Two recent industry studies examine different parts of the AI and maintenance landscape. Each is useful on its own terms; neither provides a directly comparable benchmark for the other.
MaintainX reported that 65% of organisations expected to implement AI-powered maintenance solutions by 2026, based on a 2025 survey of 1,320 maintenance professionals in the United States and Canada. Separately, Fluke reported 18% predictive-maintenance adoption in a 2026 Censuswide survey of more than 600 senior decision-makers and maintenance professionals in the United States, United Kingdom and Germany.
The studies use different samples, geographies and measures, so their percentages should not be subtracted. Read separately, they still support a practical question for asset leaders: what must be in place before AI interest can become governed operational work?
Four practical checks for maintenance-AI readiness
Technology is only part of an operational deployment. The following checks are a practical starting point for maintenance use cases, not an exhaustive or validated readiness model. The evidence required depends on the decision and its consequences.
1. Consistent asset records
For asset-specific decisions, records need a consistent identity and meaningful context. Registers can contain outdated structures, duplicate equipment or mismatched identifiers across systems. Test the records your use case will actually consume rather than assuming a complete register is a reliable one.
2. Current asset-criticality assessments
Work prioritisation needs consequence and service context, not just a predicted failure. Check whether the relevant assets have current criticality assessments, clear assumptions and an owner. Combine that context with condition, operating needs and available resources.
3. History suitable for the selected method
A supervised failure-prediction model needs suitable outcomes and historical context. Other approaches may use condition signals, anomaly detection or engineering rules. For each method, test whether the available history, sensor coverage and labels support the proposed decision; do not assume missing evidence can be recreated reliably.
4. Decision ownership
When a model recommends earlier intervention, who is accountable for acting—and through which governed workflow does a prediction become a work order? Without a named owner, review rules and an operational path, predictions accumulate in dashboards. A prediction that cannot become accountable work is only another information product.
How to check the evidence
Each condition is measurable. Hierarchy integrity can be sampled against defined rules. Criticality coverage is countable. Failure-history usability can be tested on representative records. Decision ownership can be mapped to review, override, abstention and escalation paths.
Turn a broad question about AI into a bounded readiness review. A scorecard can organise the discussion, but cannot establish safety, model fitness or production approval. Reuse sound data foundations where appropriate and validate the requirements of each new use case.
Privacy and platform changes to consider
From 10 December 2026, additional Australian Privacy Principle obligations require an APP entity to include information in its privacy policy when it arranges for a computer program to use personal information in decisions that could reasonably be expected to significantly affect an individual’s rights or interests. A maintenance agent is not automatically in scope; the trigger depends on the personal information used and the decision’s effect. But the obligation reinforces a sound operating principle: document the data, decision, accountability and control path before automation becomes consequential.
Spatial platforms are moving too. Esri now documents migration tools and workflows for organisations moving existing data, including geometric-network data, into Utility Network. Where an asset owner is already undertaking that migration, it is a sensible time to clarify the GIS–EAM identity, ownership and integration seam—once, with clean data and explicit lifecycle rules.
Use the checks in a planning review
If you sponsor an asset-AI initiative, begin with four questions: Can we trust the hierarchy? Is criticality covered? Is failure history usable? Does the prediction have an owner and a path to work?
- Below 40%: review the most consequential data and ownership gaps with the people responsible for the use case.
- 40–70%: check whether one bounded use case has sufficient evidence, controls and ownership for a useful proof.
- Above 70%: stronger foundations—validate the next step in a scoped use case, with production controls, review boundaries and decision ownership defined.
These are Hastha’s indicative discussion bands, not empirically validated thresholds, an audit or approval for production deployment. A material safety or data gap takes precedence over the total score.
Questions asset leaders ask
Why are the two industry studies presented separately?
The 2025 MaintainX study reports expectations about AI-powered maintenance among maintenance professionals in the United States and Canada. The 2026 Fluke study reports predictive-maintenance adoption among a different population in the United States, United Kingdom and Germany. Their samples, geographies and measures differ, so subtracting one percentage from the other would not create a valid benchmark.
What should an asset team check before an AI pilot?
For maintenance use cases, review relevant asset identity, criticality, usable history and decision ownership. These are discussion prompts, not exhaustive prerequisites for every AI method. Validate the evidence, controls and acceptance criteria for the specific decision.
What does an AI readiness score below 40% mean?
In this indicative self-assessment it flags data or ownership gaps for further review. The bands are not empirically validated thresholds, an audit or production approval. A material risk or evidence gap takes precedence over the total score.
Do Australia’s December 2026 automated-decision rules apply to every maintenance AI tool?
No. The additional APP privacy-policy obligations apply where an APP entity arranges for a computer program to use personal information in decisions that may significantly affect an individual’s rights or interests. Scope depends on the data and the decision.