Skip to Main Content

Make your workflow work for you: what industrial leaders said about AI this fall

If you work at an industrial company, you've probably seen it. You buy a big software platform, roll it out, train everyone, and a year later people are still tracking half their work in Excel or on a clipboard.

I heard some version of that story over and over this fall, first at MxD's 2026 Summit in Chicago and then again at ACG Chicago's Manufacturing Symposium. Plenty of plants have an ERP, an MES, and dashboards around the facility, and supervisors still have to piece together production data in spreadsheets.

This isn’t just happening in manufacturing. With diesel prices where they are, more freight is moving to rail, and the rail and supply chain companies we talk to have the same problem. There's a system of record but it doesn’t include status updates (which live in email) or handoffs between companies (which happen over the phone). Workers know what's going on but leadership has no visibility.

The good news is that this problem is easier to fix than ever. Whether you want to reconfigure an off-the-shelf application or build from scratch, the tools available today make customization largely accessible. In this piece, I’ll talk about how to get started to build or streamline the workflows most likely to move your organization forward.

The hard part now is deciding what to build

The tagline at MxD was "Manufacturing at Mission Speed," and speed came up in nearly every session: speed to develop, speed to deploy, speed to ROI.

That’s exactly what I’d expect given the AI tools currently available. It’s now possible to vibe-code a working prototype in days, get user feedback, and start incorporating that feedback within the week. When writing code is no longer the bottleneck, it’s much easier for companies to build software that fits the way their people actually work instead of trying to fit workstreams into existing software.

But speed only helps if it's pointed at the right problem. Aim it wrong, and you just build the wrong thing faster. Or, as one MxD panelist put it, you go around with an AI hammer looking for nails.

Manufacturing leaders are keenly aware of this risk. A recent ICAMS study found that, while 48% of manufacturers consider AI a top-three technology, only 11% report actually using it.

To my mind, the picture this paints is that we’re in a moment of opportunity. There's a big gap between interest in AI and AI use. Streamlining custom software development is a fantastic use case, once you’re confident you’re building the right thing.

The companies that pull ahead in the next few years will be the ones that figure out how to pick the right problem and then build something their people actually use.

Start with the people doing the work

So what does that look like in practice? The best example I heard was from Alison Magyari, VP of IT at Eaton. One of their plants was growing from about $300 million toward $1.5 billion, largely on data center demand. Their biggest bottleneck was engineering, so they asked for 20 more engineers.

At first, the request seemed reasonable. But when Alison and her team dug in, they found engineers redrawing parts from scratch when similar drawings already existed. Instead of hiring (which would have locked in the inefficiency), they built an AI tool that finds past designs so engineers can start closer to the finish line.

Alison also shared a great story about adoption. Early on, about 90% of what her team built was designed at corporate and handed to the plants. Adoption was terrible. Plant workers kept using their spreadsheets.

But when the team started spending 65-70% of their time on the shop floor, adoption picked up. Alison described it like this: "AI rarely fails at the model. I often see it failing at the seams."

In other words, AI was failing at the handoffs: a bill of materials had to be re-entered because two systems didn’t match, a quote had to be typed in by hand, a schedule printed Friday was out of date by Monday. Testing prototypes with end users early and often, then iterating on their feedback, helps avoid these failures and leads to software that actually works for the people who use it.

A week later at ACG, a company president told a story from earlier in his career, when he ran distribution for a Coca-Cola bottler in Louisiana. Once a month, he’d ride along on a delivery truck to see how the warehouse loaded it, how the route ran, and what happened at each store.

"You do have to get out there with your people," he said, "so you then know what to go and transform and fix and improve."

He also offered a back-office version of the “ride-along”: As a new CFO at another company, he was trying to figure out why a price increase never hit the bottom line. He sat down with the cash applications team and found customers were taking extra discounts when they paid, so the increase never actually hit their books. He only found it by sitting with the people doing the work.

In each of these cases, someone found the costliest problem to solve by watching how the work got done. (For more on picking which problem to go after first, see our piece on how to pinpoint the AI workflow opportunity with the biggest ROI.)

Get the data right enough, then start

Once you know the best problem to solve, data comes next. A manufacturing COO who spoke at MxD was blunt about it: "If your BOM is wrong... your routing is wrong... your process is unstable and [your] data is bad, AI doesn't magically fix the factory. You may simply become… very sophisticated… at making bad decisions faster."

This echoed what 3M’s SVP of manufacturing operations said at the American Manufacturing Summit earlier this year: when they added AI without first cleaning up their data foundation, they “amplified a bad system.”

In other words, even if you’re solving the right problem, you can’t expect to solve it with bad underlying data. But that doesn’t mean waiting for perfect data to start.

At MxD, the AI panel cautioned against the practice of spending years building a pristine data foundation before doing anything useful. Their advice: choose a desired outcome, map the process of getting there, figure out what data the outcome needs, and go. If you’re already making decisions with messy data, it’s good enough to start.

I’d add one thing: do all that work before you pick the software you’ll use (whether off-the-shelf or custom), so the tool fits your process instead of trying to bend your process to fit the tool.

This is exactly what we helped one of our clients, a railcar lessor, do. They were running a complex, multi-party process almost entirely over email. If you weren’t cc’d, you had no visibility into where a car was.

They wanted to make cars visible throughout the process. We helped them map the process to get there and determined that we’d need to pull data out of inboxes and into a system where anyone could view a car’s status at any point. The company still has data silos in other departments, but fixing this one workflow has already created a lot of value.

Where to start: walk the floor, find the opportunity

Big picture, everyone is going to move faster, and that's a good thing. The tools available today let us build things for industrial companies that weren't realistic even a few years ago.

What still takes experience is knowing which problem deserves the speed. For us that means walking the plant floor, the railyard, or the back office, interacting with the people doing the work, finding where things get stuck, and putting a number on it before anyone commits to a bigger build.

Sometimes the answer is AI, and sometimes it's just a better way to reuse the drawings you already have.

That's how we approach our work with clients. We’re not looking for places to cram AI into your operations, we’re looking for the single highest-impact problem to solve. That’s what you’ll get with EdgeFinder: we spend three weeks with your team to find the highest-value opportunities in your operation and then scope out a potential solution before you make the bigger investment.

If you're staring at a list of requests and aren't sure which one to go after first, let's talk.


About the author

Gregg Wheeler is Chief Revenue Officer at TXI, where he leads go-to-market strategy across sales, marketing, and partnerships. He spends much of his time in conversation with industrial leaders, at events like the ones in this piece and with operators trying to figure out where to start. He has more than 25 years of experience in technology consulting, and his focus at TXI is helping companies that make, move, and maintain things find the problem worth solving first, then turn it into an intelligent product their people actually use.

Published by Gregg Wheeler

Let's start a conversation

Find out how intelligent solutions can accelerate growth for your organization.