CIOs, CDOs, and senior IT leaders at energy and utility organizations who have defined AI goals — predictive maintenance, grid optimization, demand response, intelligent asset management — and have a growing sense that legacy infrastructure is making all of it harder to execute. If your AI road map is advancing in meetings but stalling in implementation, read this blog post.
A pattern is playing out across the energy and utilities sector right now: An organization builds a credible AI strategy. Executive sponsors are aligned. Use cases are prioritized. Pilots are approved. Then the implementation work begins, and it immediately runs into a wall.
The data infrastructure is outdated.
That’s because the systems that hold the operational technology (OT) and IT data that AI models need were built for safety and security, not for a world of machine-readable, application programming interface (API)-accessible, governed data pipelines.
As a result, OT and IT data are siloed, but AI models require clean, accessible, well-structured data delivered through stable integrations.
“Currently, most utilities do not yet have the level of data readiness needed to fully realize AI’s value. Common challenges include limited data accessibility across the enterprise, unresolved privacy and security concerns, and inconsistent data quality,” says Chris Warren in the Electric Power Research Institute Journal.
Still, many utilities pursue legacy application modernization without integrating their siloed data.
The phenomenon is not restricted to utilities: Data accessibility and quality issues account for most delayed AI deployments in complex enterprises. However, the large number of legacy systems in utilities compounds the problem.
Modernization is not preparation for AI. It is the infrastructure layer AI runs on. Utilities that succeed in implementing safe, secure AI are those that start with siloed data before pursuing modernization.
Utility organizations that attempt to modernize first will likely fail. Let’s explore why.
The conventional enterprise AI playbook goes like this: define strategy, identify use cases, build models, then address infrastructure as needed. That sequence was designed for a world where AI was primarily a software-layer initiative. It does not work for OT environments such as utilities, where data lives in systems that predate the internet.
In those environments, a series of high-effort data engineering projects must be completed before any AI use case can go to production. Each one can add months to an AI road map that leadership believes is already funded and approved.
The alternative is not to delay the AI strategy but to design the modernization program alongside the AI strategy’s implementation. That way, the infrastructure decisions made during migration are the same decisions that determine what AI the organization can run, at what scale, and under what governance model.
That is a fundamentally different project scope — one that most utility IT organizations have not yet talked about. However, you must be aware of how your legacy systems may already be blocking your readiness requirements.
AI readiness frameworks often focus on data quality. That is necessary, but not sufficient. For energy and utility organizations, four requirements must be in place before AI use cases can move from pilot to production at scale:
Let’s explore each of these AI data readiness requirements for utilities in more depth.
AI models require data in structured, consistently formatted pipelines that systems can query. The legacy operational systems still found in utilities deliver data in batch exports, proprietary formats, and siloed schemas designed for the applications that created them, not for downstream consumption. AI-augmented modernization rebuilds these systems with data accessibility as a first-class design requirement, not an afterthought.
Use cases like predictive asset maintenance and dynamic demand response require data to flow between systems in real time. Legacy SCADA/historian time series systems that feed real-time operational data into an event-driven layer (Apache Kafka or MQTT, close to where the physical equipment actually lives) must coordinate with DERMS or ADMS architectures. Legacy point-to-point integrations and nightly batch processes cannot support this real-time integration.
Utilities already operate under the North American Electric Reliability Corporation (NERC) critical infrastructure protection (CIP) audit and data lineage obligations, and they answer to their public utility commission (PUC) requirements.
However, AI outputs have higher data lineage standards. NERC compliance gives utilities a defensible paper trail for grid reliability and cybersecurity purposes, but meeting those requirements does not produce the unified, semantically consistent, machine-ingestible data that AI agents require.
Feeding AI models with data from legacy systems that have no audit trail, inconsistent data definitions, and undocumented transformations creates governance risk that grows with every AI use case deployed on top of it.
Modernization is not just a technical change. It changes how field crews access data on the job, how grid operators monitor real-time conditions, how asset management teams understand equipment status, and how control-room operators monitor power plant performance. Organizations that design the operating model alongside the technical migration — not after it — are better prepared for safe and reliable operations, and they can adapt quickly to changes.
Imagine a midsize investor-owned utility that has board approval for an AI-driven predictive maintenance program. The business case is strong: aging transmission infrastructure, rising maintenance costs, and a workforce with deep institutional knowledge approaching retirement age.
Implementation begins. The data science team requests historical asset performance data. The operations team surfaces it from three separate legacy systems: an AVEVA/OSI PI historian, a Maximo enterprise asset management (EAM) system, and a geographic information system (GIS).
Each of these assets was built by a different vendor over a span of 15 years. However, the data formats are incompatible, the asset ID schemas do not match, and two of the three systems require manual exports with a 24-hour delay.
The AI program does not fail because the model is wrong. It stalls because the IT and OT data infrastructures were not designed with this use case in mind — and fixing the problem now, mid-program, costs more in time and money than addressing it would have during the modernization planning phase.
This is the sequencing problem, and it’s almost entirely preventable.
Across our AI strategy engagements with energy and utility clients, the same pattern surfaces: Though the goal is to lower IT costs, modernization investments were scoped in the context of rate recoverability and fiscal responsibility for regulators and shareholders. AI use case requirements are never part of the design conversation, and competing interests among the companies and stakeholders complicated the discussions.
So, by the time the AI program is funded and in-flight, the infrastructure decisions are already made. Walking them back is expensive, and working around them is slow.
The organizations that avoid this recognize that the many complex negotiations around two basic questions — “How do we modernize our legacy systems?” and “What does our AI strategy require?” — must happen with all stakeholders at the table.
That makes modernization a governance and process change initiative as much as a technical one.
If you are a CIO or CDO planning an AI program right now, these questions will tell you whether modernization needs to move up your priority list:
If the answer is primarily in legacy operational systems, you already have a sequencing dependency. The AI program is downstream of a modernization requirement that has not yet been acknowledged in the budget or timeline.
If the answer is no, the barrier is not the AI strategy. The barrier is the infrastructure the AI strategy requires. That is a different problem with a different solution — and a different funding conversation.
If the answer is yes, you have a coordination gap between two programs that are making decisions with consequences for each other — without a shared design process.
The three questions above are diagnostic. Here’s what to do if you answered yes to any of them:
Map the data your top three priority AI use cases require and identify which legacy systems hold it. Then bring that map into your modernization planning process before finalizing architecture decisions. The goal is to influence how the data layer and integration architecture are designed so the rebuilt systems deliver data in formats your AI models can use.
You have an infrastructure funding conversation to have — separate from the AI strategy conversation. The distinction can be important for timely rate recovery. The data engineering work required to bridge legacy systems to AI models is a program dependency, not a project task. It needs its own line in the budget and its own owner in the project governance structure. Naming it clearly is the first step to addressing it.
You have a coordination gap between two programs. For example, the gap may be between your IT and OT organizations. Neither of them has AI-ready architectures, and both are making decisions independently. The fix is a designated decision point.
Before your modernization program commits to any major architecture choice, require a sign-off from whoever owns the AI strategy. A two-week joint design review before architecture is finalized is typically enough to surface the decisions that matter most.
The sequencing problem is already in motion. We recommend pausing any architecture decisions in the modernization program that have not yet been finalized and running a rapid joint review with both the IT modernization team and the AI strategy team in the room.
Your goal: Identify the two or three infrastructure decisions that most directly affect your AI road map, and make sure those decisions are made with both programs’ requirements in view. That review will take less than a week. Skipping it typically costs six to 18 months of rework later.
We aren’t saying that AI initiatives should always wait for a completed modernization program. In fact, some utility AI initiatives don’t require core system modernization. Vegetation management, for example, relies on computer vision that is not part of core systems.
Furthermore, long modernization programs are themselves a risk — and AI-augmented approaches have compressed them significantly. Work delivered with modern AI agent frameworks has achieved timelines approximately 80 percent faster than traditional approaches, with cost reductions of 30–50 percent.
Instead, modernization and AI strategy should be designed in parallel because the architecture decisions made during migration determine what the AI program can do, at what speed, and with what governance structure.
Modernize without an AI strategy and you get a cleaner legacy system. Modernize with an AI strategy built in and you get a platform that was designed from the start to support the use cases you have already approved and the governance model you will eventually need.
The decision that matters most is not when to modernize. It is whether you modernize with your AI future built into the design.
For energy and utilities organizations working through both sides of this problem — modernization and AI strategy — we can help. Centric Consulting can assist with AI-augmented legacy migration and AI strategy, governance, and operating model design built on the platform the modernization creates.
