AI Adoption in Manufacturing: Future Tech’s Matt Scavetta on Avoiding Last-Mile Failures

If it’s customary to close out the year by sharing big-picture industry insights, then it made perfect sense for Machine Design to connect with industry leaders to learn about what’s coming next.

One response came from Fort Lauderdale-based Future Tech’s chief technology innovation officer, Matt Scavetta. Future Tech is known for its expertise in AI-enabled infrastructure, PC-as-a-Service and secure, large-scale technology lifecycle management. The company works with some of the world’s most mission-driven enterprises. 

Scavetta’s role as chief technology innovation officer places him at the center of this innovation, identifying emerging technologies that bring value across industries and shaping strategies to make operations more scalable and efficient.

“I also serve as a bonus resource for our customers’ IT leaders,” Scavetta explains. “This includes helping them think through pain points they’re experiencing, sharing insights into how others in their industry are approaching similar situations, and connecting them with the right technology partners.”

READ MORE: Global Manufacturing Hits $46.7 Trillion, Key Technologies Drive Resilient Machine Design

In the conversation below, Scavetta unpacks current signs of AI adoption gaps, the pitfalls of last-mile integration, what success in 2026 might look like and how to avoid common missteps.

Machine Design: How can one tell that a manufacturer's AI investment is outpacing its workforce readiness or system design? How do these signs/gaps manifest on the plant floor?

Matt Scavetta: There are a few ways to spot this, with the first being budget allocation. Companies may invest heavily in AI technology and AI-adjacent roles, but their proportional investment in change management roles and training is lacking. You’ll typically see the imbalance on the P&L side of things. On the factory floor, this shows up when dashboards and new features are released, but people fall back on implicit or tacit knowledge, shadow spreadsheets or manual workarounds, essentially recreating old processes even though they’ve been trained on new tools.

There is data that illustrates this. McKinsey recently reported roughly 80% of global companies are using the latest generation of AI but haven’t yet achieved value from AI. This resonates in manufacturing, where we know skilled roles remain unfilled. Combined with rising retirement rates, this creates a measurable gap that’s increasing over time with more near misses or frequent all-hands-on-deck moments when teams barely hit a target or deadline.

Supervisors may also see that AI recommendations are overwritten. You’ll see AI outputs ignored, spreadsheets recreated and operators ultimately complaining that the new tools aren’t making them more efficient. All of those point to a lack of true adoption.

MD: Unpack what you mean by “last-mile failure” in the context of AI deployment. Where do the breakdowns tend to occur between data-driven insight and operational execution?

MS: By “last mile,” I’m referring to the period between when a model is validated and when it actually drives change in the manufacturing process or the plant. So, a model can be accurate, but if it isn’t well integrated into the manufacturing execution system, the human-machine interface or the standard operating procedures, you’ll see failures in the last mile.

Executing and scaling technologically feasible AI use cases looks similar to the challenges of last-mile delivery in supply chains. Without proper attention to integration details, AI on the plant floor can end up as an advisory tool that no one consults, or listens to, during a disruption. Or it could be mismatched KPIs that an AI algorithm is supposed to optimize or change.

MD: As IT-driven architectures increasingly dictate what happens at the edge, how is that changing the practical boundaries between controls hardware, embedded software and engineering skill sets? Is this shift amplifying or easing the constraints you see on AI performance and human-machine workflows?

MS: As more innovation comes from traditional IT teams, the old dividing lines—where one group owned machine controls, another business applications and a third data—are starting to blur.

If this convergence is handled well, it can ease the constraints on AI performance and human-machine workflows. But when you move from a world of many single-purpose tools to a singular edge computer that runs control logic, data collection, AI models and operator screens, you can run into problems if the architecture evolves faster than the people and processes. Then it can’t feel bolted on or haphazardly put together.

As I mentioned earlier, there has to be proper investment from operational teams in the overall buy-in and in changing of workflows so they’re centered around these AI advancements.

MD: When you say it’s “not a tooling problem,” what does that imply about how manufacturers approach AI integration? Is the tech layer overvalued in relation to human factors and organizational design?

MS: When I say it’s not a tooling problem, I’m reacting to the pattern where manufacturers obsess over questions like: Which cloud provider should we use? What technical stack or copilot should we choose? All the while, they’re spending far less time designing how a line lead, mechanic or planner will actually use the system at 2 a.m. on a Tuesday.

The focus needs to shift to operational workflows, adoption, training and collaboration. We know that very few companies have successfully put the entire package together and achieved real efficiency and scalability. But the high performers who have done it are the ones whose operational teams consciously redesigned their workflows, roles and governance around AI. These teams are putting pen to paper to clearly define who owns AI outputs, how those outputs flow into human–machine workflows, and how roles, KPIs and incentives change as a result.

Tooling usually isn’t the problem or limiting factor. The real work is in leadership, operational model design, aligned incentives and ultimately skills development. If your org chart and SOPs aren’t changing, then the AI tools you choose won’t matter much. Many organizations already have access to robust AI platforms—either cloud services or on-prem infrastructure like Dell PowerEdge or Dell APEX. The real differentiator is redesigning workflows and roles to take advantage of them.

READ MORE: Bringing AI to the Factory Floor: It’s Not as Simple as Plug-and-Play

MD: Please offer a real-world example where AI performance outpaced the surrounding operator/human interaction. What can manufacturers learn from this scenario?

MS: There was a company in the processing industry that adopted a digital assistant designed to help operators respond to abnormalities and disruptions in the manufacturing process. The results were initially positive as operators made better decisions and kept the process closer to target. But over time, the operators stopped scanning their monitoring applications as frequently and their overall understanding of the dynamic of the situations declined.

So, the risk was high that if the digital assistant failed or encountered a situation it hadn’t been trained on, operators might miss early warning signs and struggle to regain control. The takeaway is that training and interface design need just as much attention as the model itself. You have to make sure that users stay engaged and that those human-in-the-loop workflows remain effective.

MD: Looking toward 2026, what changes will be essential for manufacturers to turn AI potential into measurable productivity?

MS: The companies that win won’t be the ones running the most pilots. They’ll be the ones that wire AI into a small number of critical value streams—like uptime, yield, energy or change—over time. They’ll be tracking those hard metrics like OEE, safety and other core manufacturing KPIs.

To do this, there needs to be shifts. First is standardizing data and edge architecture so models can run closer to the production line. Second is treating the IT/OT convergence and digital twins as core infrastructure not [as] nice-to-haves or proof-of-concepts, but mission-critical technology. Finally, and really most importantly, it’s investing heavily in skills and continuous learning so the boots-on-the-ground operators and engineers can effectively partner around AI.

At this point, it’s less about discovering new use cases and more about leadership execution. The path forward is choosing a small portfolio of use cases, aligning cross-functional teams and measuring impact ruthlessly against the KPIs that matter in aligning these investments.