Pluim noted. “And then they’d ask, ‘do I have custom code that needs to be removed? How do I take advantage of innovation?’ There’s more parallel thinking now which allows customers to go directly from SAP ECC to Cloud ERP while simultaneously updating their data strategy, allowing them to try their first AI use cases and see value immediately. ”

Operational evidence supports this view. Pluim noted he has never seen so much code cleanup and remediation as in the past 24 months. At the same time, SAP BTP integration usage is accelerating at breakneck speed, and the partner ecosystem has shifted from ABAP-heavy delivery toward clean-core, side-by-side extension work supported by the RISE with SAP Validated Partner program and clean-core dashboards. Customers are asking smarter questions of their systems integrators as a result.

SAP has completed roughly 3,500 ERP transformations for customers. These upgrades feed centers of excellence now wired into the AIOps engine. The learning from each transition becomes reusable across the next thousand engagements. The talent bench customers access grows deeper every quarter without requiring them to pay for that depth twice.

Managed Integration and the Hire-to-Retire Pilot

The forward-looking element is managed integration. Today, even on Cloud ERP, customers own integration, identity and access management, and the SAP BTP layer. Pluim stated that SAP has about 30 customers where it runs the full hire-to-retire process spanning SAP SuccessFactors, SAP Concur, and SAP S/4HANA as an integrated managed service. SAP takes responsibility for the connective tissue.

“We’re taking our cloud operations to become truly integrated; that’s not going to happen overnight,” Pluim said. “Ultimately, the aim is to make it as easy as possible for customers to drive the business, to take the business outcomes and not worry about the nuts and bolts and the underlying architectures.”

This section of the operational roadmap warrants attention. If SAP can credibly own the integration layer for end-to-end business processes, the gap between buying RISE with SAP and achieving an outcome closes. If it cannot, the Autonomous Suite narrative loses a critical assumption.

Customer Feedback as Portfolio Input

Asked how feedback shapes the portfolio, Pluim provided clear examples. Intercontinental long-distance disaster recovery, immutable backup, customer-managed keys, and heightened network security groups were developed in response to direct customer requests and in coordination with hyperscaler partners. The Middle East geopolitical situation has made multi-region data placement an urgent requirement rather than a theoretical concept.

He offered an authentic reflection on customer relationships. “I always make the joke that the more you cuddle SAP, the better service you get from us. And I think that’s very, very true,” he said. Engaged customers receive more attention because they provide a stronger signal to the product and operations teams.

What This Means for SAPinsiders

Reset internal SRE and operations targets in absolute numbers. Pluim’s shift from percentage-based availability to absolute failure counts is a transferable management practice. Translate every percentage SLA in operations into the absolute number of incidents, failed changes, or affected users it represents. Set quarter-over-quarter reduction goals against the absolute count. Percentages flatter the dashboard, but absolute counts drive corrective behavior.

Run transformation tracks in parallel and start AI prerequisites on day one. The customers extracting value run SAP ECC-to-SAP S/4HANA migrations, Cloud ERP adoption, data strategy, custom-code cleanup, and initial AI use cases concurrently. Build a parallel program plan that includes identity and access management in SAP BTP, custom-code triage, and initial AI use cases as workstreams starting alongside the migration rather than waiting until it concludes.

Evaluate managed integration as a procurement option. With dozens of customers running hire-to-retire as an SAP-managed integrated service, the boundary of what SAP will operationally own is expanding. Add managed integration scope, integration SLAs, and end-to-end process ownership to your next Cloud ERP or cloud renewal discussion. Ask what SAP will manage beyond the application instances.