Data Precedes Artificial Intelligence

There is a simple proposition at the heart of the debate over artificial intelligence in military sustainment, and it is easy to lose sight of amid the excitement about new tools. Data precedes AI. The quality of what an AI system concludes can never rise above the quality of what it is given to reason over, or above the ability of the enterprise to act on what it concludes.

That proposition matters now because a familiar instinct is shaping how the Pentagon approaches maintenance. AI is treated as a tool, and tools can be dropped into workplaces. Buy the software, hand it to the maintenance department, and readiness will follow. A recent NAVAIR draft request reflects that approach. The motive is sound, but the approach starts in the wrong place.

AI Is a Logic System

Strip away the marketing and an AI system is a logic system. It takes inputs, applies rules or learned patterns, and produces conclusions. Logic has an old lesson here: a valid argument built on false premises yields conclusions that are perfectly well formed and perfectly wrong. In maintenance, the premises are data.

If the underlying records are incomplete, inconsistent, late, or locked in systems that do not talk to one another, the AI will process them faithfully, quickly, at scale, and with an air of authority. The result is not better decisions. It is fast management of bad data.

Good data in this context has a specific meaning. It is accurate. It is traceable to a specific, serial-numbered part on a specific aircraft. And it is captured at the point where the work is done, by people who trust it enough to use it.

Two Halves of One Problem

AI in sustainment has two targets. The first is the aircraft. Condition-based maintenance, fatigue damage monitoring, and the digital twin all replace statistical guesses with the actual history of a specific component. Instead of changing a part after a fixed number of hours, you change it based on its real condition.

The second target is the industrial base: the suppliers, lead times, demand signals, and inventory that determine whether a part is where it needs to be. Sikorsky’s Ariel tool, as I understand it, is aimed here, working with the Defense Logistics Agency across more than 6,000 spare parts.

These are not competing applications. Predicting that a component will fail in 400 flight hours is worth little if the replacement has a 14-month lead time and the supplier has not been told. One half tells you what will be needed. The other tells you whether the enterprise can deliver it. A force with only one has half an answer.

The Enterprise Comes First

What joins the two halves is the digital thread, the unbroken record that follows a part from design through manufacture, operation, repair, and retirement, and a contract that rewards the right outcome.

The CH-53E experience is instructive. In 2018 Sikorsky won a performance-based logistics award from NAVSUP that added an aircraft availability metric to the traditional measures of parts delivery. That small change shifted the incentive from having parts in a warehouse to having aircraft that fly. The data environment that came with it, the Fleet Common Operating Environment, allowed the parties to link a change in parts to a measurable change in availability. That is precisely the connection an AI needs in order to learn anything.

The sequence is the lesson: enterprise, then data, then AI. Not the reverse.

The cautionary side comes from a 2021 Department of Defense Inspector General audit of DLA’s long-term Captains of Industry contracts. The auditors found estimated savings that had not been validated against performance, and a contract without performance metrics in which on-time delivery fell to 53 percent against a 90 percent goal. A performance-based arrangement without trustworthy data is an act of faith. AI working on top of it will either sharpen accountability or conceal its absence.

Why It Matters for a Distributed Force

The Marine Corps is building a distributed force that must operate for defined periods in contested environments, far from fixed support. The CH-53K is central to that force, and it is a smart, digital aircraft. But it is a poor bargain to place a smart aircraft inside a dumb enterprise.

If a unit must sustain operations for a defined period, planners can work backward. Which components will reach their limits in that window? Which have long lead times? Which depend on a single supplier? Answering those questions requires good data and an enterprise able to act. Made with bad data by a system that accelerates the error, the decision about where to place a part becomes a hidden vulnerability.

That is why sustainment is a combat function, and why the questions that decide the outcome are old ones: Is the data fit to reason over? Is the enterprise built to act? Do the two halves connect? How will we know it works?

The real choice facing AI in sustainment is whether we build the data and the enterprise first, or buy the tool and hope the rest follows.