The Asset Journal

The Asset Journal, Volume 20, Issue 2

AI for Maintenance sounds like the answer many of us have been waiting for. But what should we realistically expect from AI in this context? Too often, we look for a ‘universal solution’ that tells us everything we think we want to know, rather than what we truly need to know, and we assign that role to AI. Before we apply AI across the business, we need to define clearly what Maintenance needs to know and what the wider business needs to know about Maintenance.

Many practitioners and analysts have found that AI itself is often not the problem. Failures usually stem from limited capability in using it effectively and a poor understanding of where it should be applied. In Maintenance, we need methods and tools that provide meaningful insight. This includes access to key information on asset performance, utilisation, condition, and reliability and more, ideally at the touch of a button. It also includes visibility of workflow issues familiar to planners, supervisors, and superintendents. At the same time, finance teams want a clear view of maintenance expenditure so they can support better investment decisions. The key question is: what exactly should AI be searching for and answering?

In my view, defining the information we need is only part of the challenge; we must also ensure the data is complete and reliable. Many organisations still work with fragmented, inconsistent datasets spread across multiple systems, which can lead AI—whether agentic or anthropic—to produce biased or misleading results unless the boundaries of the data being queried are clearly set. Claims that AI can reduce maintenance effort by 10% to 30% are certainly attractive, but we need to ask what those reductions would actually mean in practice. Could risk increase? Would the savings be sustainable? How do we guarantee performance? These are some of the fundamental questions for anyone considering AI in Maintenance. It should also go without saying that every analytical result, including AI-generated opportunities, still requires human judgement to validate the actions to be taken.

Given the complexity of Maintenance, particularly if it is to become a future profit centre, we first need to understand the foundations required for AI to deliver good results. Asset registers and asset information must follow a consistent structure and syntax across the data repositories and the organisation so that people can access the right information reliably. That will require effort, as will defining clear boundaries for AI and aligning purpose across departments to protect the integrity of the information retrieved. Before expecting AI to deliver its full promise, we may achieve early value by using it to identify inconsistencies in maintenance strategies and tasks for similar assets across the business. It may also help reveal planning, scheduling, and other workflow issues that often remain hidden. Used this way, AI could become a practical training ground for engineers, maintainers, support teams, and managers as they shape the future of Maintenance in the age of AI.

As always, we value your experiences and feedback and trust you will enjoy this edition of “The Asset “.

publications@amcouncil.com.au