The first wave of enterprise AI could spread across industries with relatively little specialization. Assistants summarized documents, generated text, searched information and helped write code whether they were deployed by a bank, manufacturer, utility or healthcare provider.
That changes as AI moves closer to operational execution. Releasing a pharmaceutical batch, rerouting production after a supply disruption or coordinating repairs after a grid fault requires more than general reasoning. Each process operates within its own data structures, business rules, regulations, exceptions and tolerance for error. Few companies have made that move yet: SAPinsider research found 74% of SAP customers still identifying use cases or experimenting, and only 14% described their AI maturity as optimized or transformational. Moving deeper into industry operations raises a different question: How much business context does AI need before companies can trust it to act?
What Is Industry AI?
Industry AI combines general AI capabilities with the domain knowledge needed to operate inside a particular sector. That context can include industry terminology, process logic, data relationships, regulatory requirements, exceptions, and the systems where work takes place.
Manufacturing provides a useful example. NIST’s 2026 smart manufacturing roadmap describes AI operating across heterogeneous sensing, control, and production environments where reliability, explainability, interoperability, and safety matter alongside model performance. SAP is pursuing a similar idea commercially through an Industry AI portfolio that combines general AI capabilities with industry process knowledge and enterprise data. The industry layer supplies the context that determines whether an AI-generated action fits the process around it.
Why Does Industry Context Matter More as AI Takes Action?
Industry context becomes more important as AI moves from producing information to influencing or executing consequential decisions. A drafting assistant can generate an imperfect first version for a person to review; an AI system acting inside a financial, clinical, industrial, or energy process may have much narrower margins for error.
Financial services illustrates the difference between adoption and authority. A Bank of England and Financial Conduct Authority survey found 75% of responding firms already used AI and 55% of reported use cases involved some automated decision-making, yet only 2% involved fully autonomous decisions. Healthcare shows a similar boundary: 81% of physicians told the AMA they use AI professionally, while liability, validation, and physician involvement remain important conditions for trust. Adoption can spread much faster than delegated decision authority.
Does Industry AI Require Specialized Models?
Industry AI does not necessarily require a model trained specifically for one sector. Specialization can come from the information and controls surrounding a general-purpose model, including enterprise data, retrieval systems, ontologies, business rules, application integrations, and permissions.
Domain knowledge can sit at several layers of the architecture. A general model may provide language or reasoning capabilities while another system supplies the current production order, customer contract, regulatory rule, or equipment state needed to reach a valid conclusion. SAP’s Industry AI approach reflects this layered model by combining its broader AI platform with industry data and process knowledge. The relevant measure of specialization is increasingly how much domain context reaches the AI at the point of a decision.
Which Processes Are Moving From AI Assistance to Execution?
Industry AI is moving furthest toward execution in processes where AI can interpret changing conditions while operating inside defined controls. The transition is uneven because industries assign very different consequences to automated decisions.
Healthcare remains heavily assistive, with the AMA finding AI use concentrated in tasks such as clinical documentation and summarization. Energy provides examples closer to operational execution: the International Energy Agency says AI-based fault detection can reduce outage durations by 30% to 50%, while AI can also support grid optimization and maintenance decisions. Manufacturing introduces physical production, quality, and asset constraints, while financial services already uses AI within fraud and anti-money-laundering workflows. Across these sectors, execution tends to expand where the process supplies clear boundaries for what the system can observe, decide, and change.
What Determines Whether Industry AI Can Scale?
Industry AI scales when domain knowledge can be converted from customer-specific expertise into repeatable data, processes, integrations, and controls. The harder industry problems often begin with significant customization because operational knowledge may live across employees, legacy applications, documents, and informal exception handling.
SAP’s use of forward-deployed engineering illustrates one approach: engineers work directly with customers on industry problems, with recurring patterns potentially becoming reusable products. The broader scaling challenge extends beyond SAP. NIST’s manufacturing roadmap identifies interoperability, data management, integration, trustworthiness, and standards as obstacles to broader industrial AI deployment. The pace of adoption may increasingly depend on how quickly specialized knowledge can become reusable infrastructure.
What This Means for ERP Insiders
Domain knowledge is becoming technical infrastructure. Process rules and industry semantics once lived mainly with employees and applications. AI execution increasingly requires that knowledge to become accessible in forms machines can reliably use.
Adoption and autonomy will diverge. Industries can report widespread AI use while keeping consequential decisions under close human control. Usage rates alone reveal little about how deeply AI has entered operational execution.
Customization becomes a scaling test. Early industry AI deployments may depend heavily on customer-specific engineering and process knowledge. The market will mature as those recurring patterns become reusable across companies in the same sector.
