You’re Not Behind on AI. You’re Behind on the Plumbing.

Why the manufacturers winning with AI started with the boring stuff first.

If you run a mid-sized plant, you know the feeling. Another headline, another vendor at the trade show, another competitor posting about “AI on the shop floor.” Everyone seems to be sprinting somewhere, and you’re standing at the line wondering if you missed the gun.

Here’s what nobody selling you software will say out loud: that hesitation isn’t a weakness. It’s good instinct.

Because the manufacturers actually getting results from AI didn’t start with AI. They started with the unglamorous work underneath it — the part the headlines skip.

Think about what AI actually needs to be useful on your floor. It needs clean, trustworthy data. It needs to see across quoting, scheduling, inventory, and production at the same time. It needs to know which job is late and why. If your inventory count lives in the ERP, the real count lives in a spreadsheet, and the true status lives in your senior operator’s head, no model on earth can give you a straight answer. It’ll just automate the confusion faster.

So before you buy anything with “AI” stamped on the box, do the work that makes AI pay off later:

Clean up your history. Your past quotes, job costs, downtime logs, and scrap rates are the most valuable asset you’re not using. If they’re scattered, inconsistent, or trapped in a retiring estimator’s memory, fix that first. Bad data in, expensive nonsense out.

Close the gaps between your systems. Most plants don’t have a software shortage — they have a translation problem. The ERP and the floor disagree. Finance and production work from different versions of the truth. Every patch is another spreadsheet. Connecting what you already own beats buying something new almost every time.

Find the one workflow that bleeds. You don’t need an enterprise transformation. You need to name the single place where time, margin, or sanity quietly leaks — the re-quoting, the manual rescheduling, the three-week month-end — and fix that. One real win earns more trust than ten stalled pilots.

Do this, and something useful happens. The picture of your operation gets clear. The handoffs get visible. The bottlenecks stop hiding. And only then does it become obvious where automation — and yes, eventually AI — actually belongs.

That’s the part the hype gets backwards. AI isn’t the starting line. It’s the reward for a business that already runs on connected, trustworthy systems. It’s a consequence of maturity, not a shortcut to it. Adopt it before the foundation is set, and you’re pouring intelligence into a leaky bucket — paying premium prices to scale a mess.

So if you’re feeling the FOMO, let it point you somewhere productive. Not toward the shiniest demo, but toward the quiet, structural work that makes everything after it work.

You’re not late. You’re being careful. In manufacturing, careful is how the people who win usually start.

Rafael Miranda is Managing Partner at Ultrahaus, a Canadian AI and automation studio that has paired strategic thinking with technical execution since 2003. He works at the intersection of platform integration and pragmatic AI — which is why he tends to argue that the foundation matters more than the next shiny tool.