SAP Industry AI: What Business Data, Models, and Agent Controls Have to Deliver

SAP’s Industry AI strategy raises a practical question for enterprise customers: what does an AI model already know, and what information must it receive from business systems before recommending or executing an action?

Consider the decisions involved in scheduling maintenance for a critical asset or adjusting production after a supply disruption. Each requires information about current operations, established business rules and, in some cases, regulatory requirements. A model’s training can provide useful knowledge, but company-specific information such as batch status, equipment condition and available inventory must come from operational systems.

That distinction is central to SAP’s Industry AI portfolio. At Sapphire in May 2026, SAP outlined seven initial Industry AI scenarios, describing each as a combination of applications, data models and agentic capabilities. In August, it positioned Industry AI as a focused customer transformation offering that combines industry expertise, enterprise data and forward-deployed engineering to address complex business processes.

Where SAP Industry AI Gets Its Business Context

SAP describes its Business AI Platform as the foundation for Industry AI, drawing on its industry knowledge, enterprise data, and knowledge graph. However, SAP has not published a detailed architecture showing which components power each Industry AI scenario. The technologies it has identified therefore represent capabilities within the broader platform, rather than a confirmed technology stack for every deployment.

SAP Business Data Cloud provides access to governed enterprise data, including data products that bring together information from SAP and non-SAP sources. SAP Knowledge Graph provides a structured understanding of business entities, processes and their relationships, helping AI interpret information within the context of enterprise operations.

The October introduction of SAP Reltio adds another capability. According to SAP’s Connect 2026 innovation guide, the solution, now generally available within Business Data Cloud, reconciles fragmented records across enterprise applications.

A supplier, for example, might have different identifiers, attributes or records across procurement, finance and other systems. SAP Reltio is designed to help reconcile these differences and establish a consistent view of the underlying business entity. That is important when agents must combine information from multiple systems to evaluate a supplier, assess an order or recommend a transaction.

SAP Domain Models serve a different purpose. As SAP explained in its Sapphire 2026 innovation guide, these are AI models trained on SAP domain knowledge, which can include code, data, metadata, business processes, architecture information and documentation. SAP identified initial applications in generating clean core-compliant extensions and supporting natural-language access to information in SAP S/4HANA Cloud Public Edition and SAP Ariba. At Sapphire, these models were available through SAP’s Early Adopter Care program, with general availability planned for Q3 2026.

The distinction between these technologies is important. Knowledge graphs represent business relationships, domain models contribute learned SAP knowledge, and operational systems supply current business information. An agent dealing with a production exception needs more than an understanding of how production orders relate to inventory. It also needs accurate information about the specific order, available materials and applicable business requirements.

Agent Governance Determines What AI Can Execute

However, having the right information does not automatically give an agent authority to act on it. SAP’s architecture separates user interaction, agent governance, identity management and runtime enforcement. Joule Work serves as an interface through which users can engage with assistants and agents. SAP AI Agent Hub, built on SAP LeanIX, provides a framework for discovering, inventorying, assessing and governing agents, large language models and Model Context Protocol (MCP) servers.

SAP’s AI Agent Hub reference architecture identifies SAP Cloud Identity Services as the component responsible for agent identities and access control, while SAP Agent Gateway is designed to enforce access policies as agents connect to applications, tools and other systems. SAP’s interoperability reference architecture adds that external calls through Agent Gateway are authenticated with Cloud Identity Services tokens that carry a named user’s context.

These are distinct responsibilities. An agent may be able to retrieve a maintenance record without having permission to approve a work order or post a resulting transaction. Authorization must be evaluated against the specific action and the controls enforced by the underlying systems.

The architecture is also evolving. The interoperability document, last updated in October, notes that some components are not yet generally available, and that bidirectional communication with third-party and self-hosted agents through Agent Gateway is not yet supported. Customers therefore need to distinguish the intended architecture from the capabilities supported in their particular deployment.

At SAP Connect, the company announced additional governance, risk and compliance assistants intended to extend existing identity and access controls to autonomous agents. The Access Governance and Security Assistant, scheduled for general availability in Q1 2027, is designed to coordinate identity governance across human and AI identities. Its later release does not mean existing agents operate without governance controls.

Specialized Models Add Another Layer of Business Intelligence

SAP’s model development reflects its view that enterprise AI also requires specialization within the models themselves. At Sapphire 2026, SAP announced RPT-1.5 and a retrieval-augmented prediction capability designed to retrieve relevant customer data for more contextual predictions, with general availability of that capability expected in Q3 2026. SAP-RPT-1.6 has been made available to customers, positioning tabular models and language models as complementary technologies for different business tasks.

These developments demonstrate SAP’s investment in models designed for enterprise data. However, SAP has not publicly established whether individual Industry AI scenarios use RPT models, SAP Domain Models or industry-specific fine-tuned models. The contribution of model specialization and runtime context therefore needs to be assessed at the level of the particular solution.

Industry AI Brings These Capabilities Into Operational Processes

SAP has identified seven initial Industry AI domains: Autonomous Asset Management, Autonomous Commodity Management, Autonomous Adaptive Production, Autonomous Regulated Manufacturing, Autonomous Revenue Growth Management, Autonomous Unified Commerce and Autonomous Project Delivery. Each addresses processes that extend across traditional application and functional boundaries.

SAP’s examples include energy providers anticipating spare-parts requirements and identifying suppliers, and retailers adjusting inventory and promotional decisions using demand signals and simulated scenarios. In each case, the usefulness of AI depends on the operational information available and the business constraints governing the decision. In pharmaceutical manufacturing, SAP has presented fully automated batch release as an Industry AI scenario. Its Sapphire roadmap identifies the Batch Release Assistant, scheduled for general availability in Q4 2026, as a capability designed to coordinate supply chain, sourcing, production execution and quality compliance.

SAP’s broader Industry AI rollout is taking place in stages. At SAP Connect, the company outlined additional capabilities for asset-intensive industries, commodity businesses, energy and natural resources companies, and retailers, with general availability planned for Q4 2026. Automotive industry capabilities are planned for the end of Q1 2027.

What Customers Need to Establish Before Implementation

SAP identifies three differentiators behind Industry AI: more than 50 years of industry and process expertise across 26 industries, the business data and ontologies available within customer systems, and forward-deployed engineering teams that work directly with customers on problems lacking packaged solutions.

The engineering approach is intended to turn customer-specific work into repeatable software capabilities. SAP also describes an all-in-one commercial offering covering platform consumption, solutions, forward-deployed engineering and support under a single contract, with pricing tied to customer business value. That makes implementation scope and measurable outcomes central to an Industry AI evaluation.

Deployment model matters as well. SAP has explicitly identified SAP Cloud ERP Private customers as able to benefit from the Industry AI portfolio.

SAP has described the technologies it intends to bring together, but individual implementations still need to establish how the components integrate, which capabilities are available and what operational controls apply. Those answers will determine whether an Industry AI use case can move from a proposed scenario to a dependable business process.

What This Means for SAPinsiders

Evaluate the quality and coverage of business context before selecting an AI scenario. SAP’s architecture draws on multiple technologies to supply enterprise knowledge, relationships and operational data. Customers should identify which business objects an agent needs, where their current values originate and how records from SAP and non-SAP systems are reconciled. For environments with custom SAP fields, extensions or multiple systems of record, the evaluation should establish whether critical information is available and correctly interpreted. Pilot testing should include incomplete, inconsistent and outdated records to determine whether the agent can recognize information gaps before making recommendations or executing actions.

Plan pilots around the gap between agent availability and agent governance. SAP’s Autonomous Suite assistants began reaching general availability in October, while the Access Governance and Security Assistant is not planned until Q1 2027 and Agent Gateway does not yet support bidirectional communication with third-party and self-hosted agents. Early pilots will therefore depend on existing SAP authorizations, identity services and application-level controls. Customers should test whether approval and segregation-of-duties requirements still hold when an agent, rather than a person, initiates an action, along with exception handling and audit trails, before moving from recommendations to automated transactions in regulated or financially sensitive processes.

Define measurable outcomes and long-term ownership before signing an Industry AI agreement. SAP’s forward-deployed engineering model and value-based commercial approach make clear success criteria particularly important. Customers should establish baselines for the process being automated, such as exception resolution time, manual intervention rates or maintenance delays, and agree how improvements will be measured. Doehler’s reported time savings illustrate the relevance of operational metrics, although the published account does not provide enough detail for independent assessment. Buyers should also clarify which components will become standardized products, which remain customer-specific and who will maintain the solution as business processes, data structures and SAP releases change.