Why your AI and your ERP need to work together

How AI turned ERP modernization from a project you finish into a capability you continually evolve

For thirty years, the hardest question in any ERP program has been: when will it be done? Until recently, at least it had a clear answer. You picked a platform, configured it, trained people, went live, and closed the project.

Artificial intelligence (AI) has made that question unanswerable, because the premise is now obsolete. An enterprise resource planning (ERP) system used to be a system of record, a place that captured what already happened so people could report on it and reconcile it. Now AI is turning that same system into a tool for action: flagging an exception and resolving it, seeing a cost driver and responding to it, all without waiting for a person to catch up.

ERP programs have never been easy to land, and AI raises the stakes rather than lowering them. Most organizations are ambitious about AI and unprepared to act on it, especially in manufacturing, energy, and utilities, where the appetite to run leaner and predict better runs well ahead of the readiness to do it safely. That gap between AI ambition and AI value is, overwhelmingly, a breach in the data foundation underneath it.

TL;DR

How has the integration of AI changed the way ERP works across the enterprise?

AI has shifted ERP from a system that records what happened to one that acts on what is happening in real time, but not only inside the ERP. AI now works across a layered architecture: some agents act inside the ERP, others reason over a shared semantic layer that harmonizes data across your systems, and a coordination layer above the ERP orchestrates action across CRM, ERP, WMS, MES, EPM, and beyond. Your ERP becomes one system of action among several, and the governed one at the center. That single change cascades: the data has to be trustworthy enough to act on, the AI agents taking action have to be governed like users, and the people accountable for outcomes have to trust the system. And because AI keeps redefining what the system can do, ERP modernization is no longer a project you finish. It is a capability you operate, because AI structurally requires it.

The shift is away from a siloed installation and toward the ongoing coexistence of AI within and around your ERP. Understanding that arc is the difference between an initiative that scales and one that becomes another stalled pilot.

Five AI-led shifts that require thoughtful integration into your ERP program

As AI gets further and further embedded into all our tools and processes, including ERP, the potential consequences of less than perfect data, governance, and adoption become ever more stark.

The shift	What it means
AI does more than analyze; it acts across layers.	AI now pinpoints discrete signals and acts on them, across a layered architecture with your ERP as the governed system of action at the center.
Action raises the stakes on high-quality data.	To realize the value, your data foundation must be trustworthy enough for an agent to act on it autonomously and asynchronously, which is what the semantic layer is for.
AI agents are new ERP users.	Humans are no longer the only users of your ERP. Agents need identities, scoped access, and governance in the ERP and across the systems they touch.
Trust becomes a prerequisite to adoption.	If your team doesn't trust the data or how it's being acted upon, they'll find reasons to dismantle or ignore any system you stand up.
The pace of AI change keeps accelerating.	As AI-led change heightens, every part of your business needs closer, more frequent inspection. That includes your ERP.

01. Operational AI acts across a layered architecture, with ERP at the center

Start with what kind of AI capability you are actually looking to build.

Analytical AI generates insight. It consumes your data and tells you something: a forecast, a pattern, an anomaly. Then it hands the result to a person. It sits beside your ERP, pulling data as needed. The relationship stops there.

Operational AI goes a step further and acts on what it detects. It does not just flag that a production line went down; it reroutes the schedule. It does not just predict a late payment; it holds the order.

The market increasingly describes this as a spectrum rather than a two-box split: from pattern-matching and predictive analytics, through copilots and assistants, to task-specific agents and multi-agent orchestration, measured by how much autonomy you grant. We use “operational AI” as shorthand for the agentic end of that spectrum: AI that acts, not just advises.

Here is the part most modernization plans get wrong. They assume operational AI simply “runs through the ERP” because that is where the action lands. The reality in 2026 is more layered:

Some agents run inside the ERP, acting directly on transactional data. This is the right home for regulated, high-stakes actions that must stay close to the system of record and its controls.
A semantic layer sits alongside it, syncing and harmonizing data from the ERP and every other system so agents can reason over trustworthy, well-defined information. A growing share of the analysis and decision-making happens here, not inside the ERP.
A coordination (orchestration) layer sits above and outside the ERP, directing agents and actions across CRM, ERP, WMS, MES, EPM, and external tools, increasingly through open agent-to-agent protocols.

So your ERP is not the only place AI acts. It is one system of action among several — but a central, heavily governed one. The vendors themselves are split on how much to embed AI inside the ERP versus orchestrate above it, and the debate is genuinely unresolved as of 2026. The most durable position is a hybrid: keep regulated, high-risk actions close to the system of record, and orchestrate everything else across a layer above it. ERP governance stays central even as coordination moves outside the ERP.

That reframes the modernization question. It is no longer “can AI run through my ERP?” but rather “is my ERP clean, connected, and governed enough to be trusted as the system of action at the center of this architecture, and to feed the semantic and coordination layers around it?”

In plain terms

What is operational AI in ERP?

Operational AI takes action inside business workflows rather than only producing insights about them. In an ERP context, it triggers, routes, or completes a step of the process, which means the ERP becomes more than a data source it reads from; it becomes a system it acts through.

What is a semantic layer?

A semantic layer is a governed, business-friendly view of your data that sits between raw systems (ERP, CRM, WMS, MES, EPM) and the tools that use it. It harmonizes and defines your data (what a “customer,” an “order,” or “margin” actually means) so people and AI agents reason over the same trusted, well-labeled information instead of pulling conflicting numbers from separate systems. It is where much of an agent's analysis safely happens, and where the lineage and explainability your finance team wants actually lives.

For a manufacturer, that is the difference between a report telling you a line stalled and a coordinated response, with the ERP, MES, and warehouse systems rebalancing the run together before the shift is lost, while your team works higher-value, strategic levers.

For an energy operator, it is the difference between a maintenance prediction and a maintenance schedule that adjusts in real time as signals move across your operational and enterprise systems.

02. AI automation makes data quality even more important

As long as AI only analyzes, weak data is only a reporting problem. A wrong number produces a misleading chart, and a person catches it.

The moment autonomous AI agents begin taking action, weak data becomes an operational risk. One wrong number can produce a negative event, across a spectrum of severity. This is where the cost of a fragmented ERP rises from “misleading” to “damaging,” because there is no longer a human pause between the bad data and its consequence.

The logic here is unforgiving: AI inherits the gaps in your data rather than fixing them, so weak data no longer just distorts a report, it corrupts an action. AI agents are only as trustworthy as the data they use to inform their moves, which is exactly why the semantic layer matters: it is where fragmented data becomes actionable, governed trust that both agents and people can rely on.

The moment autonomous AI agents begin taking action, weak data becomes an operational risk.

For a CFO, this is concrete: an AI agent acting on month-old or unreconciled data could mislead the board, or approve a supplier payment against an invoice that was already disputed. For a supply chain leader, an agent acting on inaccurate inventory counts does not just distort a report; it places an unnecessary order.

03. AI agents: The new type of user your ERP was never designed for

Every ERP ever built assumed its end user would be a human on the other side of the screen, someone to interpret a number, sanity-check an entry, and pause when something looked wrong. Agentic AI removes that assumption.

We are now provisioning AI agents much the way we once provisioned employees: granting access, defining what they can touch, and deciding what they can do without supervision. And they are arriving fast, moving from isolated pilots into production across finance, supply chain, and operations, embedded directly inside the enterprise applications people already use.

Humans need to be in the AI loop without turning AI agents loose with no oversight on day one. A new governance layer is emerging to govern AI activity across all layers.

Modern cloud platforms have begun including pre-built agents, scoped access, and audit trails as of 2026, but many legacy and on-premise ERPs still have no concept of a non-human user, no agent-grade permission model, and no guardrail tuned for something that acts at machine speed without human judgment. For the mid-market firms running older systems, that gap is the live risk.

And the employee metaphor only gets you so far. Human governance is built around onboarding dates, managers, and offboarding. Non-human identities have none of those. They spin up and disappear constantly, and they already outnumber human identities many times over in most enterprises. That is why the security consensus is moving from static, provision-once access toward dynamic, just-in-time, runtime governance for agents. Treat the employee model as the on-ramp, not the destination.

The risk is not hypothetical. Treating agent governance as all-or-nothing is a common failure mode: lock the agent down and it is useless; trust it fully and it can act on bad data, or beyond its remit, at machine speed. A large share of agentic projects stall for exactly this reason, over cost, unclear value, and weak controls. Governing these agents is a question of organizational strategy and change management, the same way our human workforces have always been. Defining and evolving a cohesive governance layer that works with your ERP is a key component of enabling AI at scale.

04. Finance’s skepticism is the spec, not the obstacle

Is your finance team struggling to embrace autonomous AI? Modern advice often suggests finance teams need to “get on board,” but this function's insistence on traceable numbers and human-in-the-loop review is exactly the discipline that makes operational AI safe. Regulators, auditors, and risk functions are moving the same way, toward explainability and human oversight for automated decisions, preserving human authority over exceptions and high-risk cases.

Finance leaders in industrial, manufacturing, energy, and utility companies are demanding the same right now. The real gap is not that they are skeptical. It is that most ERPs cannot give finance the lineage, explainability, and control that would earn their trust, the very things a well-designed semantic layer and governed orchestration are built to provide.

Most ERPs cannot give finance the lineage, explainability, and control that would earn that team's trust.

When an AI agent proposes an action, can the controller trace the data it used? Can they see why the action makes sense? Can they intervene before it commits? If not, the problem is in the way your ERP and the layers around it were configured — not the skepticism. An AI-ready ERP is one built to satisfy finance's curiosity with auditability and human checkpoints designed intentionally and proactively, rather than bolted on as an afterthought. A modern semantic layer provides one source of the truth with true data lineage and auditability.

05. Today, ERP is less a project than an ongoing capability

You are no longer operating a single system, either. You are operating a living architecture made up of the ERP, customer relationship manager ICRM), and warehouse management system (WMS) integrated with a semantic layer and the coordination layers above it. And it keeps evolving. There is no stable end-state that would let you call an ERP project “finished.” So it moves away from being a “project” and toward an ongoing enterprise capability.

Project thinking vs. capability thinking

	ERP as a "project"	ERP as a "capability"
Time horizon	Defined end date at go live	Ongoing; evolves with AI and the business
Team	Assembled then disbanded	Standing function that operates the system
Configuration	Frozen at launch	Continuously optimized
Training	A one-time event	Ongoing, personalized, and tuned to AI shifts
Success measure	System is live	The business makes better decisions and acts more quickly

The companies that will struggle with AI in their ERP systems are the ones still budgeting, staffing, and governing ERP as a one-time capital project, with a team assembled toward a distant go-live and then disbanded, configuration frozen at launch, training wrapped by launch week. The companies that pull ahead will treat this work as a permanent function.

Connect with our expert

Greg Pitstick

Managing Director

Greg has more than 35 years of experience helping commercial, manufacturing, distribution, technology, and healthcare organizations achieve digital transformation, optimize their operations, innovate new global supply chains, and prepare them for the future.

View Profile

Contact Greg

Loading form…

Let's connect

Tell us a little about what you’re working on. We’ll connect you with the right experts to help you move forward.