How to Get Predictive Maintenance Off the Ground
Driven by increased adoption of Industry 4.0 technologies, specifically AI, manufacturers still rank predictive maintenance use cases at the top of their priority lists. Yet despite strong interest and recognition of key advantages, manufacturers are struggling to parlay compelling proof of concepts into something that delivers value at scale.
According to Grandview Research, the global predictive-maintenance market is expected to grow from $14.29 billion in 2025 to $98.16 billion by 2033, expanding at a CAGR of 27.9%. A 2022 report from Deloitte shows early adopters already enjoying quantifiable benefits from predictive maintenance use cases. Among them: Up to a 15% reduction in downtime, a 20% increase in labor productivity, and a 30% reduction in inventory levels with less need to stock just-in-case parts, the Deloitte research found.
Automation World tapped two experts in this area to dig into what’s stalling predictive maintenance, as well as how to accelerate adoption. Michael Cooper, manager of post-sales solution engineering and reporting at Fiix by Rockwell Automation, and Matt Bernhard, senior solutions architect at TwinThread, shared their perceptives.
Q: What do you see as the top three barriers preventing manufacturers from adopting predictive-maintenance systems?
Bernhard: The first one is typically data readiness, meaning having the data available to actually drive value for these use cases. Customers know they need to build the data systems and infrastructure to drive value, but the executive team is pushing for an AI solution first. While AI is the shiny new object, you have to find something that can actually drive value.
Context is also critical to gaining value. Take, for example, a predictive-monitoring use case for rotating assets—they don’t always know what products are running on the line, what the operational state of the equipment is or what the quality metrics are. Those are things that can have an impact on overall performance and asset health. Without that data and context, you are not getting the whole picture. That could lead to maintenance-reliability teams not trusting the data from a predictive-monitoring system because it’s missing that important context.
Cooper: From what I’ve seen, there are three core issues to adopting predictive maintenance. Deployment complexity as most implementations still feel like custom projects and don’t scale well across sites or assets. Many teams assume the need for a full transformation instead of knowing where to begin and how to phase it. Finally, there’s typically an ownership gap so even when deployed, customers struggle with who owns it long term and how to sustain it. These show up differently by size. Smaller manufacturers tend to lack resources to even start, while larger ones struggle with scaling across multiple plants consistently.
Q: How significant is the lack of in-house technical expertise as an obstacle, and what specific skill gaps do you encounter most frequently?
Bernhard: Most manufacturers, unless they are very large companies, don't have a very large internal data-science team. And if they do, they tend to be deeply technical and not as familiar with the maintenance or processes happening on the plant floor. There ends up being a disconnect. On the sensor side, the biggest challenge is with the translational layer, which goes back to that idea of needing context.
In order to make all this successful, you need to have someone in the loop who is the subject-matter expert—they understand what the data is and have a general understanding of what’s important to drive value for the use cases. That person could be a process engineer or a reliability leader, but they are driving this because they understand both the business implications and technical needs.
Cooper: The biggest gaps are not data science or machine learning (ML). They are hardware (OT) integration and getting machine signals into a usable state; translating signals into actual maintenance actions; and ongoing ownership of thresholds, tuning, and system behavior. Change management is also a challenge. Teams need to redefine what they do differently or stop doing once Condition-Based Maintenance (CBM) is in place. And if the system requires a developer to maintain, it rarely sticks.
Q: Beyond the initial software licensing costs, what are the hidden or often-underestimated expenses that cause sticker shock for potential customers?
Bernhard: We tend to not run into this issue because we have a fairly straightforward licensing approach that is clear to our customers. It’s not like these LLMs that are token-based, which you start using and you end up with a $500,000 bill that you didn't expect. Where sticker shock tends to occur is when software solutions weren’t built thinking about scale so users are priced out of expansion.
Cooper: The software itself is rarely the biggest cost. Integration effort, internal resources, and change management are what ultimately drive the total cost of ownership.
Q: What's the realistic timeline and investment required before manufacturers start seeing measurable ROI from predictive maintenance, and how does this align with typical corporate budget cycles?
Bernhard: This is one of the most important things to talk about with your potential suppliers. Manufacturers need to focus on the business value first, not the technology. We work with customers to find those high-value use cases that can prove value in 90 days or less. Then we talk about expansion planning in the six-month range so they can continue to see value across either multiple lines within the same facility or potentially multiple use cases across different facilities.
Cooper: You can see value relatively quickly on targeted assets, often within months. For example, with Fiix CMMS + FactoryTalk Optix, maintenance teams can quickly realize value through faster response times to issues, reducing inspections, and more accurate downtime tracking. The real long-term ROI comes from scaling these solutions across assets and sites once change management is complete, and new best practices are implemented to optimize for CBM, which is where most programs struggle. There is often a mismatch with budget cycles that expect immediate, predictable returns.
Q: How are you positioning predictive maintenance? Is it a gradual journey rather than an all-or-nothing investment? Do you offer pilot programs, phased implementations, or entry-level packages that allow manufacturers to start small?
Bernhard: Our philosophy is think big, start small, move fast. Companies can leverage our solution templates for things like asset reliability or asset anomaly detection use cases to start. They start small, solving a specific problem, and we help them expand with other solutions templates for things like center lining machines or optimizing throughput on a line. We'll build out the solutions that are going to be relevant to them as they start to scale and stabilize the process of optimizing their maintenance and reliability operations.
Cooper: This is the most important shift. Predictive maintenance must be positioned as a journey. Start small with high-value assets, deliver quick wins, standardize how deployments are repeated, and gradually shift ownership to the customer. If predictive maintenance is positioned as a large upfront transformation, most customers will not even start.
More maintenance insights from Automation World:
How a Two-Week Maintenance Overhaul Cut Downtime by 10% at a Major Snack Food Manufacturer
