Maybe you bought a subscription. Maybe your team went to a workshop. Someone tried a few prompts, got a good answer, and meant to come back to it.
Then the next rush job came in.
That doesn't mean AI can't help. It means it's worth looking at what happened between trying the tool and using it in the work.
Trying it and using it aren't the same thing
In a 2025 survey from the Manufacturing Leadership Council, a division of the National Association of Manufacturers, 66% of respondents said they had difficulty scaling AI pilots into full production. And 78% named legacy systems, poor connectivity, or integration as a challenge.
Those numbers describe the people who answered that survey, not every manufacturer. They also don't mean 66% of projects failed. They do show that getting past the trial stage is a real concern in manufacturing.
How AI connects to the everyday work matters too. McKinsey's March 2025 research found that out of 25 factors it tested, redesigning workflows had the strongest link to reported profit impact. But only 21% of respondents whose organizations used generative AI said they'd fundamentally redesigned at least some workflows.
That's a pattern in survey answers, not proof that changing a process guarantees a return. Still, it points to a better question than "Which AI tool should we buy?"
Where does the work get stuck, and what would need to change?
Why this is hard in a small shop
A small team has to find time to learn while the work keeps coming.
The OECD's 2025 research on small and medium-sized businesses puts it plainly. With fewer employees, small businesses have less flexibility to pull someone off revenue-generating work for training.
Among businesses in that survey that weren't using generative AI, 49.8% said employees didn't have the right skills. And 52.5% were concerned about what happens to information entered into the tools.
That research covered seven countries, not just U.S. manufacturers. But those are practical worries, and they deserve real answers. Who has time to learn this? Can we use it with customer information? Who checks the answer?
A good demo also doesn't tell you whether the setup fits the job. Sometimes the tool really is the wrong one. Messy source information, unreliable answers, and no connection to the systems you already use can all get in the way.
Kathie Mahoney, president of MassMEP, said it well in a June 2026 Manufacturing Dive interview:
“It’s more about the people and how do you integrate everything.”
To me, that means listening to the people doing the job before deciding what to build.
Start with one job worth fixing
NIST's guidance for smaller manufacturers recommends starting with a focused problem, getting the data ready, and building support among supervisors and the people who'll use it.
In a small shop, here's how I'd put that into practice.
Follow one recurring task. Watch how a request moves from the inbox to the person who has to act on it. Look for the reading, copying, searching, and chasing that happens over and over. Ask whether a checklist or a simple automation would fix it before you add AI.
Pick one result that matters. Choose a task that happens often enough to test. Decide what better would look like, like less time gathering quote information, without promising a number before you've measured it.
Use the tools your team already knows, when it makes sense. If your current software can handle the next step safely, put it there. Don't make keeping a second system updated part of someone's job without a good reason.
Let a trusted person try it first. Ask the estimator or office lead who knows all the exceptions to test it on real work. Give them room to say what's wrong. Their judgment should shape the setup.
Make ownership part of the handoff. Name who checks the output, who keeps the instructions current, and what happens when something breaks. Keep a way to do the task without AI.
In a 2026 Design News interview, Tim Gaus, Deloitte's smart manufacturing leader, put the ownership question this way:
“They don’t want to be reliant on a partner for their entire journey.”
A good handoff helps your team run the work on their own, including knowing when to ask for help.
Example: getting a quote request ready
This is a made-up example to show the idea. It isn't a BloomingRock client project or a measured result.
Picture a small fab shop. A request comes in with an email thread, a PDF spec, and two drawing revisions. The estimator has to pull the details together before pricing can start.
The first task could be an intake summary in the shop's existing job folder or spreadsheet. Using approved tools, AI could draft a list of the requested quantities, the documents supplied, and the open questions, with links back to the source material.
The estimator checks every entry against the originals. Anything missing or conflicting stays flagged. Pricing and technical calls stay with the people responsible for them.
Before trying it, the shop writes down how long prep takes today. During the trial, it tracks review time, corrections, and whether the team actually uses the summary.
If checking it takes longer than doing the task the old way, the setup needs work. If a checklist works better, use the checklist.
Make it useful in a normal week
The research says to look past the demo. Fit, reliable information, training, and ownership all need attention.
In your shop, the place to start can be one recurring problem and the person who knows it best.
If you'd like to look at one together, Book a Look. I'll start with how the work gets done today.
Sources
- Manufacturing Leadership Council / NAM, Shaping the AI-Powered Factory of the Future, May 2025, pp. 20-22 and methodology p. 27. Report.
- McKinsey, The state of AI: How organizations are rewiring to capture value, March 12, 2025. Fieldwork July 2024. Research.
- OECD, Generative AI and the SME Workforce, November 5, 2025. Preparedness chapter and methodology.
- Cole Rosengren, MassMEP president talks AI adoption, funding and the evolving industry, Manufacturing Dive, June 16, 2026. Interview.
- NIST MEP National Network, Artificial Intelligence: Key Consideration and Effective Implementation Strategies, July 2025. Guide.
- Rob Spiegel, Deloitte Expert Says AI Enhances Manufacturing Instead of Replacing Human Workers, Design News, March 5, 2026. Interview.