Nearly every manufacturer is looking at AI. Almost none of them have the data infrastructure to run it. That's not an AI problem — it's an operations problem.
New research from Redwood Software puts a precise number on something most of us already suspected: 98% of manufacturers are actively exploring AI. Only 20% describe themselves as fully prepared to deploy it.
That's not a technology adoption gap. That's an operations gap dressed up as a technology problem.
And if you're a mid-market manufacturer — $100M to $1B in revenue, running SAP or an aging ERP, with a compliance function that still lives mostly in spreadsheets — this gap is aimed directly at you.
Here's what the 80% who aren't ready tend to have in common. Their data lives in silos. Their ERP was implemented a decade ago and is held together with custom integrations nobody fully understands. Their automation efforts are fragmented — an RPA bot here, a Power Automate flow there — with no coherent orchestration layer connecting any of it.
They've run AI pilots. Some of them even worked, technically. But when it came time to scale — to move from proof-of-concept to production workflow — the seams showed. Quality data wasn't where the model expected it. Change control processes slowed down deployment. The compliance team asked questions nobody had answers to.
The pilot got shelved. Or worse, it got deployed anyway and started producing outputs nobody trusted.
This is the pattern. It's not unique to manufacturing, but it's particularly acute there because the stakes are higher. A hallucinating AI in a content marketing workflow is embarrassing. A hallucinating AI touching production scheduling, supplier quality data, or safety documentation is a liability.
Most manufacturers are having the wrong conversation. They're debating which AI platform to buy, which vendor has the best demo, which large language model performs best on their use case. These are secondary questions.
The primary question is: do you have clean, structured, accessible data that an AI workflow can actually operate on?
For most mid-market manufacturers, the honest answer is no — not yet.
This isn't a permanent condition. It's a sequencing problem. The companies that are pulling ahead aren't the ones who found the best AI tool. They're the ones who invested first in the boring infrastructure: data governance, ERP integration hygiene, process documentation, orchestration layers that connect fragmented systems into something coherent.
That work isn't glamorous. It doesn't generate press releases. But it's what separates the 20% who are ready from the 80% who are still exploring.
If you're trying to move from the 80% to the 20%, here's a useful framework. Think in three layers:
Layer 1: Data Foundation
Can you reliably answer these questions: Where does your production data live? Who owns it? How clean is it? How current is it? If your answer to any of these is "it depends" or "I'm not sure," you're not ready to deploy AI in that domain. Fix the data before you build the workflow.
Layer 2: Process Clarity
AI workflows need documented processes to run against. Not aspirational process maps — actual current-state documentation of what happens, in what order, with what inputs and outputs. Most manufacturers have this for their shop floor. Very few have it for back-office operations like supplier onboarding, quality exception handling, or regulatory reporting. Those are exactly the workflows where AI can deliver fast ROI — but only if the process is legible first.
Layer 3: Integration Architecture
This is where most mid-market manufacturers get stuck. AI doesn't operate in isolation. It has to connect to your ERP, your MES, your quality management system, your document repositories. If those systems aren't integrated — or if the integrations are brittle and undocumented — you can't build reliable AI workflows on top of them. You need an orchestration layer that makes these connections stable before you start automating decisions.
Stop evaluating AI vendors until you've done an honest internal audit on these three layers. Not a polished assessment for the board — a working document that your ops team, IT, and compliance leads can all agree is accurate.
For each candidate workflow, ask: Is the data clean and accessible? Is the process documented and stable? Are the system integrations reliable?
If the answer to all three is yes, you're ready to build. If any answer is no, that's your next project — not the AI deployment.
The 20% who are ready got there by doing this work before they needed it. The rest are going to spend 2026 learning the same lesson the hard way.
You don't have to be one of them.
Dealing with a similar challenge?
We work with mid-market companies in regulated industries to build AI workflows that actually hold up.
Let's TalkSean Cummings
Founder of Laminar Flow Analytics. Specializes in AI workflow automation for regulated industries — medical device, financial services, and complex logistics operations.