Nashville brought the biggest Epicor announcements in years: six AI capabilities, a world model, headless ERP, and a migration program promising 90-day go-lives. Here’s what it means and what to watch.

KEY ARTICLE HIGHLIGHTS

  • Epicor Insights 2026 drew 4,100 attendees to Nashville — a 37% jump from 2025 — signaling a community energized by, or anxious about, the company’s accelerating cloud-and-AI pivot.
  • The company announced six AI capabilities simultaneously: Industry Data Model, Vertical Agents, Lux Agentic Design System, Agent Foundry, Headless ERP, and Time to Value via the Ascend program, claiming to be the first ERP company in the world to ship all six at once.
  • Headless ERP was the sleeper announcement: Epicor is decoupling its business logic from its UI, allowing external AI tools like Claude and ChatGPT to interact with ERP data through open MCP connections, a structural shift in what “running ERP” means.
  • The Prism Agent Foundry lets customers build, verify, and share their own AI agents inside the Epicor ecosystem, a marketplace play that could meaningfully expand Epicor’s stickiness if adoption follows.
  • The Freight Spend Intelligence agent surfaced $700,000 in two-year freight variance for one customer and drove that figure down from 19% to 1.5% after deployment, one of the few Insights demos anchored to a real, auditable business outcome.
  • CEO Steve Murphy opened with a billing apology and an unusually honest AI primer, including an acknowledgment that the best models hallucinate more than 10% of the time. The credibility spend before the product claims was deliberate and effective.
  • Agentic AI pricing remains unsettled. Epicor confirmed in the analyst Q&A that it is still working out outcome-based pricing models. Customers negotiating new or renewal contracts should demand specific terms before signing.
  • The cloud migration barrier is falling: Cornell Pump migrated from NetSuite to Kinetic in two weeks using the AI-assisted Ascend program. Epicor is targeting a 90-day qualified go-live standard.

The majority of the 4,100 attendees who showed up at the Gaylord Opryland for Epicor Insights 2026 are the kind of customers who’ve been running Epicor software for a decade or longer; manufacturers, distributors, and building supply companies for whom the platform is embedded in daily operations. They weren’t evaluating alternatives. They were there to understand what Epicor is becoming, and what that means for their businesses.

What Epicor is becoming, if you take the keynotes at face value, is something materially different from what it has been: an ERP company that can act on data, not just record it. The announcements at Insights 2026 are arguably some of the most consequential the company has made in years. Some are shipping now. Some are on the near horizon. And some are directional bets whose value won’t be clear for another 12 to 24 months. All of them are worth understanding.

Before the Product Slides: Steve Murphy Sets the Table

Epicor CEO Steve Murphy opened the Insights 2026 keynote with a move that was either brave or strategically calculated — probably both. Before a single product announcement, he candidly addressed, and apologized for, the billing problems Epicor’s customers have been experiencing. Directly. By name. He committed to personal involvement until they were resolved and offered customers access at a Thursday support session for customers who wanted to discuss it.

That’s not a typical conference opener. For a room of long-tenured customers with strong opinions about vendor support quality, it landed right. More importantly, it bought credibility for everything that followed.

The second move was an AI primer that was likewise unusually honest. Murphy laid out five positions before touching a product slide:

  • AI is powerful and frightening simultaneously, and anyone telling you it’s purely one or the other isn’t being straight with you.
  • Layoffs attributed to AI mostly reflect over-hiring or market share loss, not automation displacement. AI is replacing parts of what people do, not entire roles.
  • Net-Net: AI will create more jobs than it eliminates, including businesses we haven’t imagined yet.
  • The knowledge gap between what people write about AI and how it actually works is real. Murphy explicitly called out the tendency to make things up when you don’t understand the underlying mechanics.
  • Creativity remains a human competitive advantage. The technology is a mirror of what’s out there. It has no creativity of its own.

“The best models hallucinate more than ten percent of the time. Some are closer to ninety. If you hired someone who said they were great at the job but hallucinated between ten and ninety percent of the time, you’d worry about that.”

— Steve Murphy, CEO, Epicor — Insights 2026 Keynote

Murphy also addressed something that matters strategically for an ERP company making AI claims, and that’s the problem of recursive errors — AI training on its own outputs, reinforcing hallucinated content as fact. His point was direct and applicable: the ERP system of record is exactly the kind of grounded, structured, authoritative data source that gives AI something real to work with. That’s not just vendor positioning. It’s architecturally accurate.

Murphy shared that Epicor has more than doubled in size over the past nine years and tripled its R&D budget to over $200 million. For a private company that doesn’t file a 10-K, this is the kind of self-reported financial signal the market appreciates. Murphy also committed that Epicor’s R&D investment level will continue.

Cognitive ERP: Six Capabilities, One Architecture Argument

Epicor President and Chief Product & Technology Officer Vaibhav Vohra had the most architecturally substantive portion of the keynote. His argument wasn’t just that Epicor has added AI features — it was that Epicor has built a coherent stack that connects industry-specific data structures to AI action in a way that general-purpose tools cannot replicate. That’s a bolder claim, and it’s worth examining both what supports it and where it remains to be proven.

Epicor’s Cognitive ERP framework sits on top of what the company calls its Industry ERP Cloud and Data Engine and comprises six interconnected capabilities announced simultaneously at Insights 2026. These include:

  • 6 Cognitive ERP capabilities shipped simultaneously
  • 4,100 Insights 2026 attendees (up 37% YoY)
  • $700K Freight variance recovered by one customer demo
  • 2 wks Cornell Pump migration from NetSuite to Kinetic

Industry Data Model (Data Ontology)

The Industry Data Model (Data Ontology) is Epicor’s foundational differentiation claim, and it’s the most analytically interesting one. The company argues that its 50+ years of manufacturing and distribution data have produced a structured industry-specific ontology — not just a database schema, but a semantic model of how manufacturing objects (parts, jobs, routings, vendors, warehouses) relate to each other in real production contexts. The argument follows: AI without ontology is guessing. AI grounded in that ontology is accurate, actionable, and auditable. Every answer traces to a specific job, part, or operation. Every action respects existing ERP rules and approval flows.

If this holds up operationally, and I believe that’s the test every customer and analyst should be applying, it is a genuine competitive moat. No horizontal AI vendor selling into manufacturing has built this from the transaction layer up. The question is whether the ontology advantage produces demonstrably fewer errors and less human correction than generic alternatives in live customer environments. That’s not a conference claim, it’s a 12-month proof-of-concept.

Vertical Agents (Epicor Prism)

Epicor Prism is the company’s network of vertical AI agents embedded directly into Kinetic and Prophet 21. At Insights 2026, Prism agents were demonstrated performing real tasks: generating sales orders via natural language, auto-setting order priority when expedite dates change, reconciling freight invoices against carrier manifests, running MRP explainability analysis, and executing shop floor job resequencing. These were live demos against actual customer environments, not staged mockups.

The new agents announced at or ahead of Insights 2026 include the Freight Spend Intelligence Agent, the MRP Log Agent (which explains why MRP made specific recommendations), a Quoting Intelligence Agent that demonstrated a 92% same-as-except match rate and 41-minute quote cycle for job shop scenarios, and a Supply Chain Optimization Agent capable of reducing late jobs from 8 to 2 and improving on-time delivery by four percentage points in a demo simulation.

The Prism Business Rules Agent — demonstrated live — allows users to generate new ERP business rules in natural language. In the demo, the agent interpreted a request, identified the relevant fields (expedite date, order priority), drafted the conditional logic, displayed the code, and deployed the rule to Prophet 21 — all within the Prism interface. The underlying code remained accessible to developers who wanted to verify or modify it. That’s the right approach: AI as a productivity accelerator, with human oversight preserved.

Lux Agentic Design System

The Lux Agentic Design System is the governance and UX layer for how humans and AI agents interact with Epicor systems. Chief Innovation Officer Arturo Buzzalino described it as not just buttons and colors, but the mechanism by which agents create context-appropriate interfaces dynamically — an approval flow when that’s what’s needed, a dashboard when data visualization is the task, a recommendation when analysis is required. In a headless ERP model, Lux is what ensures that AI-generated experiences look, feel, and behave like trusted Epicor products rather than ad-hoc outputs.

Lux also governs the agent publishing process: when customers build custom agents via the Agent Foundry and submit them to the Prism Agent Market, Lux enforces a consistent security and UX standard. It’s the design contract that makes a marketplace viable.

Agent Foundry

The Agent Foundry is Epicor’s platform for customer-built AI agents. Using natural language prompts and guided configuration, customers can build agents tailored to their specific business context, deploy them inside their Epicor environment, or submit them for review and publication to the Prism Agent Marketplace. The marketplace itself, a channel inside the broader Epicor Marketplace launching later in 2026, allows customers and partners to discover, deploy, and in the future transact on certified agents.

The Agent Foundry demo was notably specific: a supply chain optimization agent built live, submitted for compliance scanning, deployed, and run against actual data, reducing revenue at risk from a simulated $1.2M to $125K by resequencing jobs and expediting purchase orders. That’s hard not to find impressive. The governance layer (code scanning before deployment, ERP security integration, Lux design enforcement) is what I see that separates this from a generic AI playground. Whether that governance holds as agent complexity increases is a question worth monitoring.

Headless ERP

I felt like Headless ERP was the sleeper announcement from Insights 2026, and arguably the most strategically significant. Headless ERP means Epicor’s business logic, data structures, and transactional system can now operate without requiring the standard Kinetic or Prophet 21 interface. External AI tools, external agents, and external platforms can interact with ERP data and trigger ERP actions through governed API connections.

The mechanism is Model Context Protocol (MCP) — the open-source standard for AI-to-system connectivity that has gained rapid adoption across the enterprise AI ecosystem. Epicor has explicitly embraced MCP rather than building a proprietary integration wall, and I think that’s a smart move. In the Insights 2026 demo, a manufacturing persona ran a part substitution impact analysis through Claude connected to Kinetic, then created a follow-up action item inside Prophet 21 through ChatGPT. The ERP data model provided context and governance; the external AI provided the interaction layer.

“We are opening up Prism’s agentic capabilities and Epicor’s vertical ERP ontology so outside agents can interact with Epicor in a governed, secure, and permission-aware way. Whether the agent starts inside Epicor or outside, Prism becomes the trusted layer.”

— Arturo Buzzalino, Chief Innovation Officer, Epicor

The strategic implication here is significant: Epicor is positioning the ERP core as infrastructure, not interface. Companies that want to build AI-native workflows on top of trusted, structured transactional data can now do so without being locked into Epicor’s UI paradigm. That’s a meaningful architectural shift. It also raises a legitimate governance question that Epicor will need to answer as adoption scales: what happens when external agents make mistakes inside the ERP? What are the rollback mechanisms, the audit trails, and the liability boundaries?

Time to Value — Ascend Program Expansion

Epicor expanded its Ascend cloud migration program with AI-assisted data mapping that automatically identifies schema relationships between legacy systems and Kinetic, generates default values, eliminates duplicate records, and enables users to define data scope (e.g., migrate only the last two years of order history) without SQL expertise. The program is currently consultant-led, with self-service capabilities on the roadmap.

The headline customer example: Cornell Pump migrated from NetSuite to Kinetic in two weeks. That number will rightly get scrutiny, and obviously not every migration will be that speedy. Cornell’s implementation had specific characteristics (clean data, relatively standard configurations) that enabled the speed. But the directional signal is real: Epicor is investing heavily (reportedly 50-60% of total AI spend) in migration tooling, and the gap between the on-prem experience and the cloud experience is widening every quarter.

Epicor is targeting a 90-day qualified go-live standard for new cloud implementations via Ascend. That claim, if it holds across a meaningful customer sample, would be a material competitive differentiator against peers whose implementation timelines still run 12 to 24 months.

The Industry World Model: Epicor’s Long-Term Data Bet

Vohra’s most ambitious announcement, and the one with the longest runway before it can be evaluated, was the Industry World Model.The concept: Epicor holds anonymized, opted-in transaction data from a significant share of North American manufacturers and distributors. If aggregated and structured correctly, that dataset becomes a macro-level intelligence layer that no individual company — and no general-purpose AI vendor without ERP-level transactional grounding — can replicate. That’s clearly a pretty big deal.

Vohra cited a specific example: Epicor’s aggregated auto parts distribution data can predict Bureau of Labor Statistics motor vehicle parts CPI movements more accurately and earlier than the public index itself. It can also signal, in operator language, when the channel is shifting from clearance mode back to pricing power, or when real demand weakness is separating from channel shift. He framed this as the difference between reporting (what happened) and prediction (what operators need to do next, before public data catches up).

This is a legitimate strategic asset if Epicor can execute on the governance requirements: opt-in participation, rigorous anonymization, community value return, and transparency about what is and isn’t in the model. The analyst community should be watching whether Epicor publishes methodology, whether the macro signal claims are independently verifiable, and whether customers actually see the benefit or just the pitch.

The World Model is also an answer to a question that should be on every enterprise AI buyer’s mind: what is the sustainable moat for an industry-specific AI product? Epicor’s answer is the data network that accrues from being the transactional system of record for manufacturing and distribution at scale. That is a more defensible position than “we fine-tuned a general-purpose LLM on some industry docs.”

What Nashville Left Unanswered

Insight into what’s real requires noting what the keynote didn’t resolve and there are a few areas that I believe warrant scrutiny. These include:

Agentic AI Pricing Is Still Being Figured Out

This matters more than it might appear. Epicor acknowledged in the analyst Q&A that outcome-based pricing for AI agents is still being worked out; no firm structure, no committed caps, no overage policy. The direction is outcomes-based: charges trigger when an agent completes a defined task. But as we discussed with execs in the executive Q&A, “still working it out” at this stage of deployment is a material gap for customers writing contracts today. If you’re negotiating a new or renewal agreement that includes Prism or agent capabilities, you need specific pricing terms in writing before you sign a contract. As we all know, what looks affordable at current AI model costs may look very different in 18 months as usage scales.

The On-Prem International Question Got a Non-Answer

Epicor’s on-premises sunset is real: final feature releases for Kinetic, Prophet 21, and BisTrack in 2028, Active Support through 2029, then Sustaining Support (no new features, limited bug fixes). For the majority of the Insights audience — mid-market manufacturers with small IT teams in North America — the cloud migration path is viable and the AI capabilities are a compelling pull.

But in the analyst session, a direct question surfaced: highly regulated industries outside the U.S. are scared of and/or often reluctant to migrate the cloud, which is understandable. Is there a bridge? Epicor’s answer pivoted to the majority case. Kinetic runs on Microsoft Azure; data goes to the geographically nearest region; legal advice is recommended for data residency questions. That is not a data sovereignty guarantee. Customers in defense-adjacent manufacturing, financial services supply chains, or government contracting contexts face constraints that regional Azure deployment does not automatically resolve. Epicor has time — Active Support runs through 2029 — but that time is the planning window, and an opportunity to figure out the pricing questions, not a reason to defer starting the conversation.

Governance at Scale Remains Theoretical

The Agent Foundry’s code scanning, ERP security integration, and Lux design enforcement are the right governance mechanisms on paper. What isn’t yet clear is how they perform as agent complexity increases, as customers push the boundaries of what agents can access and act on, and as the marketplace grows beyond a curated early set. Epicor’s argument: that ERP safeguards provide the accountability layer that standalone AI tools lack, is structurally sound. Proving it at production scale is a different challenge. The appropriate posture for customers right now is to treat AI governance as a parallel workstream to agent deployment, not an afterthought.

What the Customer Stories Actually Show

The 2026 Customer Innovation Award went to Tilton Group, a family-owned manufacturer that migrated to Epicor Kinetic Cloud and achieved 99.5% service levels, near-perfect inventory accuracy, and what they described as precision seasonal planning. That’s a meaningful set of outcomes for a company in a sector where forecast accuracy directly drives margin.

The CEO Champion Award went to Gevint, a Canadian electrical and automation distributor founded in 1906 that migrated to Prophet 21 Cloud across 171 locations, integrated Epicor EDI and ECM for AP and sales automation, and is now expanding into new sites and acquisitions without increasing back-office headcount. That last point, scale without proportional cost growth, is the business case for cloud ERP that should really resonate with CFOs.

Cornell Pump’s two-week migration from NetSuite to Kinetic is the story that will travel farthest from Insights 2026, and appropriately so. The IT manager’s advice was notably candid: engage your Epicor account executive early, because the Ascend tooling may not be visible to customers who aren’t having the right conversation with their account team. That’s a sales execution observation worth filing.

The Path Ahead for Epicor: Opportunity, Execution Risk, and What to Watch

Epicor is making a coherent and ambitious bet. The architecture argument — industry-specific ontology as the durable AI moat, ERP as the accountable action layer, community data as the intelligence network — is more intellectually honest than most of what we hear at enterprise software conferences. I appreciate that. The product announcements made at Insights 2026 were substantive. The customer proof points are real and impressive. And the company is investing at a level ($200M+ R&D, majority directed to AI and migration tooling) that suggests commitment rather than positioning.

But the path ahead has genuine execution risks that must be considered.

  • Migration momentum vs. competitive defection. The on-premises sunset is the most powerful forcing function Epicor has to drive cloud adoption. It’s also a window for competitors. Every on-premises customer evaluating the migration decision is, by definition, also evaluating alternatives. Epicor’s ability to retain that base depends on whether the Ascend program delivers the migration experience it’s promising, not just in showcase cases like Cornell Pump, but across the heterogeneous, heavily customized deployments that characterize the long-tail installed base.
  • Pricing clarity before scale. Outcome-based pricing for AI agents is the right long-term model. Unresolved pricing at the point of customer contract negotiation is a near-term risk — both for customer trust and for Epicor’s ability to capture the value it’s creating.
  • Change management is the actual constraint. Multiple customer panelists at Insights said the same thing independently: the bottleneck on AI adoption is organizational, not technical. Epicor’s own research reportedly found 92% of manufacturers call smart manufacturing critical to long-term strategy while most say they’re not ready to deploy it at scale. Closing that gap is key and will require investments in adoption support that go beyond great demos.
  • Governance at production scale. The ERP-as-accountability-layer thesis is compelling. Proving it as agent complexity grows and the marketplace expands requires ongoing transparency about what the agents are doing, when they’re wrong, and how errors are caught and corrected. Epicor needs to publish that track record, not just the use case wins.
  • The World Model requires community trust. The data aggregation strategy only works if customers opt in, and they will only opt in if the value exchange is clear and the governance is credible. Epicor will need to be specific about what data is included, how anonymization works, and what each participant receives in return before the World Model becomes a real competitive asset rather than a slide.

The question is not whether AI is coming to ERP. That question is settled. The question is whether your ERP’s data structure, governance model, and domain expertise give the AI something real to work with, or whether you’re just bolting a general purpose chatbot onto a system of record that the AI doesn’t actually understand.

Epicor’s answer to that question is the most architecturally honest in the mid-market ERP space right now. Whether it translates from Nashville keynotes to production reality is what the next 18 months will reveal.

Analyst Take: What Epicor Customers and Prospects Should Do Now

What do I see ahead for Epicor customers and prospects? If you’re an Epicor on-prem customer: treat 2026 as the year to start the migration planning conversation in earnest. The cloud is where the AI capabilities live. Remember that the Active Support runway through 2029 is the planning window, not a reason to wait.

If you’re in contract negotiations, it’s imperative that you require specific pricing terms for agentic AI and Prism capabilities before signing. Outcome-based pricing is the stated direction; what’s defined as an outcome, what caps apply, and what happens at renewal are questions that need written answers now.

If you’re building a governance framework, start before you expand agent deployment, not after. Define the use case, set approval checkpoints, establish regression testing, and identify who owns accountability when an agent takes a wrong action.

If you’re evaluating Epicor as a new platform: the Cognitive ERP architecture and vertical ontology arguments are the most differentiated positioning the company has had in years. Test the freight variance, MRP explainability, and quoting intelligence agents in your own environment before the contract is signed. Conference demos are the floor, not the ceiling — what matters is whether the agents perform in your data context.

And lastly, if you’re watching from the competitive sidelines: Epicor is making a serious, coherent AI infrastructure play. The companies that should be most concerned are those competing in mid-market manufacturing and distribution without a comparable industry data model or migration acceleration story.

 

This article was originally published on LinkedIn.

 

Read more of my coverage:

Beyond the Front Door: Why Locking Access Isn’t Enough in the Age of AI and Agentic Risk

From Hype to Operational Reality: Mitel’s Vision for Enterprise Communications in the AI Era