This article was originally published on LinkedIn.
We're Seeing Real ROI — But We've Only Scratched the Surface
Across finance, operations, and supply chain, the impact of AI is already clear. We're seeing faster closes, smarter forecasting, sharper fraud detection, and huge productivity gains from tools that can summarize, reconcile, and explain data in seconds.
For companies running on platforms like NetSuite, AI has moved well beyond buzzword status. It's now delivering measurable ROI.
And yet, as powerful as these early wins are, they represent only a fraction of what's possible.
Most ERP-AI integrations today still primarily focus on automation and reporting. They help us see the data faster, but not necessarily think with it more intelligently.
I believe the next wave will go much deeper, and it'll happen when AI is woven directly into the ERP itself. Not just as a reporting assistant, but as a reasoning partner. That's when the ERP stops being a system of record and starts becoming a system of understanding.
Below are ten possible applications of AI in ERP that I think we'll see in the future, and for each one, I've noted what I believe it will take to make it a reality.
1. From Queries to Intent
Most ERP users don't really want another report. Instead, they want clarity. They're asking questions like:
"Why did margin drop last week?" "Which products are becoming unprofitable?"
An intent-driven ERP would interpret those questions and deliver the reasoning, not just the numbers.
How We Get There
- A semantic layer mapping natural language to ERP logic.
- Fine-tuned LLMs trained on accounting and operational vocabulary.
- Conversational interfaces built into workflows.
Feasibility: High. All of the necessary components exist. So this is really more of an integration challenge, not a research problem.
2. Dynamic, Self-Adjusting Policies
Every ERP hides static rules: approval limits, credit thresholds, discount caps. AI could monitor context and self-tune those policies, tightening during cash crunches, loosening when liquidity is strong.
The result would be an ERP that behaves like a living system, continuously balancing safety and agility.
How We Get There
- Continuous monitoring agents tied to KPIs.
- Policy simulators to test changes safely.
- Audit-ready guardrails that log every AI adjustment.
Feasibility: Medium. From a technical standpoint, this isn't all that difficult. But organizations will need to build trust and governance first.
3. A Memory for Decisions
ERPs remember what happened, but not why things happened. AI could create a narrative memory connecting data changes to their context.
Then we'd be able to ask questions such as:
"Why did we switch vendors last spring?" "Who approved that exception, and what was the reasoning?"
The ERP would become a living record of organizational intent.
How We Get There
- APIs linking ERP logs with collaboration tools like Slack and email.
- LLM summarization to extract decision rationales.
- Knowledge graphs to store relationships between events.
Feasibility: Medium-High. The tech is available now. I think that privacy and retention policy are going to be the challenging parts.
4. Modeling Culture and Ethics
Every company's culture leaves fingerprints in its ERP: who overrides approvals, who delays entries, who cuts corners.
AI could surface those signals, helping leadership detect bias, burnout, or ethical drift early. It's not so much about surveillance as it is organizational awareness.
How We Get There
- Behavioral signal tracking (approvals, timing, frequency).
- Ethical baselines defining "healthy" behavior.
- HR collaboration to interpret the signals responsibly.
Feasibility: Low-Medium. While this would be technically simple, I think the challenge is going to be implementing this in a way that is culturally sensitive.
5. Cross-ERP Collaboration
Many large enterprises run multiple ERPs, and they do so as a result of acquisitions, divisions, or global operations. Integrating them can be slow, expensive, and fragile.
AI could act as a semantic interpreter, understanding how each ERP describes the world and translating across systems. That could finally make "multi-ERP" mean "single source of truth."
How We Get There
- Schema embedding models that learn each system's metadata.
- Ontology mapping to align concepts ("Customer" = "Client").
- Validation layers for human review.
Feasibility: Medium. This concept is already emerging in next-gen integration platforms.
6. Simulating Alternate Histories
What if we could replay history with different assumptions? Then we could use AI to simulate how results might have been different if things like pricing, payment timing, or vendor choices had been different.
This would turn the ERP into a decision simulator, a sort of "sandbox" for strategic foresight.
How We Get There
- Causal modeling trained on historical data.
- Scenario simulation engines embedded in financial modules.
- Explainable outputs that quantify trade-offs.
Feasibility: Medium-High. This is already taking shape in FP&A tools; ERP-native versions are next.
7. Preventing Anomalies Before They Happen
Most anomaly detection is reactive. It finds issues after they occur. But AI can go further, identifying precursors to problems: rushed approvals, irregular vendors, late-night activity.
Think of it as a self-learning safety net.
How We Get There
- Real-time behavioral monitoring.
- Predictive classifiers trained on past errors.
- Adaptive UX that nudges users before submission.
Feasibility: High. The data already exists. It just needs to be applied in real time.
8. Modeling the Company as a Micro-Economy
Every organization is essentially a small economy, with supply, demand, liquidity, and cost flows. AI could model those internal forces to optimize working capital, internal pricing, and resource allocation.
ERPs would become self-tuning economic models, not just ledgers.
How We Get There
- Cross-module data graph linking finance, inventory, and HR.
- Agent-based simulations that test changes safely.
- Visualization layers to show dynamic equilibrium.
Feasibility: Medium. I think this is already computationally feasible, but its success depends on data maturity.
9. The AI CFO
The CFO role is evolving very quickly, from reporting to reasoning. AI copilots are already producing board decks, running forecasts, and reconciling data. Soon, they'll manage liquidity, simulate scenarios, and surface insights autonomously.
I think we'll see the CFO role evolve from operator to strategist, guiding a finance function that's increasingly enhanced by AI.
How We Get There
- Autonomous process agents for close, forecast, and reporting.
- Trust and validation layers for compliance.
- Conversational analytics for interactive decision-making.
Feasibility: Medium-High. This is already underway in some large enterprises.
10. Measuring the Mood of the Business
ERPs quietly record human rhythm, including approvals, exceptions, delays. In the future, AI could interpret that rhythm to measure organizational mood.
- Slower approvals → fatigue
- Rising exceptions → stress
- Sudden spend changes → friction
This "Business Mood Index" could give leaders a real-time pulse on organizational health.
How We Get There
- Behavioral signal extraction (time, tone, volume).
- Correlation models linking to engagement data.
- Privacy-safe design for ethical use.
Feasibility: Low-Medium. This is technically possible now, but it requires strong ethics governance.
The Road Ahead: From System of Record to System of Understanding
We've already proven that AI can deliver measurable ROI. Now it's time to move beyond efficiency, toward understanding.
When AI stops just summarizing data and starts reasoning about it, the ERP transforms from a system of record into a system of insight, and eventually into a system of reasoning.
I believe that's the new frontier: the moment when ERP and AI stop existing as separate tools, and start thinking together. And when companies reach that moment, they won't just run smarter businesses, they'll run more human businesses.