Why AI Requires A New Enterprise Operating Model
The next enterprise transformation is a fundamental operating model redesign, not merely a technology upgrade. It demands integrating ERP, supply chain, data, AI and security to move business decisions from context to controlled action. Systems are evolving from record-keeping to proactive execution, with AI interpreting information and coordinating workflows. However, AI requires clear decision rights, governance and human judgment for exceptions. Trusted data is now an operational necessity and security must trace every automated action across platforms. People are crucial, designing the model, providing oversight and ensuring accountability. Leaders must map specific processes, test tough scenarios and prioritize outcomes to avoid fragmented AI failures.
The next enterprise transformation is not another technology implementation. It is a redesign of how the enterprise operates. ERP, supply chain, data, AI and security create value when they work together to move a business decision from context to action, with clear ownership at every step.
For years, transformation was judged by technical milestones: implementing Enterprise Resource Planning (ERP), migrating to the cloud or establishing a data platform. Those milestones still matter, but they are incomplete. A system can go live on time and within budget while decisions remain slow, data remains disputed and work still depends on manual handoffs. The technology changed; the operating model did not. In a recent Forbes article, “People, Process, Technology And The Shift Nobody Saw Coming,” I argued that technology may drive transformation but people drive technology. AI now raises the next question: what must the enterprise become as technology starts participating in decisions and execution? The answer is a single operating model linking systems, decision rights, controls and human judgment.
Disclosure: KramerERP provides paid research, advisory and consulting services to technology companies, including ERP and data vendors discussed in this industry.
The need for that redesign is increasing as enterprise systems transition from being systems of record to becoming systems of action. These modern systems can now flag exceptions, recommend next steps, trigger workflows and execute tasks across applications. According to McKinsey’s State of AI report November 2025, nearly nine in ten organizations now use AI in at least one function, yet most report no significant effect on enterprise-wide EBIT. The report’s top AI performers, part of the small group capturing value, follow two key practices: they redesign workflows from start to finish and assign senior leaders responsibility for AI governance. The main insight is that adding more AI does not automatically change how a business operates.
For many years, ERP systems have been instrumental in managing transactions across finance, procurement, manufacturing and human resources. Yet many processes still rely on manual handoffs and spreadsheets. An invoice exception can sit unresolved because the right person was never notified, while finance loses days reconciling data before close.
Modern ERP platforms can reduce those delays by flagging exceptions, recommending next steps and starting controlled workflows. In collections, an AI-supported process can rank overdue accounts by payment history, customer value and risk, recommend an action and route higher-risk cases to a manager instead of treating every account the same.
ERP’s role is transactional control. It should move routine work through governed workflows while reserving material exceptions for human judgment. The business sets approval thresholds and is responsible for the results, while ERP consistently enforces these rules. As I wrote in “AI Is Changing ERP, Not Replacing It,” AI can accelerate ERP, but it does not replace the controls and data discipline that make transactions reliable.
Most organizations can see inventory, supplier performance, delays and demand more clearly than before. But visibility without decision logic only shows the problem in higher resolution.
The supply chain’s specific role is connecting planning with physical execution. When a shortage hits, a connected operating model can identify affected orders, check inventory across locations, estimate the cost of alternatives and recommend a response based on customer priority and operational impact.
Planners and procurement teams still resolve the exceptions that require experience and judgment, but they should not have to assemble the context manually. Better coordination can reduce stockouts, premium freight and recovery time when planning and execution run as one process. I made a similar case in “Why Sports Has Become a Blueprint for Real-Time Enterprise Execution.” NFL Next Gen Stats and MLB Statcast show why real-time visibility matters only when the organization knows who should act, what to decide and how quickly execution can follow.
Every business decision depends on clear definitions, but many organizations still face challenges due to duplicate records and unclear ownership. ERP defines a customer one way, the customer platform another and the supply chain system a third. The problem is not technical alone; the business has not agreed on who owns the definition or resolves conflicts.
The risk increases when AI relies on those records to make recommendations or take action. While AI can process data quickly, it doesn’t resolve disagreements over definitions, which can cause confusion or errors. It might make a recommendation seem very reliable, even though the background context is still uncertain.
Data provides important context. Clear definitions, ownership, quality standards, and source corrections ensure ERP, supply chain and AI collaborate effectively, offering a reliable foundation for action. Gartner reported in 2026 that organizations with successful AI initiatives invest up to four times more in data quality, governance, AI-ready people and change management than organizations reporting poor outcomes. The foundation is not separate from AI value; it is what makes that value repeatable.
Early enterprise AI efforts focused on generating content, summarizing documents and helping people find information. Those uses can improve productivity, but the more consequential stage begins when AI evaluates information across systems, coordinates workflow steps and acts within ERP, supply chain and customer processes.
AI’s primary function is to interpret information and coordinate actions. The critical question is not only what an agent can do, but who authorizes it, on whose behalf and within what limits. Organizations need to define which data an agent may access, what it may recommend, what it may execute, when approval is required and how an action can be reviewed or reversed. NIST’s AI Risk Management Framework and Generative AI Profile treat governance, defined roles and human oversight as operating requirements. AI can perform work, but it cannot own the consequences.
Enterprise security has traditionally focused on applications, networks and user accounts. Those controls remain essential, but AI agents, service accounts and automated workflows create a chain of activity that can cross several platforms in seconds.
A compromised identity, excessive permission or poorly designed workflow can change records, initiate transactions or expose information across functions. Security therefore has to trace which identity accessed the data, which permissions were used, what action followed and whether the required approval occurred.
Security’s key role is to enforce least privilege, approval and traceability throughout that chain. Sensitive actions should be limited to the task, recorded for review and designed so teams can stop or reverse them. That control gives the business confidence to use AI in financial, operational and customer-facing processes without creating unacceptable exposure.
Consider a manufacturer facing a supplier delay. AI can identify affected orders and recommend moving inventory. ERP can enforce purchasing controls, supply chain systems can assess production impact and security can determine what the agent may execute. People still decide which customer takes priority, who approves the purchase and when an override is justified. That is where the operating model is tested.
People’s unique role involves judgment, challenging others and ensuring accountability. Mindset matters here, though not as a substitute for technology, governance or process design. Ricky Kalmon describes mindset as the software between your ears. Employees need to be ready to evaluate a recommendation, manage an exception and question the system when the context does not fit. As I wrote in “ERP Shifts to Industry 5.0 to Enable Smarter Human-Centric Operations,” the point is collaboration: technology handles more routine tasks as people shift toward strategic roles, focusing on judgment and exception handling. People remain the design center, not the part left over after implementation.
The most useful design unit is not the application. It is the decision. Start with one process where delay or exception cost is visible, such as an invoice exception, a supply disruption or a customer credit change. Map the required data, system actions, owner, approval thresholds, exception path and audit evidence from beginning to end.
Test the tough scenarios, not just when everything goes right. What happens when data conflicts, an approval is delayed or someone overrides the system? Focus on outcomes like faster cycles, fewer exceptions and lower costs, not the number of agents deployed.
Start with one decision, prove it works and build from there. The goal is steady progress, not full autonomy overnight. For many companies, waiting now carries more risk than modernizing.
The more immediate risk is layering AI onto fragmented processes and aging systems that cannot provide real-time context, integration or controlled execution. Gartner projects that through 2026 organizations will abandon 60 percent of AI projects that lack AI-ready data, and RAND research puts the enterprise AI failure rate above 80 percent, roughly twice that of conventional software, with weak process and decision structures cited more often than the models themselves. That forecast is already visible in the field. S&P Global Market Intelligence found that the share of companies abandoning most of their AI initiatives before production rose to 42 percent, up from 17 percent a year earlier.
ERP, supply chain, data, AI and security are not separate transformation destinations. They are parts of one operating environment. A decision may begin with data, be evaluated by AI, execute through ERP, affect the supply chain and require security controls throughout.
Vendors can make integration, identity management, auditability and decision controls easier to implement. Customers still have to define the work, the decision rights and the level of autonomy the business will accept. No platform compensates for an organization unwilling to change how work gets done.
The next chapter will be defined by whether the enterprise can turn trusted context into controlled action. That requires technology working across functions and people prepared to govern, challenge and improve it. Technology may drive transformation. People still determine the outcome.