Earlier this year, McKinsey & Company published an article on the growing divide between AI agents and ERP systems, arguing that many organizations are struggling to realize meaningful value from AI because their core systems, data, and workflows are not ready to support it. The article is worth reading not because it introduces a radically new idea, but because it articulates a problem that many large enterprises are already experiencing in practice, often uncomfortably so.

The central observation is simple. AI ambitions are accelerating faster than the operational foundations that are meant to sustain them. Companies invest heavily in generative AI, agents, and intelligent automation, yet the underlying enterprise processes remain fragmented, inconsistently governed, and difficult to change. As a result, AI initiatives proliferate as pilots and proofs of concept, while sustained business impact remains limited and uneven.

This is not primarily a technology problem but an execution problem rooted in AI operationalization.

When AI ambition outruns operational reality

Across industries, organizations are reallocating budgets toward AI programs while postponing or minimizing investments in less visible but more consequential areas such as ERP modernization, data architecture, and integration and ERP integration for AI. The symptoms are familiar. Promising demonstrations that cannot be scaled. Intelligent recommendations that stop short of execution. Automation initiatives that still rely on manual intervention at critical points.

What goes wrong is rarely the quality of models or the creativity of use cases. The failure usually occurs where AI meets reality, namely in end-to-end workflows that cut across functions, systems, and external partners. In such environments, every weakness in the chain matters. An outdated ERP module, a brittle interface, an exception handled outside the system, or an ungoverned partner interaction can all become hard limits on what AI can safely automate.

ERP sits at the center of this tension as it encodes the operating logic of the enterprise. Orders, inventory, production, invoicing, settlement, and planning all converge there. AI agents that are meant to influence or automate decisions in these areas must ultimately rely on ERP data, rules, and transactional authority. Without a modernized and accessible ERP foundation, AI-driven workflows remain observational rather than operational.

This much aligns closely with McKinsey’s argument. Where the discussion needs to go further is in recognizing that ERP itself is only one part of the system.

From isolated use cases to domain level workflows

One of the most valuable shifts in thinking highlighted in the article is the move away from isolated AI use cases toward domain level workflow transformation. Instead of scattering AI initiatives across sales, finance, supply chain, and operations, leading organizations concentrate their efforts on a limited number of domains. Then they work through the full set of interconnected processes within them as part of a coherent ERP modernization strategy.

This shift matters because value is rarely created by a single intelligent decision in isolation. It is created when decisions propagate reliably through operational workflows and produce real world outcomes.

Consider a common supply chain scenario such as dynamic inventory allocation. The decision logic depends on master data, transactional data, configuration rules, and event signals that reside in ERP. Yet the business impact only materializes when the resulting decisions reach suppliers, logistics providers, and customers, and when their responses flow back into the system in a timely and interpretable way through SAP S/4HANA integration and B2B channels.

Designing an AI-enabled workflow requires working backwards from the decision that should be made and executed, and explicitly identifying the ERP objects, events, and rules it depends on. It should also include the external messages and partner events that must be incorporated and governed. This is fundamentally different from building a model and assuming integration can be addressed later.

At this point, B2B integration stops being a secondary concern and becomes part of the core design problem.

Ontology does not stop at the ERP boundary

The McKinsey article also emphasizes the importance of a shared ontology. This refers to a consistent way of representing business entities, relationships, and processes so that AI systems operate on a stable and unambiguous understanding of the enterprise. Grounding AI in ERP data models and business rules reduces inconsistency. It also keeps automated decisions aligned with how the business actually runs.

In practice, however, most large enterprises do not operate solely within the boundaries of their ERP systems. Their operating model extends across suppliers, contract manufacturers, logistics providers, distributors, banks, and customers. All of these groups generate data and events that influence decisions and must be incorporated into AI ERP integration patterns.

If the ontology stops at the ERP boundary, a substantial part of the business reality remains invisible or poorly represented. Supplier identifiers, customer references, document states, and exception events must reconcile cleanly with internal master data and process states if AI-driven decisions are to be trusted. External events such as delays, partial shipments, or rejections must be represented in a way that AI systems can interpret consistently and act on without ambiguity.

In this sense, an ERP grounded ontology becomes, in practice, an ERP plus network ontology. Trustworthy and scalable AI depends on semantic alignment not only inside the enterprise, but across the extended business network.

Embed AI in workflows to cross the network boundary

A recurring theme in the article is that AI delivers value when it is embedded directly into the steps where work happens, rather than being positioned as a detached analytical layer. Approvals, planning cycles, and exception handling are cited as examples in which AI can augment or automate decisions in context.

Yet many of these workflow steps originate outside the ERP system. An approval may relate to a price discrepancy on a supplier invoice. An exception may be triggered by a carrier status message. A recommendation may involve adjusting collaboration with a key customer or supplier.

If AI is embedded into ERP workflows while the B2B edges are treated as static or secondary, automation remains partial and fragile. Manual workarounds persist precisely where partner behavior, data quality, and timing matter most. This is where much of the potential value of ERP integration for AI is lost.

Why B2B integration becomes a first-class dependency

In traditional ERP modernization programs, B2B integration is often discussed as a necessary set of interfaces that must not be forgotten. The AI and ERP conversation creates an opportunity to reconsider this hierarchy.

When AI agents are expected to act on real-world signals and execute decisions across organizational boundaries, B2B integration becomes part of the execution fabric. External transactions and events are no longer just messages to be processed. They are triggers for decisions and actions that must be auditable, governed, and reversible if needed.

For OpenText Business Network, this reframing is significant. The value is no longer limited to connecting systems but to providing a stable and semantically aligned interaction layer between ERP and the extended enterprise. This allows AI-driven workflows to operate reliably even as models and logic evolve.

Partners do not need to adapt to every AI experiment. They interact through a consistent network surface, while intelligence is introduced and refined behind it.

ERP modernization to SAP S/4HANA as a forcing function

For organizations moving to SAP S/4HANA, these questions are no longer theoretical. The shift toward a more event-driven, API-oriented ERP architecture exposes weaknesses in legacy integration approaches. These were previously masked by batch processing and custom code, particularly in the area of SAP S/4HANA integration with external partners.

This creates a narrow window of opportunity. ERP modernization can reinforce existing fragmentation by lifting old integration patterns into a new system. Alternatively, it can be used to redesign workflows across ERP and network boundaries with AI readiness in mind.

Approach ERP modernization with B2B integration and AI to sequence change more deliberately, and reduce rework and avoid dead ends that limit future AI operationalization.

Close the AI, ERP, and network gap

The most important takeaway is that the challenge is not limited to bridging AI and ERP. There is a broader divide between intelligent systems, core enterprise platforms, and the external networks in which businesses actually operate.

AI agents require clean, timely, and semantically consistent data and event from both internal systems and external partners. This is essential if they are to move beyond impressive demonstrations into sustained operational impact through AI ERP integration.

For organizations modernizing ERP, this is an opportunity to address ERP, B2B integration, and AI as one connected transformation. With OpenText Business Network, this transformation is not plumbing but part of the operational fabric that makes intelligent, domain-level workflows possible across the extended enterprise.

Value, in the end, is not created by intelligence alone, but by intelligence that can act.

The post AI, ERP, and the missing middle: Why integration determines whether modernization delivers value appeared first on OpenText Blogs.