SAP Customers Set Priorities at ASUG Tech Connect

At ASUG Tech Connect 2026, held for SAP leaders, enterprise architects and users, attendees compared their priorities with vendor messaging and shared progress on data, process and migration work. The event had more attendee momentum than last year, and some customers presented early agentic AI results. SAP representatives joined community and “ask me anything” sessions, where they heard feedback and sometimes offered answers customers had not known about. The author says the costs and return on investment (ROI) of AI remain unsettled.

Attendees focused on modernizing and securing SAP ECC systems, challenges with RISE, and requested Business Data Cloud features, while SAP was emphasizing Joule Work. Gartner reported that only 13% of CIOs saw significant AI value, and more than 50% of generative AI projects were abandoned after proof of concept in 2025, citing bad data, weak controls or unclear value. The author argues that stronger data context and guardrails can support agents in important business processes, while requiring care over how much autonomy they receive. At Agilent, a tariff agent used external tariff data and an internal product catalog in real time; business users said they could not imagine working without it.

SAP’s approach to agent context includes Company Memory, SAP-RPT-2 for tabular data, knowledge graphs and the Business Data Cloud Ontology Manager, which is intended to incorporate customer-specific data into Joule’s agentic inference. The author says customers still own their AI strategy and data, and notes tension between granular control of AI services and SAP’s infrastructure management in RISE private cloud deployments. Customers also need to develop their own AI skills and monitor changing agent context, outputs and costs.

Security, governance and AI ethics sessions drew strong attendance. In a session led by SAP cybersecurity and compliance adviser Srinivas Nandiwada, attendees heard recommendations to apply controls during development, monitor every agent after launch, keep an active agent registry in LeanIX, require a deployment manifest and reapproval for scope changes, and document risks and scope in a system card. Nandiwada advised building governance that can absorb new regulations, testing the highest-risk agent first and limiting fully autonomous agents to bounded, low-risk tasks rather than decisions affecting people. An audit log of agent actions was also described as necessary. One customer recommended cataloging agents, including ones the organization may not control, and auditing security to give colleagues a more objective view.

The author says SAP customers showed greater sophistication than at last year’s event in scaling and governing agents, grounding them in suitable data, establishing data and process foundations, winning business support, demonstrating business value and managing inference costs. Many popular sessions instead covered ERP migration and “clean core” work; the Hershey’s session on clean core and RISE migration drew heavy attendance and included Business Data Cloud experiences. The author says migrations now feel more strategic, and argues that effective AI use has no shortcuts. SAP partners were also helping with migrations, specialized data products and AI operations. A smartShift presentation illustrated using deterministic, rule-based automation where it is sufficient, which the author says can reduce inference costs and help customers make informed choices.

SAP value advisory services representative Todd Hassell emphasized aligning IT with business strategy. The author argues that business goals, culture, processes, data quality and AI context all matter, and says SAP’s public cloud ERP stories are more compelling to him than private cloud examples, though harder to find. SAP has said it will bring ECC customers along and is making it possible for them to start using SAP Business AI before a major migration. The author notes that ECC customers still face decisions about modernization, business priorities and limiting security exposure in older systems.

At ASUG Tech Connect 2026, held for SAP leaders, enterprise architects and users, attendees compared their priorities with vendor messaging and shared progress on data, process and migration work. The event had more attendee momentum than last year, and some customers presented early agentic AI results. SAP representatives joined community and “ask me anything” sessions, where they heard feedback and sometimes offered answers customers had not known about. The author says the costs and return on investment (ROI) of AI remain unsettled.

Attendees focused on modernizing and securing SAP ECC systems, challenges with RISE, and requested Business Data Cloud features, while SAP was emphasizing Joule Work. Gartner reported that only 13% of CIOs saw significant AI value, and more than 50% of generative AI projects were abandoned after proof of concept in 2025, citing bad data, weak controls or unclear value. The author argues that stronger data context and guardrails can support agents in important business processes, while requiring care over how much autonomy they receive. At Agilent, a tariff agent used external tariff data and an internal product catalog in real time; business users said they could not imagine working without it.

SAP’s approach to agent context includes Company Memory, SAP-RPT-2 for tabular data, knowledge graphs and the Business Data Cloud Ontology Manager, which is intended to incorporate customer-specific data into Joule’s agentic inference. The author says customers still own their AI strategy and data, and notes tension between granular control of AI services and SAP’s infrastructure management in RISE private cloud deployments. Customers also need to develop their own AI skills and monitor changing agent context, outputs and costs.

Security, governance and AI ethics sessions drew strong attendance. In a session led by SAP cybersecurity and compliance adviser Srinivas Nandiwada, attendees heard recommendations to apply controls during development, monitor every agent after launch, keep an active agent registry in LeanIX, require a deployment manifest and reapproval for scope changes, and document risks and scope in a system card. Nandiwada advised building governance that can absorb new regulations, testing the highest-risk agent first and limiting fully autonomous agents to bounded, low-risk tasks rather than decisions affecting people. An audit log of agent actions was also described as necessary. One customer recommended cataloging agents, including ones the organization may not control, and auditing security to give colleagues a more objective view.

The author says SAP customers showed greater sophistication than at last year’s event in scaling and governing agents, grounding them in suitable data, establishing data and process foundations, winning business support, demonstrating business value and managing inference costs. Many popular sessions instead covered ERP migration and “clean core” work; the Hershey’s session on clean core and RISE migration drew heavy attendance and included Business Data Cloud experiences. The author says migrations now feel more strategic, and argues that effective AI use has no shortcuts. SAP partners were also helping with migrations, specialized data products and AI operations. A smartShift presentation illustrated using deterministic, rule-based automation where it is sufficient, which the author says can reduce inference costs and help customers make informed choices.

SAP value advisory services representative Todd Hassell emphasized aligning IT with business strategy. The author argues that business goals, culture, processes, data quality and AI context all matter, and says SAP’s public cloud ERP stories are more compelling to him than private cloud examples, though harder to find. SAP has said it will bring ECC customers along and is making it possible for them to start using SAP Business AI before a major migration. The author notes that ECC customers still face decisions about modernization, business priorities and limiting security exposure in older systems.