What It Really Takes to Be Ready for Operational Intelligence, Part 2
“Go live, to me, is kind of just the start of where the value gets created.”
It’s not what most people expect to hear about a technology project, and it’s one of many ideas from Rob Bansek and Devon Keane’s podcast conversation that challenge how organizations think about AI readiness.
The timing couldn’t be better. Deloitte’s AI in Manufacturing 2026 survey found that 84% of manufacturers already see measurable value from AI, yet only about one in five use cases has been scaled across sites or the enterprise. So what’s getting in the way? In Part 1, Rob and Devon explained why technology is often the easier part of operational intelligence (AI)
When Smart Isn’t Enough
On paper, Devon’s first story should have been a success. The R&D group behind it was, in his words, “some of the smartest and most impressive people who I’ve ever worked with.” They had data scientists, hardware engineers, and a lab where they built their own sensing kit and a machine learning model to guide frontline operators.
Then two things happened that no amount of engineering could fix. The business shifted its priorities, but the model kept optimizing for the old ones. And when its recommendations reached the floor, operators weren’t allowed to make the changes it suggested.
Here’s what makes the story worth hearing: neither problem showed up in the data. Both came from teams that saw the problem differently and never walked the process together. If your organization has a promising pilot that never made it past the first line, there’s a good chance some version of this story is behind it. In the full episode, Devon explains how it unfolded and the lesson he carries into every project now.
Rob’s story is simpler, and that’s what makes it stick. At a water utility, an AI model recommends moving water first thing in the morning, when energy is cheaper. It’s a perfectly logical call, right up until an afternoon thunderstorm rolls through the South and the water is in the wrong place. “There are brilliant people that run these models, way smarter than me, that understand the math behind them,” Rob says. The math isn’t the problem. The model just doesn’t know what the people running the process know.
Now raise the stakes. The episode also looks at a new AI recommendation engine for air traffic controllers, where a single recommendation can affect thousands of lives. What does it take to build a system people can trust when the cost of being wrong is that high? It’s the same question every plant faces, just at a different scale, and the answer always comes back to context: a model only knows what it’s been given.
Choosing Where to Start, and What Comes After
So how do you keep your project from becoming the next cautionary tale? Since joining InSource, Devon has been leading use case discovery workshops that put leadership and the frontline in the same room to decide which problems to solve first. That means weighing value and effort, but also data readiness, compliance, competitive advantage, and what the business needs most right now. Devon doesn’t sugarcoat why it matters: “If you’re going to invest some dollars into a new technology, into a new process, you better have a return on investment. Otherwise, that technology is probably going to get cut.”
But here’s the twist: the biggest number on the board usually isn’t the best place to start. The use case with the highest potential ROI is often the most complex, with more data to connect, more teams to align, and more change for the floor to absorb. If an organization isn’t ready for that, the project is more likely to stall. A smaller, well-defined win builds confidence, proves value, and creates momentum for the bigger projects that follow.
Even then, picking the right problem is only half the battle. Rob has watched organizations define the problem, feel ready, and then get blindsided by everything that comes next. People have to adopt new ways of working, someone has to own the results, and the impact has to be measured against what success was supposed to look like. Most teams treat go-live as the finish line. For Rob, it’s the starting line, and the rigor that comes after launch is often what separates a project that scales from a pilot that quietly fades away.
Manufacturers Aren’t Short on Data
“Our data isn’t good enough.” It’s one of the most common reasons organizations give for holding off on AI.
Industrial organizations have been collecting data through historians, MES systems, and other tools for 40 or 50 years. The data is there. What’s changed is what we’re asking it to do. Data collected for reporting is now being asked to power analytics and AI. Treat that as a data problem, and it’s easy to spend months and a lot of money chasing a unified namespace, a data lake, or a move to the cloud before anyone knows what’s actually needed.
Devon agrees that data is real work. He calls it probably the largest technical challenge behind today’s AI solutions. But trying to fix it all at once can feel like trying to “boil the ocean.” His answer is to flip the order: start with the use case, and the data questions get much easier. Which data matters? Where does it live? Who owns it? Each use case builds a little more of the foundation and solves a real problem along the way.
That practical, problem-first approach is part of what drew Devon to InSource, which works across the entire stack, from HMI and SCADA to historians, MES, CMMS, and the analytics and AI layer on top. It’s also the moment Rob decides to make the sales pitch, “just because I can.” After roughly 30 years in business, InSource serves more than 8,000 active manufacturing and industrial customers. The new practice builds on that foundation, helping organizations define the problem, understand its value, and get ready to solve it. And maybe even “bring some joy” along the way.
Source:
1: https://www.deloitte.com/ch/en/Industries/industrial-construction/perspectives/ai-in-manufacturing.html
Hear the Full Conversation
Hear Rob and Devon’s full conversation on what it really takes to make operational intelligence work, from the shiny tool to the people who make the final call.