Closing Status Gap Between Maintenance and AI

John Carrier

Ai Maint

At the 2017 MIT CIO Conference, Thomas Siebel, CEO of C3.ai, boldly declared that “predictive maintenance is the next killer app.” It was a sentiment that echoed through boardrooms globally, sparking a gold rush toward Industrial AI. Fast forward to today: many of those early tech pioneers have seen their market valuations corrected sharply, and the industry is entering a period of aggressive consolidation.

This creates a puzzling "Predictive AI Paradox." If the technology is the "killer app" of the decade, why are so many implementations failing to deliver a return on investment? The answer doesn't lie in the code; it lies in the culture and the lack of a seat at the executive table. 

 

The Predictive AI Paradox: A System Dynamics View 

To understand why predictive tools often fail, we must look at them through the lens of System Dynamics. Predictive AI is not a standalone solution; it is a feedback loop within a much larger, more complex system of human labor, spare parts logistics, and production pressure.

While there are numerous success stories applying AI to predictive maintenance, for every success, there may be several "quiet failures" where AI led to a degradation in system capacity. A key action enabled by Predictive AI is the ability to re-prioritize maintenance tasks based on an estimation of specific equipment’s Remaining Useful Life (RUL) on the PF Curve. Let’s examine two pathological dynamics at play: 

 

Pathological Dynamic 1: The Uncertainty Dilemma (The "Blame" Gap) 

As Yogi Berra famously noted, “it’s tough to make predictions, especially about the future.” When data is noisy, the AI produces an uncertain prediction. This can put the maintenance manager in an impossible "lose-lose" situation:

In this scenario, the AI becomes a tool for shifting blame rather than solving the underlying decision problem. As shown in Figure 1, it only serves to introduce more uncertainty and ambiguity into the decision-making process, not less.  

 

John Carrier Figure 1

 

Pathological Dynamic 2: The Maintenance Debt Spiral

When Predictive AI is introduced into an environment already burdened by excessive deferred maintenance, it can inadvertently trigger a "maintenance debt spiral." While AI accurately identifies "soon-to-be-critical" failures, the initial uncertainty and lack of confidence can become a source of paralysis rather than efficiency if the organization lacks the bandwidth to act on all the problems it identifies. Because the current capacity is entirely consumed by today’s urgent "fires," these impending failures—flagged by the AI—are systematically pushed back to make room for more immediate repairs. This turns the AI into an early-warning system that provides a high-fidelity countdown to a crisis it cannot prevent —a veritable digital Cassandra. As shown in Figure 2, each new maintenance break-in results in a spiral where a scheduled maintenance task is repeatedly recycled back into deferral until it becomes the next degraded equipment to be flagged by Predictive AI. 

 

John Carrier Figure 2

 

 The "Dirty Data" Reality: A Plant Tour Reflection

The lure of mathematical algorithms often blinds us to physical reality. During a recent plant tour, I was shown a state-of-the-art predictive system featuring over 300 wireless sensors. The dashboard was a sea of green and yellow lights.

However, within the first 50 feet of the walk-through, I pointed to a conveyor belt’s electric motor. It was visibly oscillating at approximately 10 cycles around its shaft centerline. Even an untrained eye could "predict" impending failure—no advanced tech needed!

This highlights a fundamental truth: You cannot "algorithm" your way out of poor fundamental maintenance. Without sufficient maintenance capacity and execution discipline, the potential value of Predictive AI cannot be realized. 

 

Buy the Result, Not the Technology

There is a well-known adage regarding power drills: the customer doesn't want a drill; they want a hole in the wall.1 Yet, in the current industrial climate, we are over-investing in the "drill." A recent industry whitepaper noted that up to $108 billion in annual AI investment is being wasted because of legacy data infrastructure.2

In the case of the 300 sensors mentioned above, nearly 20% were not collecting useful data due to poor placement or connectivity issues. The remaining data sat in a "data graveyard," unanalyzed. A better approach is the 80/20 rule:

  1. Start Small: Install 30–60 sensors on the "Bad Actors" (the most critical/unreliable assets).
  2. Close the Loop: Ensure the data leads to a Work Order, the Work Order leads to a Repair, and the Repair leads to a Root Cause Analysis.
  3. Scale Only After Success: Don't justify a $1M investment just because you’ve already spent $500k. 

 

The Dynamics of Deferral: The Maintenance Status Gap

In many plants, a familiar tragedy plays out: an important customer order conflicts with routine maintenance. Production takes precedence, and maintenance is deferred. In a facility running near capacity, this is not a one-time event; it is a cycle.

Why does this happen? Because the value of an additional sale is easy to quantify on a spreadsheet TODAY. The cost of deferred maintenance—increased wear, lower efficiency, and the risk of a 48-hour outage next month—is "invisible" to the current accounting systems.

This is where the Technology-Maintenance Status Gap occurs. We spend millions on AI to tell us when a machine will fail, but we don't give the Maintenance Manager the authority to say "No" to the Production Manager when the data says a shutdown is required. 

 

Lessons from the Pits: The Racing Analogy

Professional auto racing is perhaps the only industry where maintenance and production have equal prominence. A Formula 1 pit crew isn't a "cost center"; they are a "performance center." The efficiency of the tire change is a direct input into the race strategy. 

British humorist Jerome K. Jerome once said, “I like work; it fascinates me. I can sit and look at it for hours.” In racing, everyone looks at the work.

When was the last time your organization looked at how maintenance is done? Have you invested in the lighting, the tools, and the ergonomics that make it easier for a technician to be successful? If you are funding a $500,000 AI software package but your technicians are using 20-year-old wrenches in a dark basement, priorities are misaligned. 

 

Call to Action: Close the Maintenance–Technology Status Gap

Strategy exists to prevent organizations from optimizing for today at the expense of tomorrow. Today, most firms give technology a formal voice at the executive table. Production has one as well. But maintenance—the function responsible for the physical integrity, availability, and lifespan of billion-dollar asset bases—is too often treated as a downstream service, not a strategic partner.

This imbalance is the root cause of the Predictive AI Paradox. We invest heavily in algorithms that see risk earlier, while denying the maintenance organization the authority, capacity, and legitimacy to act on that information. Technology gains status; maintenance absorbs the consequences.

Closing this gap does not require a new title. It requires parity.

Organizations must raise maintenance to the same footing as technology when decisions are made—with equal authority to: 

Predictive AI only creates value when it is embedded in a strong maintenance planning, scheduling, and execution system with the power to act. Starve that system in favor of the “technology of the month,” and AI becomes little more than a high-resolution dashboard documenting decline.

The voice of maintenance must be as loud as the voice of the AI. When those voices are equal, predictive tools amplify human judgment. When they are not, technology accelerates failure — quietly, expensively, and predictably.

 

1 Christensen, Clayton M., Scott Cook, and Taddy Hall. "What Customers Want from Your Products." Working Knowledge, Harvard Business School, 16 Jan. 2006, www.library.hbs.edu/working-knowledge/what-customers-want-from-your-products.

2 Legacy Data Infrastructure Is Driving $108 Billion in Annual Wasted AI Investment, New Survey Finds, Hitachi Vantara Infrastructure Report January 27, 2026