Everyone wants their ERP to be smarter. Better dashboards. Better forecasts. Better answers to "what should we do next?"
But here's the thing: your ERP was never designed to answer that question. It was designed to record what happened, enforce your business rules, and give you a single version of the truth. Somewhere along the way, we started expecting it to also be the brain of the organization. And it tried. But that's not where it shines.
What I'm seeing now, across the NetSuite ecosystem and ERP more broadly, is a quiet but significant shift. AI is taking over the insight function. ERP is settling back into what it does best: being the authoritative system of record.
This isn't ERP's decline. It's ERP's specialization.
How We Got Here
ERP started as an integration play. Before these systems, most companies ran separate applications for finance, manufacturing, HR, and supply chain. Each had its own data model and its own version of the truth. ERP promised one platform where transactions would be recorded once and flow consistently across functions.
That promise was largely delivered. But as adoption grew, so did expectations. Leaders didn't just want accurate data. They wanted insight from that data. So reporting modules, analytics dashboards, and decision-support tools got bolted onto transactional cores. Vendors obliged, embedding BI capabilities and, more recently, machine learning features directly into their platforms.
The result? A system asked to serve two masters: operational fidelity and strategic intelligence. It handled both, but rarely excelled at the latter. Native analytics were rigid, slow to adapt, and constrained by the transactional data model. Teams wanting richer insight typically extracted data into warehouses or lakehouses, where it could be shaped for analysis.
ERP remained the source of truth. But the source of insight lived elsewhere, often in fragmented, hard-to-govern configurations. Sound familiar?
What AI Changes
Three things are accelerating the separation.
Agentic AI has matured. LLMs and agentic systems can now reason across structured and unstructured data, generate recommendations, and execute multi-step workflows with limited human intervention. We've moved past assistants that answer questions. We're now building agents that complete workflows.
Composable architecture has moved from theory to practice. The idea of assembling capabilities from modular, interoperable components rather than relying on a monolithic suite is no longer a conference-talk abstraction. It's how forward-looking teams are building. ERP becomes one component in a broader, API-connected ecosystem. It does what it does best and delegates the rest.
The data infrastructure got cheaper and faster. Cloud-native platforms can ingest ERP data in near real-time and make it available to AI systems without the batch-processing delays that used to hobble analytics. The technical barriers to decoupling insight from record have dropped significantly.
Agents as the Insight Layer
The clearest sign of this shift is the rise of AI agents embedded in enterprise workflows. Every major ERP vendor is racing to deploy them: Microsoft's Copilot agents in Dynamics 365, Oracle's 50-plus agents in Fusion Cloud, SAP's Joule assistant, IFS's agent orchestration framework.
These aren't chatbots. They're designed to take autonomous action within guardrails. A supplier communication agent might parse vendor emails, update orders, and escalate exceptions without anyone clicking a button. A financial reconciliation agent might identify discrepancies, propose adjustments, and prepare audit documentation.
The pattern is the same everywhere you look: ERP holds the authoritative data. The agent generates the insight and initiates the action.
I used to think of this as ERP getting smarter. Now I think the more accurate framing is that ERP is getting a partner. The insight work is moving to systems purpose-built for reasoning, while ERP handles what it was always meant to handle: the record.
What the Architecture Looks Like
In practice, here's the pattern that's emerging:
ERP stays the system of record for core transactions: orders, invoices, payments, inventory movements. Its value is data integrity, audit trails, and regulatory compliance. It answers "what happened."
A data platform (warehouse, lakehouse, or streaming layer) aggregates ERP data with signals from CRM, e-commerce, IoT, and external sources. It's optimized for analytics, not transactions.
AI agents and analytics tools operate on that data platform to surface anomalies, generate insight, and recommend actions. They answer "what should we do." Some push recommendations back into ERP for execution. Others orchestrate across multiple systems.
This is specialization, not replacement. ERP's role as source of truth isn't diminished. It's clarified.
Where ERP Still Owns the Room
It's worth being explicit about what ERP is still the right tool for. This isn't a "move everything to AI" argument.
- Transaction processing. Recording orders, payments, shipments, and inventory movements with guaranteed consistency.
- Master data governance. Maintaining authoritative records for customers, vendors, products, and org hierarchies.
- Regulatory compliance. Providing auditable, tamper-evident records for financial reporting, tax, and industry-specific requirements.
- Process enforcement. Ensuring that approval workflows, segregation of duties, and tolerance limits are applied consistently.
None of that is trivial. It's the foundation everything else depends on. AI agents generating recommendations from corrupted or inconsistent data will produce confident nonsense. The separation of concerns only works because ERP continues to do its job well.
The Trade-Offs
This shift isn't free. A few things to weigh honestly.
Integration complexity goes up. Decoupling insight from record means more moving parts: data pipelines, API connections, sync logic, agent orchestration. If your integration discipline is weak, this gets harder, not easier.
Data latency becomes visible. When analytics lived inside ERP, data was more or less current. With external AI layers, you have to manage the gap between when a transaction posts and when it's available for analysis. For some use cases, near-real-time is fine. For others, it's not.
Governance has to grow up. AI agents making autonomous decisions need clear guardrails, audit trails, and escalation paths. Bad data in means bad actions out. If your data governance isn't mature, you won't trust what the agents tell you.
Explainability gaps persist. ERP reporting, for all its limitations, was deterministic and auditable. AI-generated recommendations often aren't. You'll need to decide which decisions require explainability and which can tolerate probabilistic reasoning.
Questions Worth Asking
If you're evaluating ERP modernization or AI integration, here's where I'd start:
Where is your insight actually generated today? If it's already in external BI tools, data warehouses, or ad hoc spreadsheets, you've already separated insight from record. You just haven't formalized it.
What is your ERP's real value? If you stripped out reporting and analytics, would the transactional core still justify its cost? For most organizations, the answer is yes. That tells you something about where to invest.
Do you have the integration maturity for a composable model? If your organization still treats integration as a one-time project rather than an ongoing capability, composable architecture will disappoint.
How much autonomy are you prepared to grant agents? The value of agentic AI scales with the scope of decisions it can make. If every action requires human approval, the efficiency gains will be modest.
There's no universal answer here. Organizations with stable, mature ERP environments may find that embedded AI features deliver enough value without architectural upheaval. Those with fragmented landscapes or aggressive AI ambitions may benefit from a more deliberate separation.
Specialization, Not Displacement
The narrative that AI will replace ERP misreads the situation. ERP isn't being displaced. It's being relieved of a burden it was never ideally suited to carry.
This is what mature technology ecosystems do. Databases didn't disappear when data warehouses emerged; they specialized. Operating systems didn't vanish when application platforms abstracted them; they became infrastructure. ERP is following the same trajectory: from monolithic suite to specialized backbone, valuable precisely because it no longer tries to be everything.
The practical shift for leaders is to stop asking "How do we get more insight out of ERP?" and start asking "How do we build an insight architecture that treats ERP as its authoritative data source?"
The answer will involve AI agents, analytics platforms, and composable integration. But it will still run through ERP's transactional core. That's not a limitation. It's a foundation.