Better technology can improve analysis. But it cannot compensate for unclear definitions, poor decision logic, or inconsistent practices.
Artificial intelligence is rapidly becoming part of the conversation around maintenance, supply chain, reliability, and spare parts management.
That makes sense.
The potential is significant.
AI can analyze large volumes of data, identify patterns that would be difficult for a person to detect, help predict requirements, highlight anomalies, and improve the speed with which organizations process information.
Recent industry research also suggests that organizations are already applying AI and advanced analytics to maintenance and spare parts management.
But there is another question that needs to be asked:
What decisions are we asking the technology to make?
Because better analysis does not automatically produce better decisions.
Consider some of the apparently simple questions involved in spare parts inventory management.
Should we stock this item?
How many should we hold?
Is this part critical?
Can we reduce the current quantity?
Should this item be classified as obsolete?
Can another site provide backup stock?
Should we make a Last Time Buy?
Each question may eventually produce a number, classification, or system setting.
But getting there requires judgment.
Take criticality.
One person may call a spare part critical because failure could stop production.
Another may use the same term because the item has a long lead time.
Someone else may associate criticality with high cost.
Another person may simply mean, “This part is important to me.”
If those definitions have never been resolved, adding AI does not solve the problem.
It simply introduces another mechanism for processing an undefined concept.
A great deal of current discussion around AI focuses, quite rightly, on data quality.
Poor catalogue descriptions, duplicates, missing manufacturer information, incorrect lead times, and unreliable transaction histories all limit what any analytical tool can achieve.
But an organization can have relatively clean data and still make poor spare parts decisions.
Why?
Because data quality and decision quality are not the same thing.
Imagine that your ERP contains an accurate reorder point of four units.
That tells us what the system currently says.
It does not tell us whether four is the right number.
Perhaps the reorder point was established ten years ago.
Perhaps the equipment has since been modified.
Perhaps demand has changed.
Perhaps another site now holds the same part.
Perhaps the supplier lead time has fallen from six months to six weeks.
Perhaps nobody knows why four was originally selected.
The data may be perfectly accurate.
The decision behind it may no longer be valid.
This presents another challenge when applying AI to spare parts.
Historical data is not necessarily a record of good decisions.
It is a record of what happened.
If inventory levels have traditionally been excessive, historical purchasing patterns may reflect excessive inventory.
If every request to stock a new part was approved without meaningful challenge, the material master contains the accumulated result of those decisions.
If different sites have used different definitions of criticality, the historical classifications reflect those differences.
If reorder parameters have not been reviewed for years, historical settings may represent conditions that no longer exist.
This does not make historical data useless.
Far from it.
But it does mean that organizations need to understand what the data represents before assuming that technology can learn the “right” answer from it.
This is where the real opportunity lies.
AI can potentially make good decision processes faster and more scalable.
It can help identify parts that warrant review.
It can highlight inconsistencies.
It can detect unusual settings.
It can bring together information from different systems.
It can help analyze alternative stocking scenarios.
It may even challenge decisions that would otherwise go unnoticed.
But for this to work well, the organization still needs a basis for judging the answer.
For example:
What consequences justify holding insurance spares?
How should predictability of failure influence stocking?
How should lead time be considered?
When should stock held elsewhere in the network affect a local stocking decision?
Who can approve exceptions?
What evidence is required?
When should an existing decision be reviewed?
These are management and governance questions.
Technology can support them.
It cannot decide what your organization believes the rules should be.
Before worrying too much about whether an organization is “AI ready,” I would ask whether it is decision ready.
That requires several things.
First, people need a common understanding of the terminology they use. Words such as critical, obsolete, excess, and required need to mean something specific enough to support a decision.
Second, important decisions need transparent logic. People should be able to explain why an item is stocked, why a particular quantity is held, and what would cause that decision to change.
Third, the supporting data needs to be fit for the decision being made. That does not necessarily mean perfect data. It means knowing which information matters and having sufficient confidence in it.
Fourth, responsibilities need to be clear. Someone needs to own the decision process, not merely the data field in the ERP.
Finally, there needs to be a mechanism for reviewing and improving decisions over time.
Without these foundations, adding more sophisticated technology may simply make an inconsistent process operate faster.
None of this is an argument against AI.
The opposite is true.
The emerging technology is potentially extremely useful for spare parts management precisely because our inventories can be large, our data complex, demand intermittent, and the relationships between operational risk and inventory investment difficult to analyze manually.
But technology works best when it supports a sound management process.
The starting question should therefore not be:
“How can we use AI for spare parts?”
A better question is:
“Which spare parts decisions do we need to make better, and what logic, information, and governance should support those decisions?”
Once you can answer that, technology has something meaningful to amplify.
Without it, you may simply be automating the uncertainty.
Industry context referenced in this article: Sphera, 2026 MRO Survey Report, August 2026; McKinsey & Company, “Maintenance Meets AI: A Proven Approach for Asset-Heavy Industries,” September 23, 2026.
If this resonates with you, it is time to take action. Start here >>> Review Your Options.
Posted by Phillip Slater