Guilherme Studart is the co-founder of Delfos Energy, an AI Asset Performance Management solution for Wind, Solar and Storage.

getty

In 2025, our team monitored a wind fleet for over four months. During this time, there was no drama, failure of any turbine or overdue inspection. The operator simply wanted better visibility into what was coming.​

Over the first three months, AI models learned the normal operating behavior of every monitored component: generator, gearbox, transformer, pitch, yaw and anemometer. By month three, 28 predictive alarms had fired across the fleet. By month four, those 28 alerts had been distilled into four clear maintenance priorities. The team ordered parts, scheduled maintenance and planned the necessary interventions.​

By month five, every issue had been resolved before it became a major failure. There were no major component replacements, emergency mobilizations or unplanned downtime.​ The result was nearly 10,000 MWh of annual energy production recovered, around $875,000 in annual value protected and 2.3% of annual output that would otherwise have disappeared.

What stands out here is that there wasn’t one single event that would have caused these losses, but cumulative events that were invisible if you looked only at the data reported on the OEM tools. ​​

The Question Most Investors Aren't Asking

Europe now has 285 GW of installed wind power capacity, with 187 GW more expected to come online between 2025 and 2030, according to WindEurope's 2024 statistics report. That is an enormous amount of capital being deployed into assets whose long-term returns depend heavily on operational performance.​ Yet the way most of that performance is evaluated has not kept pace with the scale of deployment​.

O&M costs represent 20-30% of the total lifetime cost of a wind project, making them the largest controllable variable in the operational phase by a significant margin. And within that O&M budget, unscheduled maintenance (the reactive kind triggered by failures rather than foresight) is consistently the most expensive and most variable component. The same research shows that predictive maintenance strategies have demonstrated reductions of around 25% in unplanned downtime across wind assets.​

​Most investment frameworks still evaluate wind assets primarily on capacity factor, contractual availability and grid connection reliability. These are lagging indicators, and by the time they move, the value has already been lost.

​Capacity factor is typically measured monthly or quarterly, and availability only registers a problem once a turbine has actually stopped producing. A bearing degrading over 12 weeks won't move either metric until the failure has happened and the downtime is logged, by which point a scheduled, low-cost fix has become an unplanned, high-cost one.​​

The more important question, and one I encounter far too rarely in due diligence conversations, is a straightforward one: what is the O&M intelligence model behind this portfolio?​

What The Data Actually Costs When You Can't See It

​The annual value protected across the fleet in my introductory example is useful not because it represents an extraordinary outcome, but because it illustrates how operational intelligence changes maintenance decisions.

Rather than waiting for failures to surface through conventional operational metrics, the AI models detected deviations from each component's expected behavior early enough to investigate them. Over four months, 28 predictive alerts were prioritized into four maintenance actions, allowing the operator to focus resources where they were most likely to matter.

The value did not come from predicting the future with certainty. It came from distinguishing meaningful signals from routine operational variation, giving maintenance teams enough lead time to plan interventions instead of reacting to failures after they occurred.

This is what operational intelligence often looks like in practice. It is not another dashboard or a stream of SCADA alarms. It is continuously modeling component behavior, identifying changes that warrant attention and supporting better maintenance decisions before problems become significantly more expensive to address.

Choosing not to monitor does not eliminate operational risk; it simply means more of that risk remains hidden until it appears as unplanned downtime, emergency maintenance or lower-than-expected energy production.​

The Due Diligence Question That Changes The Conversation​

When evaluating a wind portfolio today, investors and asset managers typically bring rigorous frameworks to bear on resource quality, grid risk, counterparty exposure and regulatory environment. Those frameworks are necessary, but they are not sufficient.​​​

The underlying operating model matters more than any single deployment, and there are four things I believe define a mature one:

1. Data Integration: Does it unify SCADA, CMS and inspection records into one component-level history or leave each in its own silo?

2. AI Governance: Who validates alarm accuracy and retrains the model as the fleet ages?

3. Workflow Integration: Does an alert actually reach a work order or just a dashboard?

4. Organizational Readiness: Will teams act on early warnings or override them to keep a reactive maintenance calendar?​

The operational phase of a wind asset has increasingly become 30 years or more. The performance gap between a reactively managed fleet and a predictively managed one compounds across every year of that period, and at the portfolio level, that gap is not a rounding error.

The question worth adding to every evaluation is simple: does this O&M model tell you what is going to fail or only what already has?​​​​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?