Why Maintenance Orchestration Is Still So Difficult

Maintenance Data Has Not Been Productized

The raw ingredients for AI-driven maintenance exist at most energy and utility organizations. Work order histories go back decades. Asset records cover thousands or millions of operational assets, from poles, transformers, and meters to pipelines, processing equipment, generation assets, and treatment facilities. The problem is not that the data is missing, but rather that the data has not been productized. It has not been structured, governed, and made reliably accessible in ways that downstream consumers, ML models, automation workflows, and AI agents can actually use.  

Failure codes illustrate the issue clearly. A single organization might have dozens of different ways to describe the same failure across multiple sites, facilities, and CMMS instances, often the result of system migrations, local customization, or years of inconsistent entry by field crews. An ML model trained on that data cannot reliably learn failure patterns because the label for the same failure is different across sites and facilities. Cleaning and standardizing that taxonomy is not a data hygiene exercise. It is the prerequisite for every AI investment that follows.  

The same pattern shows up in asset master records, which are frequently fragmented across EAM systems, GIS platforms, and legacy databases with different hierarchy structures and uneven coverage. It shows up in sensor and historian data that generates readings from production equipment but is not connected to the analytics environment where those readings could drive decisions.  

Data productization means treating these assets as products with defined consumers. It means asking what an ML model needs from work order history to train on, what an agentic workflow needs from asset records to query reliably, what a compliance reporting tool needs from PM completion data to extract consistently, and then building the data infrastructure to deliver those things. That work comes before the AI investment, not after it. 

Planners Spend More Time Coordinating Than Planning 

Maintenance planning across most energy and utility organizations is labor-intensive in ways that are not visible on an org chart. Planners spend the majority of their time not making decisions about maintenance priorities but gathering the information they need to make those decisions.  

Planners routinely gather information from:  

• Asset history from the EAM  
• Parts availability from the inventory system  
• Permit requirements from the safety system  
• Crew schedules from the workforce management tool  

None of these systems communicate with each other natively, so the planner navigates between them by hand.  

The coordination overhead is huge. Each day, planners have to spend hours gathering data and coordinating instead of actually planning maintenance work. A single work package, assembling asset history, confirming parts, checking permit lead times, and validating crew availability can take two to six hours to put together manually. At the volume a mid-size energy or utility company generates in a week, that is a significant hidden cost that does not appear in any maintenance budget line. Instead, it shows up as deferred work, schedule slippage, and crews arriving without what they need.  

Much of this coordination is rule-based and repeatable. Tasks such as extracting PM completion data and assembling it into a compliance report, creating a work order from an inspection finding, and triggering a parts reorder when stock drops below a threshold do not require human judgment. These are process automation opportunities that are consistently underutilized.

The coordination that does require judgment — assembling a complex work package from multiple information sources or routing a permit with conflicting constraints — is where Agentic AI creates leverage. The agent handles retrieval and assembly. The planner reviews a complete picture and makes the decision, rather than spending the first two hours of the day building that picture from scratch.

Condition Monitoring Is Generating Work, Not Enriched Work Orders 

Condition monitoring has improved significantly. APM platforms, asset monitoring systems, operational analytics, and drone inspection programs are surfacing equipment issues that would previously have gone undetected until failure. That capability is real and the investment is justified.  

The problem is what happens after the alert fires. In most programs, the step from detection to work order creation is still manual. Someone reviews the alert, decides it represents a real maintenance need, and creates a work order by hand. The work order that comes out typically describes the symptom and not much else: 

• There is no job plan attached  
• There is no bill of materials  
• No tooling list  
• No indication of which permits will be required or how long they take to obtain  
• Relevant safety procedures are not included  

That work order lands in a planner's queue and waits. The planner researches what the job actually involves, finds or writes a job plan if one exists, confirms parts, and checks permit requirements, before the work can be scheduled.  

If the work order came from an automated detection system, the detection was automated but everything after it was not. This means that condition monitoring programs generate work without generating ready-to-execute work orders. Crews eventually arrive at jobs where the most critical information, the right procedures, parts confirmation, permit status, relevant asset history, was assembled at the last minute or not at all. First-time fix rates suffer, rework follows, and a portion of the value of the detection investment is lost in the execution.  

"It's relatively normal to be assigned a work order and discover you don't have the parts or tooling needed to do the job. Everyone is exasperated, but it still happens."

The opportunity is automating the path from condition detection to an enriched work order that already contains what the planner and crew need:  

• The job plan  
• BOM  
• Tooling requirements  
• Permit checklist  
• Relevant JHAs  
• Asset history

That assembly is a job for AI-driven workflow automation, not a job for the planner to do from scratch each time an alert fires. 

The Problems Feed Each Other 

There's another challenge across the industry: a meaningful portion of the energy and utility maintenance workforce is approaching retirement. As that institutional knowledge disappears, the underlying maintenance processes become even more difficult to standardize and scale. That's a challenge deserving of its own discussion.  

Stale data limits what automation can reliably do. The coordination work still falls on people. Unenriched work orders and incomplete maintenance records make it more difficult to standardize maintenance processes over time. And all of it affects the quality of the data that feeds back into the next generation of AI models and analytics.  

The good news is that because these problems are not independent, solving one often creates momentum for solving the next. For many energy and utility organizations, the first step is identifying where the data gaps actually exist and establishing a clear sequence for what comes next.  

Perficient works with energy and utility organizations, including oil and gas operators, across the full maintenance orchestration stack. Explore more of our Automotive & Industrials expertise.