It’s Tuesday morning. Our monitoring system flags a developing temperature anomaly on a critical bearing. It tells us what, but not much about why. We spend the next six hours digging through design documents, old reports, and service logs, only to eventually find a buried note about a known sensor calibration issue.

We lost hours chasing a ghost that was already known and documented.

That isn't some rare “one-off”; it's a daily reality for many asset managers. It’s “the 80% problem”: our digital tools are focused on 20% of our data, while the answers we need to realize value from those tools are buried in the other 80%.

The real cost of wasted time

If you have a technical background, you expect your job to be about solving tough engineering problems. The reality is you spend a shocking amount of time just trying to find stuff.

And it’s not just a feeling. A McKinsey report found that employees spend about 1.8 hours every day just searching for information. Other data from IDC suggests this can be closer to 2.5 hours, or nearly 30% of the workday.

Think about that from a C-level perspective. It’s like hiring five highly skilled engineers, but one of them is permanently stuck in the archives. In a crunch, this leads to bad habits; we find a couple of data points that confirm our gut feeling and move on, completely missing the contradictory note on page 75 of some forgotten report that would have pointed to a safer or more profitable fix.

Smart systems, dumb context

It’s a well-worn stat that around 80% of a company’s data is unstructured. The messy stuff in reports, contracts, emails, and service logs. But in renewables, our automation (for anomaly detection, performance optimization, etc.) is built almost exclusively to leverage the neat and tidy 20% of structured sensor data.

This creates a huge disconnect. An anomaly detection system is great at flagging that something has changed, but it often lacks the context to determine what that change actually means or how to best address it. Is a critical failure coming? Is it a harmless quirk? A good sign from a recent upgrade? A known issue?

To get that answer, asset managers turn to their documents, emails, and notes, hunting for “I’ll-know-it-when-I-see-it” clues in a data room compiled over years of operation and across several asset managers. The result? Instead of acting decisively, our best people waste their time debunking alerts. The very value these expensive systems promised is compromised.

A market with a hole in the middle

The digital tool market for renewables is booming, but it’s all at the extremes.

At one end, you have powerful Asset Performance Management (APM) tools from companies like Bazefield, BaxEnergy, and Green Power Monitor, all focused on that 20% of sensor data.

At the other, you have slick energy trading and forecasting platforms from Node Energy, Solcast, and Vaisala that manage market risk.

What’s left out? The big(ger) picture tasks. The tools that enable the asset managers responsible for the overall performance of the asset across commercial, environmental, and regulatory expectations and requirements. Manually piecing everything together using the 80% of data no one built a tool for.

This is where It gets interesting (and YES, AI shows up)

For a long time, this inefficiency was just the cost of doing business. Not anymore. With falling CAPEX for renewables, OPEX is now center stage. Add in aging assets, complex hybrid assets, and the end of subsidies, and the pressure on asset managers has skyrocketed. We can't afford this operational drag.

This is where Large Language Models (LLMs) enter the conversation. They can sift through our document “graveyards” and give us answers in seconds.

But let's be realistic: pointing an AI at a messy SharePoint drive isn't a solution. Just as large language models excel at digesting inhuman amounts of textual information, they also excel at being confidently wrong. So-called "hallucinations" are a documented feature (simulating some level of creativity), not a bug. In our own testing, we’ve also seen that LLMs are eager to please: if the information being asked for is not present in the source, the likelihood of the model still returning something (that looks completely believable) increases significantly.

Studies show that even top-tier models can invent facts 3-5% of the time, while others can be wrong in over a quarter of their responses. OpenAI and Stanford testing noted low single-digit hallucinations in highly constrained QA tasks but much higher for open-ended reasoning. For industry, even a 3% error rate isn't an inconvenience; it's a potential safety incident or a multi-million dollar mistake.

A recent KPMG survey found that 68% of executives are worried about the accuracy and bias of AI models. Another report from Salesforce shows nearly 60% of workers are concerned about generative AI's inaccuracy. The trust just isn't there yet.

The fix is not AI; it is building systems that make AI a reliable tool.

We need a trustworthy "co-pilot" (sigh) for asset managers that proactively finds information and creates insights, all while keeping the lineage back to the source.

The technology is finally here. The challenge is no longer whether we can solve the 80% problem, but how we build the tools and processes to do it without introducing undue risk. The future of efficient and profitable renewable energy operations depends on it.