TL;DR

A few days ago I was flipping through the October issue of Manufacturing Engineering & Technology (yes, I still read print magazines 😅) and stopped on a short piece by Tim Beatty and John Liddy from Bullen Ultrasonics. The title asked the question I hear in almost every first meeting: why do manufacturing AI projects fail?

I read it twice. Then I took pictures of it so I could share it with you.

Working in operations, on shop floors and in planning rooms, I could feel the pain in this article. A lot of us had to learn this the hard way.

It starts with data you can trust

Their opening is simple: “Successful adoption, however, requires more than buying software and connecting machines.”

Then they get honest about what the real world looks like. Equipment bought across several decades. Machines that collect information inconsistently. Systems that don’t talk to each other. And process knowledge that lives “scattered across spreadsheets, databases and employee experience.”

Sound familiar? Literally every company I’ve walked into has some version of this.

Bullen didn’t wait for the AI wave to fix it. They began extracting machine data in 2019, upgraded hardware, and improved how information is collected and placed in context. That work has produced more than 2 billion data points tied to specific machines, processes and conditions.

And here’s the line I love most: “Volume alone does not create value.”

Two billion data points. And they still say that. That’s humility, and it’s the right kind.

Where the FDO comes in: this is the first job. Make the data trustworthy and keep its context. Which machine, which material, which shift, who changed the setting. A number without its story is just a number. The FDO is the person who sits with the team and asks, “where does this actually come from?” until everyone gets the same answer.

Name the problem before you name the tool

The second point is the one I’d tape to every boardroom wall: “AI should solve a defined manufacturing problem, not simply prove that the technology can be used.”

Their examples are concrete: increasing yield, reducing cycle times, improving equipment utilization, finding the variables behind inconsistent output. Not “let’s do something with AI.” A real problem, with a number attached.

Who owns that problem? What does “better” look like? How will you know?

If you can’t answer those three, you’re not ready for the tool yet. And that’s fine. That’s information.

Where the FDO comes in: someone has to turn “do something with AI” into one problem, with one owner and one number. That’s not a tech task, it’s an operations task. The FDO walks the process, finds where the pain really is, and writes it down so plainly that nobody can misread it. One problem. One owner. One number.

Let readiness set the pace

This part is so practical. Bullen’s departments are at different stages. Some have years of organized data. Others still need machine upgrades. So instead of forcing everyone onto the same schedule, “we allow readiness to determine the pace.”

They start where usable data, operational need and employee expertise meet, test it under real production conditions, and only then expand. Their words: “Scaling should follow evidence, not enthusiasm.”

I’m telling you, that sentence alone would save a lot of budgets.

Where the FDO comes in: evidence doesn’t collect itself. The FDO tests tools on the real work (not the demo data!), compares the results, and makes the call on which department is ready now and which one needs the foundation first. Honest calls, based on what the test showed. Not on who is the most excited.

The people who know the machines

Bullen learned this before the AI boom. When they brought in their first robot, they held open sessions so employees could see how it would work, ask questions, and understand why.

They say AI needs the same approach. Operators, engineers and technicians “understand materials, machines and process conditions that may never be fully represented in a dataset.”

AI can point to a correlation. People decide whether it’s physically realistic, commercially useful and safe to test. That’s the 80/20 I always talk about. AI does the heavy lifting on the patterns. You keep the final say.

Where the FDO comes in: the FDO sits between the floor and the tech. They listen to the operator who knows which machine acts up and when, and make sure that knowledge makes it into the system. And when the model suggests something, they bring it back to the people who know, so humans keep the final say. Translator in both directions.

Why this role matters

I want to be humble here. Bullen did the work, and I’m a reader cheering from the sidelines. But reading their piece, I kept seeing the same thing: every step needs an owner.

Companies are very good at buying tools. What’s usually missing is the person who owns readiness end to end. Not just the data. Not just the pilot. The whole chain, from the machine to the margin.

That’s the Forward Deployed Operator. Someone who stays current on the technology, builds on a solid foundation, and always knows the answer to one question: how does this business make money, and where can AI help it make that money faster?

Without that person, you get a pilot. With that person, you get a capability.

A quick readiness check for your Monday

  1. Can you name one problem, with one number, that AI should move?
  2. Is the data for that problem collected the same way every time, with context?
  3. Who on the floor (or in the office) knows that process best, and are they in the room?
  4. Have you tested more than one tool on it, on your real work?
  5. Do you know how solving it shows up in revenue or cost?

If you said no to some of these, welcome to the club. Most of us are somewhere in the middle. Bullen took years. You don’t have to do it all this quarter. You just have to start with the next honest step.

Resources

AI readiness isn’t a purchase you make once. It’s a habit you build, one honest department at a time.

Know an ops leader who’s been asked for “an AI plan” by next quarter? Forward this to them. It might take some pressure off.

Listen to the Forward Deployed Operator podcast | Join the FDO School waitlist