The Data Imperative: Why AI Cannot Compensate for Poor Information
Amit Golhani, Director of Research and Development
Every energy enthusiast I speak with today has AI somewhere on their agenda. The ambition is real, and the urgency is justified due to challenges in the energy sector be it scarcity of natural resources, energy independence, harnessing new but intermittent sources of energy coupled with storage, transition in mobility, decarbonization at speed - all simultaneously. AI, in theory, is purpose-built for this level of complexity where the human brain stops analyzing data in multi-dimension. But after working across utilities, grid operators, product organizations and energy transition businesses, I have arrived at a conviction that most organizations are missing that data is the foundation for any credible AI strategy - and domain expertise is the glue that makes it hold. Without both, you don’t have an AI strategy. You have an expensive experiment, and you might have noticed that many organizations lose their annual budget in weeks and months in such experiments.
The Foundation: Data You cannot build an AI strategy on a broken data foundation. And in the energy sector, the foundation is, for most organizations, cracked. Not because data is scarce - it isn’t. Systems - be it metering, operational, market- and customer-facing - generate extraordinary volume every second. The problem is that this data is siloed, inconsistent, poorly labeled, and disconnected from the decision it is supposed to inform. AI does not fix bad data. It amplifies it. I believe the energy organizations that will lead the AI era are making a board-level decision to treat data as critical infrastructure - with the same rigor they apply to physical assets. That means knowing what data you have, what it is worth, how fresh it is, and what decision it can reliably support. It means designing data architecture around decisions, not around legacy systems. And it means treating data quality as an operational discipline like security and privacy, not an IT housekeeping task. In the energy sector, poor data quality is not merely an analytical inconvenience. It is a safety, reliability, and reputation risk. An AI system making real-time dispatch decisions on stale or corrupted data is not a productivity tool. It is a liability.
The Glue: Domain Here is where I see most AI initiatives, in energy and beyond, underperform. Organizations bring all the elements of the AI jigsaw puzzle and generate outputs, not outcomes. Then the operations teams ignore them. Why? Because the model doesn’t speak the language of the domain. Domain expertise is what transforms a statistically valid model into an operationally trusted one. It is the knowledge that the same asset exposed in different conditions and generating different outcomes is not anomalies but behavior. This knowledge does not live in data. It lives in people in the organization who spent decades reading the energy system, and witnessed the transition. The organizations that are winning with AI in this sector are the ones that have found a way to encode that institutional knowledge into the design of their models, their training data, their feature selection, and so on. Domain is not a soft skill. It is the mechanism by which AI outputs become decisions that operators will act on. Without it, the most sophisticated model in the world sits unused - a monument to misaligned investment.
The Formula That Actually Works I have learned various formulas and my favorite was that the equation of decarbonization is more electrification and more digitalization. Now the new formula which excites me is that the compounding operational advantage over the competition is a multiplication of data foundation, domain expertise, and AI capabilities. Weak data makes domain knowledge harder to encode and AI models less reliable. Missing domain expertise means no one knows which data matters or how to validate model outputs. And AI without the first two is just pattern-matching on noise. The energy transition is generating data at a scale and variety the sector has never encountered. Every renewable resource, new mobility option, storage action, and shift in energy demand is a signal. The organizations that build the data infrastructure to capture and integrate these signals - and the domain-informed AI to reason over them - will compound their advantage with every passing year. Those that treat AI as a technology purchase rather than a strategic capability built on data and domain will find themselves making slower decisions with less confidence at precisely the moment the energy system demands the opposite.
Data is the foundation. Domain is the glue. AI is what you build when both are in place.
