The next phase of ERP is not a replacement. It is building connected ecosystems around the transactional core, where AI has clean, governed data to work with.
getty
A new wave of AI-native enterprise resource planning (ERP) startups is targeting a market dominated for years by Oracle, SAP, Microsoft, Infor, Epicor, QAD, Sage, NetSuite, IFS, Acumatica and Odoo. Investors are backing these companies because ERP, despite its size, is vulnerable to criticism for its complexity, high costs and prolonged implementation processes.
Reducing ERP complexity is not the same as replacing ERP. Most of what is being sold right now falls into the first category, not the second. That does not make it irrelevant. It simply means enterprises need to be clear about which problem is being solved, what risks are involved and where it fits within the broader operating model.
These startups understand the challenges ERP users face. Implementation timelines run long, interfaces age poorly and manual workflows persist in finance and procurement even when automation is available. Rillet, Campfire and DualEntry, among new AI-native ERP and accounting platforms, offer faster close processes, cleaner migration paths and automated reconciliation.
The challenge is that workflow tooling is not a system of record. ERP is the heart and transactional backbone of the enterprise. It runs the chart of accounts, procurement controls, inventory logic, manufacturing dependencies, supply commitments, payroll, compliance, financial close and the security and access controls that determine who can see, change or authorize what across every transaction. In regulated industries, that list also includes the documentation trails, audit requirements and industry-specific process controls built directly into the transactional core. That is not a workflow layer. It is operational control across business units, geographies and regulatory environments.
AI can speed up tasks already defined in a specific environment, but it cannot replace the essential governance that supports them. It does not substitute for approval processes, data ownership, audit logs or segregation of duties. Deploying agentic AI within ERP workflows without strict guardrails, immutable audit logs and a defined security posture is a liability, not a capability. That security posture includes enforcing role-based access at the agent level and containing breaches to prevent a compromised agent from propagating access across connected systems.
One risk that gets overlooked in ERP is vibe coding, using AI to generate code through prompts without much structure, governance or testing behind it. That can work in prototypes or lightweight apps where mistakes are not risky and easy to catch. ERP is the opposite. Poor ERP logic can create financial, inventory, procurement, tax or reporting issues across the business. In regulated industries, the risk gets much higher. In pharma, one mistake can violate FDA 21 CFR Part 11. Revenue errors can create SOX exposure, and poor data handling can create GDPR problems. Dropping AI into revenue recognition or financial controls without discipline is a real risk. ERP needs processes that are deterministic, repeatable and traceable. Mostly right is not good enough.
Every AI ERP vendor emphasizes speed. Data discipline determines whether that speed produces better decisions or faster mistakes. ERP systems work best when the data model underneath is consistent, well-governed and trustworthy. When item masters are inconsistent, supplier records are duplicated, cost structures are disorganized and process ownership is unclear, AI can make these problems worse rather than fix them. Moving faster with bad decisions does not make them better; it makes the mistakes more impactful. In industries with strict documentation or traceability requirements, such as aerospace, defense, life sciences and food and beverage, the data governance problem is also a compliance issue. AI operating on unclassified or ungoverned data in those environments does not just produce unreliable outputs. It can produce outputs that violate industry-specific documentation standards and create audit findings that take months to remediate.
In a January 2026 Forbes analysis of covering Salesforce’s research into enterprise AI adoption, I noted that weak data management consistently ranks among the top constraints on AI delivering real operational value. That holds across every ERP-heavy industry I cover. The constraint is rarely the model. It is the data quality, lineage, ownership, access control and policy enforcement underneath the model. Those are governance problems. They have to be solved before the AI layer can be trusted to act.
In a prior Forbes article, “How ERP Data Fits Into the Enterprise Data Ecosystem,” I made the case that ERP is no longer the sole core of the enterprise data ecosystem. This transition requires deliberate architectural choices that many enterprises have yet to adopt.
Snowflake, Databricks, Microsoft Fabric, Cloudera, Teradata, AWS, Google Cloud, IBM, Salesforce and Informatica are no longer just analytics infrastructure. They are becoming the connective layer between ERP systems, supply chain platforms and the AI services built on top. This is less about dashboards and more about making operational data governed, shareable, and usable by analytics and AI. At FabCon 2026, Microsoft made clear that Fabric is moving beyond analytics into a broader data control plane, with OneLake, databases, governance and AI operating as a more unified platform. SAP is moving in a similar direction with Business Data Cloud, positioning it as the layer that connects application context, governance and AI across the broader data estate. Oracle Fusion Applications data is natively integrated, and embedded AI, agents and agentic applications automatically inherit existing Fusion access controls and governance capabilities. This allows Fusion Agentic Applications to execute decisions within business processes while aligning with workflows, policies, approval hierarchies, permissions and transactional context.
When an organization consolidates transactional data from ERP, operational data from manufacturing systems and commercial data from CRM into a governed environment, AI applications become reliable. Demand forecasting improves. Procurement pattern recognition becomes more accurate. Anomaly detection in financial close shifts the organization from investigating problems after the fact to catching them in time to act.
KramerERP offers paid services to technology companies, similar to those provided by other tech-industry research and analyst firms. These services include research, analysis, advising, consulting, benchmarking, acquisition matchmaking, video and speaking sponsorships. KramerERP, an ERP industry expert, has collaborated with or is currently working with the companies listed in this article.
Infor and QAD offer examples of AI built around operational discipline rather than marketing position.
Infor’s argument for its Agentic Enterprise framework is that industry-specific context has to come before agentic automation, not after. “At Infor, our agents are grounded in micro vertical process catalogs and customer data,” said Vignesh Subramanian, Infor SVP of Product Management, Platform and Technology. “They know the difference between waste in food and beverage yield, manufacturing scrap and clinical waste. That precision eliminates hallucinations. But precision alone is not enough. When agents execute business workflows autonomously, the governance requirements are identical to those applied to humans doing the same job.”
That operational context keeps automation tied to reality and reduces the risk of hallucinations. It aligns far more closely with how complex manufacturing, distribution and supply chain decisions get made than most AI-native ERP pitches. Infor’s four-stage model concludes with governed velocity, treating governance as the control layer that enables autonomous execution. Its governance architecture, built around centralized identity, zero standing privilege, immutable logging and embedded GRC services, is planned for release in October 2026. That matters in Infor’s core verticals, where manufacturers operating under ISO requirements, food and beverage firms under FDA FSMA, and distributors managing trade compliance all need governance that supports auditability, traceability and human review, not just general enterprise security controls.
QAD addresses the same problem from a different starting point. Its acquisition of Redzone brought connected workforce technology directly into the QAD Adaptive ERP portfolio. Redzone captures frontline worker activity, production data and operational signals at the shop floor in real time. Manufacturing ERP has always struggled with the gap between what the system says should happen and what is happening on the floor. Redzone can help close that gap at the data-origin level, where the AI signal must be clean before anything built on top of it can be trusted. Its ChampionAI approach extends beyond the shop floor, using AI to automate workflows, migrations and modernization.
QAD reports implementation times have been reduced to a quarter of what they were previously. “Agentic AI is powerful, but by itself it does not modernize the enterprise,” said Sanjay Brahmawar, QAD CEO. “You cannot simply add lightweight or vibe-coded agents to a system of record where there is zero tolerance for error and expect real transformation. The real breakthrough comes from an AI-ready ERP foundation that enforces process discipline, maintains data integrity and turns decisions into action, not just insight.” That is the model that makes AI operationally useful in ERP. Not automation layered on top of inconsistent data. In automotive supply chains, where IATF 16949 and customer-specific requirements mandate full production traceability, that data-origin discipline is a certification requirement. Operational resilience means the system produces a trustworthy signal even when components fail.
The credibility test for AI-native ERP startups that claim to be alternatives to established systems is recognized. But can they demonstrate they have control, build trust and deliver consistent results at scale? A proof-of-concept or mid-market deployment with lower risk tolerances is not the benchmark. In large enterprises, processes such as financial closing, procurement controls and compliance reporting rely heavily on the system’s reliability across different quarters and regions. In regulated industries, the bar is higher. A startup ERP vendor without a documented track record in FDA-regulated manufacturing, aerospace documentation or financial services compliance is probably not a credible alternative to established vendors that have spent years building industry-specific certification support. Resilience and security are baseline entry requirements in those environments, not differentiators.
In enterprise deployments, AI is entering ERP from the edges rather than the core. It enhances planning, automates reconciliation, flags exceptions and accelerates manual workflows. It’s important to remember that this doesn’t replace the core systems that keep the enterprise running smoothly, financially and operationally.
The market opportunity lies in building connected ecosystems around it. In these setups, high-quality, well-governed data flows from the transactional core to the broader platform, providing reliable input for AI. Both established and new vendors that understand this key difference will remain important over the next five years. ERP remains at the heart of the enterprise. The quality of its data, the strength of its security architecture and the robustness of its industry-specific compliance determine whether the AI layer above adds true value or merely amplifies the underlying noise. This foundation isn’t the place to take shortcuts.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。