Key Takeaways:
- Invest in data before you invest in AI in manufacturing. Before evaluating any AI tool, audit what data you have, where it lives, and whether it can be trusted across sites. The question is not which model to use — it is whether the data that model would run on is clean, connected, and complete enough to act on.
- The sequence matters more than the model. Define and connect your underlying processes before layering intelligence on top of them. Deploying AI on a broken or disconnected workflow does not fix the process — it compounds it. Get the basics right first: internal handoffs, data connections, cross-site visibility.
- Define your digital workers as a workforce category before you deploy them as tools. Scope AI agents for outcomes — what they are responsible for delivering — not for tasks. Then redesign human roles around what those agents cannot do. Adding AI without restructuring the work around it means leaving most of the return on the table.
- Design the human decision point deliberately, not by default. For every AI output, decide in advance who reviews it, what they are deciding, and under what conditions they override. Organizations that leave this ambiguous end up with outputs no one acts on, or that everyone acts on without scrutiny. Neither is the goal.
- Process literacy determines how fast AI adoption moves — not model capability. Before deploying any AI tool to a team, confirm that team can describe their own process clearly enough to recognize when the output is wrong. The teams that catch errors are the ones who understand the work. Process documentation is not overhead — it is the adoption accelerator.
Highlights from the 2026 North American Manufacturing Excellence Summit (NAMES).
The Story in Brief
“One of the sharpest lines I heard at NAMES 2026 was, ‘He was just digitalizing bad practices, and now you’ve got a digital mess that you paid a lot of money for.’ It got a laugh because everyone in the room had either seen it happen or lived it. The AI isn’t the problem. The sequence is. Every organization at this summit that had moved AI from pilot to production did the same boring work first: fixed their data, mapped how their processes actually run, and defined the human roles that make the AI trustworthy. The checklist we built is a provocation — a way to find out how far you actually are from ready, before you find out the expensive way.” — Andi Del Collins, Executive Director, Manufacturing Board
Picking Up the Story on AI in Manufacturing
Most AI Pilots Don’t Fail Because the Model Was Wrong. They Fail Before the Model Ever Gets a Real Test.
Manufacturing organizations are investing in AI at a pace that outstrips their readiness to use it well. The gap shows up the same way every time: a pilot that produces promising results in a controlled environment stalls when the team tries to replicate it — because the data isn’t consistent across sites, the process it was trained on doesn’t match how work actually happens everywhere else, or the people receiving the AI’s outputs don’t know what to do when it flags something they’ve never seen before.
Every one of those is a solvable problem. None of them are solved by a better model. The sessions at NAMES 2026 that showed AI operating at production scale had worked through all three before they launched.
“In this world of AI, my message today is: Go back to the basics before actually using smarter systems. Make sure the basic connections are there.”
Focus of the Discussion
Four Reasons AI Pilots Stall and What the Scaling Leaders Did Differently
Throughout our discussions at NAMES, four common pitfalls emerged:
- Data no one fully trusts. Manufacturing data is often cleaner in theory than in practice — Sparetech found that 22 percent of spare parts across the facilities they work with haven’t moved in five years, invisible to other plants in the same network. Mattel’s supply chain leader gave the clearest framing of the Summit: Before reaching for smarter systems, make sure the basic internal connections are in place and every function knows what it owes the next one.
- Automating a process that was already broken. Toyota’s North America manufacturing lead reframed this as a sequencing question: How do you decide where a problem is a technology problem versus a people and process problem? SAP made the architectural version of the same point — AI multiplies value only when planning, execution, logistics, and workforce data are integrated and the underlying process is sound. Technology layered on top of a broken workflow doesn’t fix it; it accelerates it.
- No one owns the response when the system flags something. Ford’s Milo texts the zone team leader when it flags an anomaly — the team leader decides whether to pull the unit. That handoff was designed before launch, not after. Without it, the most common outcome is a system that generates alerts nobody acts on, because nobody owns the response.
- Treating workforce readiness as an outcome of AI deployment, not a precondition for it. Nestle Purina spent 24 months rebuilding training infrastructure and standardizing 150 workforce personas before deploying new technology — not as an AI initiative, but as the precondition for one. Their framing: the question isn’t whether new solutions will be adopted — it’s how fast, based on the investment already made. The organizations that haven’t done this work are already behind.
About the Manufacturing Board and Assemble
The Manufacturing Board is an Assemble community that helps senior manufacturing, operations, supply chain, engineering, and plant leaders at the world’s largest companies make better, faster decisions through trusted peer insights.
NAMES — the North American Manufacturing Excellence Summit — is the Manufacturing Board’s annual in-person event where members and peers work through the operational challenges shaping the industry.
We selectively share highlights from confidential member conversations to help leaders understand how peers are navigating complex operational challenges and fast-moving transformation efforts.
If you’d like to learn more about the conversations happening inside our peer network, or how members connect with trusted peers to solve critical challenges, please contact us.
About the Author
Andi Del Collins, Executive Director
Andi Del Collins is an Executive Director at Assemble, where she has built three peer networks from the ground up: the Employee Experience Board, the Learning & Development Board, and the Manufacturing Board. Her work sits at the intersection of community design and executive peer learning — creating the conditions for senior leaders to think through hard problems together. She writes and synthesizes on behalf of those communities to help members see around corners their organizations haven’t reached yet.
