When the chief procurement officers of Toray, Mosaic, American Airlines, Wesco and ZF took the stage at ISM World 2026 at the Gaylord Rockies Resort in Aurora, Colorado from April 26 to 28, they delivered something closer to a public correction than a conference panel. The message to every industrial AI vendor in the room — and to the operating committees back home weighing 2026 capex against unproven software bets — was direct: stop selling transformation. Start naming a workflow.
The framing has hardened fast. As recently as a year ago, procurement keynotes were still organized around platform-scale ambition: agentic suites, end-to-end orchestration, source-to-pay reinvention. In Aurora, the executives running those functions in real companies said something different on the record. According to Supply Chain Dive's coverage of the panel, the discipline now starts with the business case and stops at the perimeter of a single problem.
What was actually said on stage
John Eustis, senior vice president of U.S. group procurement at Toray Industries (America), set the budget reality plainly: "we don't have a blank check." His starting question for any AI proposal, he told attendees, is "what's the business problem you're trying to solve?" — not which platform to standardize on.
Chris Martus, vice president of procurement at The Mosaic Company, gave the test its sharpest formulation: "You have to have a clear business case for what you want to achieve, and clear objectives, and then you can identify whether the juice is worth the squeeze."
Dan Bartel, chief procurement officer of American Airlines, described what disciplined adoption actually looks like inside an enterprise of that size. The airline uses Microsoft 365 Copilot Studio to let individual procurement workers build their own automations — but it has not licensed the tooling for the full team. "We're working on proving out the business case to be able to do that right," Bartel said. The default posture, in other words, is a paid pilot for the practitioners closest to the work, not an enterprise rollout in search of one.
The panel was moderated by Charlotte de Brabandt, deputy head of IT governance and compliance at ZF Group, and also featured Brandon Phillips, vice president of global supply chain at Wesco. The composition matters: this was not a vendor showcase, and the speakers were not selling. They were describing a buying posture.
The short list that's actually getting funded
What the panel made visible is the narrow band of procurement AI that survived 2026 budget review. The use cases named on stage are bounded, measurable, and tied to a workflow an operator can describe in a sentence:
-
Automated supplier bid review. Toray cited reviewing pallet design specifications — a discrete document-comparison task with a clear output and a clear baseline cost.
-
Workflow automation via Microsoft 365 Copilot Studio. Adopted at the individual builder level inside American Airlines procurement rather than as an enterprise license.
-
Embedded AI agents inside the existing procurement stack. The Levelpath platform was named as an example of agents living inside tools procurement teams already use, rather than a parallel system.
-
Contract triage and routine document handling — bounded by document type and by what the team would otherwise pay an outside reviewer to do.
What was conspicuously absent from the funded list is just as telling: full-suite agentic procurement, enterprise-wide license rollouts, and any vendor pitch organized around the word "transformation."
The data behind the discipline
The restraint on display in Aurora is not vibes. It tracks two pieces of Gartner research that bracketed the conference.
On April 7, 2026, Gartner reported that only 28% of AI projects in infrastructure and operations deliver the ROI they were sold on. In a survey of 782 I&O managers, 57% said they had experienced at least one outright AI failure, and roughly one in five projects fail entirely. Independent trade-press coverage confirmed the figures. Those numbers are now the benchmark CFOs are pricing AI pitches against in Q2 2026, and they are doing it openly.
Then on May 19, 2026 — the day before this article was filed — Gartner published the procurement-specific cut: just 36% of chief procurement officers describe themselves as "very confident" in their ability to redesign roles and processes around AI, based on a January–February 2026 survey of 101 CPOs. Gartner's read of its own data is blunt: productivity gains from generative AI in procurement remain confined to individual workers and have not yet translated into organizational gains. That is exactly the picture Bartel described for American Airlines — Copilot Studio at the individual builder level, no enterprise license, business case still pending.
The counter-headlines — and why they actually reinforce the thesis
The ISM World framing has to be reconciled with two large industrial AI announcements that hit the wires in the same window.
On May 18, 2026, Stellantis and Accenture announced a strategic partnership to deploy AI-enabled digital twins built on NVIDIA Omniverse across global manufacturing operations, with the first work running through North American plant pilots in 2026. On March 31, 2026, Hitachi Digital Services launched a Manufacturing Operations Management platform as an expansion of its HMAX Industry portfolio, integrating OT and IT for industrial operators. Six weeks later, on May 15, 2026, Hitachi was named a Leader in the 2026 ISG Provider Lens for Intelligent Robotics and Physical AI Services for the third consecutive time.
Both announcements are real, both are platform-scale, and both look — at first glance — like the opposite of the ISM World message. They aren't. Read the language the principals used. Stellantis and Accenture explicitly start with "pilots in North America." Hitachi's platform is being sold into individual manufacturing operations, not as a one-shot enterprise migration. Even the biggest 2026 industrial AI announcements are now structured as scoped pilots in selected plants. The vendor side has not retreated from ambition. It has retreated from blank-check sequencing.
The operator's playbook for reading a vendor pitch
What ISM World 2026 effectively gave procurement and operations leaders is a screening template. Run any inbound AI proposal through it before it reaches the capital committee.
Questions to ask the vendor:
-
Name the single business problem this solves. If the answer requires more than one sentence, stop.
-
What is the boundary of the pilot — which sites, which document types, which workflows are explicitly out of scope?
-
What are the exit criteria? What measurable KPI tells us to expand, contract, or kill the project?
-
Who absorbs the cost if it fails? Vendor risk-sharing or buyer risk-only?
Red flags on the pitch deck:
-
Platform-only pricing with no per-workflow unit.
-
No named KPI tied to a P&L line.
-
"Transformation," "reimagine," or "end-to-end" framing without a single specific workflow named in the proposal.
-
Enterprise-license-first sequencing — the opposite of how Bartel described American Airlines' Copilot Studio adoption.
The quality and compliance angle that gets undersold
One reason the narrow-scope posture is gaining traction beyond pure cost discipline is governance. A bounded use case — bid review on a defined document type, contract triage on a defined contract family — produces an auditable decision trail. The compliance team can name what model touched what document, what the human-in-the-loop checkpoint was, and what the override rate looks like.
Platform deployments produce the opposite problem, and it is the governance gap Gartner flagged in its April I&O research: when an AI system is everywhere, no one owns the accountability for any single decision it influenced. For regulated procurement environments — aerospace, pharma supply chain, defense — that is a non-starter on its own merits, before the ROI question is even reached. That a panel at ISM World was moderated by ZF's deputy head of IT governance and compliance is not a coincidence.
The bottom line for 2026 budgets
The vendor side of industrial AI has not retreated. The buyer side has. Procurement organizations entered 2026 with a year of generative-AI experience behind them and Gartner-grade evidence that the broad bets are not paying off. The result is a market where the same vendors are still pitching the same ambition, but the contracts being signed are smaller, narrower, and tied to one named workflow at a time.
For operators reading vendor pitches between now and the next budget cycle, the working assumption should be that ISM World 2026 is the new baseline, not a temporary cautious moment. Start small, prove the business case, expand on evidence — or, as the panelists kept reframing it, don't start at all. That is the discipline the buyer side has chosen to enforce, and the 2026 budget is the instrument.
Related reading
Sources
-
Procurement leaders urge incremental AI adoption to avoid heavy spend — Supply Chain Dive
-
ISM World Annual Conference — Institute for Supply Management
-
Gartner: Only 28% of AI Projects in I&O Deliver Meaningful ROI (April 7, 2026)
-
Gartner: Just 36% of CPOs Very Confident in Redesigning Function for AI (May 19, 2026)
-
Only 28% of AI infrastructure projects fully pay off — The Register