Key Highlights

  • Epicor’s CEO opened Insights 2026 with unusual candor — apologizing directly for billing issues and admitting top AI models hallucinate up to 90% of the time, setting a trust-first tone for the event.
  • Epicor positioned its 50+ years of manufacturing data as an “industry ontology,” a semantic model the company argues is a genuine AI moat that horizontal AI vendors cannot replicate.
  • Six “cognitive ERP” capabilities were unveiled, including Prism vertical agents, the Lux design/governance layer, Agent Foundry, and a headless ERP architecture built on MCP.
  • A freight spend intelligence agent delivered what was arguably one of the event’s most credible outcomes: one customer’s freight variance dropped from 19% to 1.5%.
  • Open questions remain on agentic AI pricing, on-prem-to-cloud sovereignty options for international customers, and governance at scale as agent adoption deepens.

Epicor’s Insights 2026 conference in Nashville produced some of the most consequential announcements the company has made in years, and the framing set in the opening keynote shaped everything that followed. CEO Steve Murphy didn’t lead with a vision statement or an AI roadmap. He led with an apology, directly addressing long-running billing problems and committing personal involvement until they’re resolved.

That choice mattered. As my fellow analyst Robert Kramer of KramerERP and I discussed on the Experience Matters podcast, ERP buyers aren’t purchasing features, they’re purchasing trust in the system that will run every facet of their business. Flipping the usual vendor script from “vision first, accountability second” to “accountability first” established the credibility that every subsequent claim in the keynote could borrow against.

Watch the full podcast conversation here:

That candor extended into Murphy’s AI primer, which pushed back on several popular AI narratives. He acknowledged that AI is simultaneously powerful and frightening, that AI-attributed layoffs mostly reflect prior over-hiring, that AI alone won’t create new jobs, and that creativity remains a human advantage. Most striking: he told a room of customers that the best AI models hallucinate more than 10% of the time, and some closer to 90%. That’s not language typically heard from a vendor CEO on a keynote stage, and it set up the core argument of the event: ERP as the substrate where machines handle repetitive work and humans retain judgment. To be fair, this is certainly a strategy Robert and I have heard before, so it’s not breaking new ground in the industry, but for Epicor, it’s the right message.

The Industry Ontology Argument

The most substantive portion of the keynote came from Viabhav Vohra, Epicor’s Va President and Chief Product and Technology Officer, who argued that Epicor has built something horizontal AI vendors can’t replicate: a data ontology — a semantic model of how manufacturing objects (parts, jobs, routings, vendors, warehouses) relate in real production contexts — tied to more than 50 years of manufacturing and distribution data.

As Robert put it, AI without that ontology is essentially guessing. With it, AI outputs become accurate, actionable, and auditable, with every answer traceable to a specific job, part, or rule already in the system. What is significant here is that no horizontal AI vendor selling into manufacturing has built this from the transactional layer up, which is what makes it a genuine competitive moat rather than a feature checkbox. The caveat for Epicor that Robert and I agree on is that this needs to be proven in complex, discrete-manufacturing environments like automotive over a sustained period, not just demonstrated on stage.

Six Cognitive ERP Capabilities — What’s Real, What’s Roadmap

Epicor claims to be the first ERP company to ship six cognitive capabilities simultaneously: Prism vertical agents, the Lux design system, Agent Foundry, headless ERP, an expanded Ascend migration program, and an industry world model.

Some of these were live demos against real customer environments; others read as directional. The quoting intelligence agent showed a 92% same-as-expert match rate and a 41-minute quote cycle — concrete and credible. Agent Foundry demonstrated reducing revenue risk from $1.2 million to $125,000 in a sample scenario, though the capability isn’t shipping until later this year. Lux functions less as a design system and more as a governance layer — the contract that keeps agent-built screens feeling like trusted Epicor, which Robert noted could become one of the company’s strongest differentiators.

The headless ERP architecture stood out as a quiet but significant shift: it decouples business logic from the UI via MCP (Model Context Protocol), and Epicor demonstrated Claude connected to Kinetic and ChatGPT connected to Prophet 21 through the same governance layer. I’m a big fan of the headless ERP move. By openly embracing MCP as an open AI-to-system protocol rather than building a proprietary integration wall, Epicor is positioning itself as infrastructure rather than interface — though it raises governance questions about what happens when an external agent gets something wrong inside the ERP.

The Freight Demo: An Outcome You Could Drop on a Financial Statement

Among all the demos, we found the freight spend intelligence agentwas the most business-outcome-anchored. One customer had absorbed $700,000 in freight variance over two years before the agent surfaced the problem; after deploying tighter controls, variance dropped from 19% to 1.5%. The agent reconciles ERP shipping data against carrier invoices, flags the root cause of mismatches, and automates controls going forward.

What made this demo land wasn’t that it was an agent, it was that it produced a number finance could act on immediately. That’s the bar every other capability at the event should be measured against.

The Industry World Model: Epicor’s Boldest, Longest-Horizon Bet

Perhaps the most ambitious claim of the event was the Industry World Model, an aggregated, anonymized, opt-in intelligence layer drawn from tens of thousands of manufacturers and distributors. Epicor’s product leadership claimed this layer could predict BLS CPI movements more accurately and faster than the public index.

That’s a striking claim, and decades of supply chain transaction data is genuinely an asset that pure-AI players can’t replicate — arguably a stronger moat than fine-tuning a general LLM on industry PDFs. But it needs receipts. Our recommendation to Epicor: publish the methodology, make the opt-in and anonymization governance public and credible, and let an independent party validate the CPI-prediction claim. That’s when the moat becomes real rather than aspirational.

Proof Points and the Gaps That Remain

Customer proof points reinforced where Epicor is winning: Tilton Group’s Innovation Award recognized 99.5% service levels and near-perfect inventory accuracy; a Canadian distributor’s CEO Champion Award recognized scaling to 171 locations without back-office headcount growth; and Cornell Pump’s two-week migration from NetSuite to Kinetic quickly became the most-discussed story of the event. These wins cluster in mid-market manufacturing and distribution, where Epicor’s vertical depth plus Ascend migration tooling create real differentiation.

Three gaps deserve scrutiny. First, agentic AI pricing remains unresolved — no firm structure, caps, or overage policy, which Epicor candidly acknowledged in the analyst Q&A. Outcome-based pricing is attractive in principle, but uncertainty around it is a real barrier to a buying decision. Second, the on-prem-to-cloud question for international and highly regulated customers got a pivot toward Azure regional availability rather than a true data sovereignty answer — a gap that matters most for defense-adjacent and government-contractor accounts. Third, governance at scale: Agent Foundry’s code scanning, Lux’s design enforcement, and ERP security integrations are the right mechanisms on paper, but they haven’t been tested against complex, expanding agent marketplaces. That needs to be next on the agenda for the team at Epicor.

What Has to Be True in 18 Months

For Nashville to represent a movement rather than a moment, several things need to happen. Ascend migration tooling has to perform across Epicor’s heterogeneous, heavily customized long-tail install base — not just showcase cases like Cornell Pump. The world model’s governance and opt-in framework needs to become public and independently verifiable. Agentic pricing needs resolution before customers make alternative-vendor decisions based on uncertainty alone. And community intelligence claims need to hold up under outside scrutiny.

The bottom line for manufacturing and distribution leaders: stop counting agents and start counting outcomes. The freight variance story — going from double-digit losses to 1.5% — is the kind of result that belongs on a P&L, not a roadmap slide. Epicor’s architectural argument, built on five decades of industry-specific data and an MCP-native approach to AI integration, is arguably its strongest positioning in years. Whether it becomes the new standard for ERP as a system of action — or simply one of the more credible attempts at it — will be determined over the next 12 to 18 months.

Read more of my coverage:

Epicor Insights 2026: ERP is No Longer a System of Record: It’s a System of Action

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

 

This article was originally published on LinkedIn.