Why AI Depends on ERP — and the Systems That Feed It
Much of the current anxiety around AI centers on whether its outputs can be trusted, and on the opacity of the inputs and assumptions models use to reach their conclusions. But the more important question is not whether we trust AI — it’s whether we are giving AI systems they can trust to operate within.
Every organization’s most powerful system of trust — often referred to as the single source of truth — is its ERP.
McKinsey recently put a sharper edge on a problem many organizations are already experiencing:
“Resources focused on AI are coming at the expense of enabling ERP to provide the system capabilities AI needs to thrive.”
Their research shows a widening gap. Companies are pouring capital and attention into AI experimentation while starving the very ERP capabilities—data quality, workflow logic, controls, and integration—that make intelligence operational. The result is predictable: pilot purgatory. Despite widespread AI adoption, only a minority of companies report any material enterprise-level EBIT impact.
McKinsey’s point is not that ERP is outdated. It is the opposite. ERP is operating DNA. It encodes clean transactional data, embedded business rules, end-to-end workflows, and institutional process knowledge. Those are not legacy artifacts. They are the prerequisites for AI agents to move beyond recommendations and safely participate in real decisions.
That distinction matters because AI only creates value when it is embedded inside workflows, not layered beside them. Agents must trigger on real events, read governed data, apply rules, and write outcomes back into the system of record. Without that, AI is just analytics with better language.
High performers understand this and don’t chase isolated use cases. They redesign entire domains—finance, supply chain, operations—so agents operate across complete workflows. And critically, they modernize ERP in a targeted way: buying standardized capabilities, customizing only where logic creates advantage, and tying everything back to measurable P&L outcomes.
Seen this way, AI is not a technology problem. It is a discipline problem. Intelligence scales only when it is anchored to systems of trust.
ERP does not create trust in isolation. It enforces it.
What often gets lost in the ERP conversation is that these systems are downstream. They assume that definitions, assumptions, and logic already exist. ERP is where rules are applied consistently, not where they are discovered.
Every ERP implementation is fed by upstream processes: planning models, reconciliations, scenarios, estimates, and judgment calls that precede formalization. If those inputs are weak, no amount of ERP rigor can compensate.
This is where many AI conversations break down. The focus stays on the system of record, while ignoring the systems that inform the system of record.
And in most organizations, that system has been Excel. Not as a replacement for ERP, but as the place where logic, assumptions, and data definitions are worked out before they are enforced.
Before ERP Enforces Trust, Logic Has to Be Built Somewhere
Excel’s enduring value has never been about formulas.
Formulas are syntax. They are the easiest part of the work to automate, abstract, or replace. What made Excel indispensable was not calculation, but context.
For decades, Excel has been where organizations reconciled:
- Data from multiple systems
- Differences in timing and granularity
- Conflicting definitions
- Assumptions that could not be encoded elsewhere
In other words, Excel became the place where meaning was imposed on raw data.
People who were “good at Excel” were rarely distinguished by their ability to write functions. They were distinguished by their ability to answer questions like:
- Which data source is authoritative?
- What does this number represent — operationally and financially?
- What assumptions are embedded here?
- What breaks if this changes?
Those judgments were rarely documented anywhere else. Excel didn’t just calculate results; it captured decision logic.
Excel as the Upstream Input to the System of Trust
This is why Excel persisted even as ERP systems matured.
ERP systems are excellent at enforcing rules once they are defined. But they are not where rules are discovered. They are not where edge cases are explored, scenarios tested, or assumptions debated.
Excel filled that gap.
It functioned as:
- A staging layer for logic
- A proving ground for workflows
- A translation layer between operations and finance
In many organizations, Excel models were where business rules were validated before they were encoded into ERP. That made Excel a quiet but critical extension of the system of trust — not a competing system, but a preparatory one.
Seen this way, Excel was never “shadow IT.”
It was pre-ERP logic, waiting to be formalized.
Why AI Makes Excel More Important, Not Less
AI changes the economics of execution, not the necessity of judgment. Tools like Copilot, Claude, and ChatGPT are exceptional at generating formulas and transformations. That matters — but it addresses execution, not definition.
As models become capable of generating formulas, transformations, and even workflows, the bottleneck moves upstream. The question is no longer how to calculate, but what should be calculated and why.
That upstream work still looks like:
- Structuring data correctly
- Defining grain and scope
- Reconciling discrepancies
- Making assumptions explicit
- Testing logic before it becomes operational
Those activities have always lived in Excel.
AI can accelerate execution once logic is sound. It cannot reliably invent the logic itself. If Excel disappears from that process, it doesn’t get replaced by intelligence — it gets replaced by implicit, unexamined assumptions embedded directly into automation.
That is not progress. That is risk.
Excel, ERP, and AI as a Continuum — Not Competitors
The mistake is treating Excel, ERP, and AI as alternatives.
They are not.
They are stages in a continuum of trust:
- Excel is where logic is explored, challenged, and refined
- ERP is where that logic becomes enforced, governed, and auditable
- AI is where that logic is scaled, accelerated, and applied at speed
When Excel is treated as disposable, organizations lose the place where understanding is built. When ERP is neglected, organizations lose the place where trust is enforced. When AI is layered on top of either without discipline, organizations scale confusion.
The systems that succeed recognize the handoff:
- Excel informs ERP
- ERP constrains AI
- AI amplifies outcomes
That sequence is not legacy thinking. It is the architecture intelligence requires.
The Real Risk
The risk is not that Excel survives too long.
The risk is that organizations confuse automation with understanding and deploy AI against foundations that were never coherent to begin with.
AI does not create trust.
It inherits it.
Excel still matters because trust still has to be built somewhere.