.jpg)
Asset and investment managers have gotten good at running diligence on assets. Evaluating an AI platform for renewable energy asset performance and portfolio management is newer ground, though. A demo is designed to show you what a product does. Diligence asks the next question: what sits underneath the platform, and does it still work on a portfolio with more OEMs, more gaps and messier tagging than a vendor's sample data?
What separates platforms is the data model underneath, the benchmark they use and whether the output holds up under investment committee questioning. I run demos for a living, so I know which questions define a serious evaluation.
The five questions in short
When owners talk about improving performance, the conversation almost always starts with production.
- What data model sits underneath the platform, and can it absorb a messy, expanding multi-OEM portfolio?
- What does it benchmark your assets against, and how much of that reference dataset is the vendor's own rather than licensed?
- Can it attribute variance to specific loss categories, or does it only report the variance?
- Does it cover your whole portfolio, including storage and sites where you have no SCADA access?
- Can the output reach the people who make the decisions, and feed the AI assistant your firm already uses?
1. Start with the data model, not the dashboard
Any renewable asset management software vendor can build a dashboard cheaply. The expensive part is the layer underneath: ingesting SCADA from multiple OEMs, multiple vintages and multiple secondary systems, then standardizing it so that a turbine in Texas and a turbine in Galicia are measured the same way.
The IEA Wind TCP's recommended practice on wind farm reliability data warned back in 2017 that the absence of common data standards was holding back industry progress on reliability. It also named the problem any fund team will recognize from an acquisition: when an asset is transferred to a new owner, they either convert all the historical data into their own structures and rerun the analysis or abandon it and start from scratch.
Ask to see the ingestion pipeline and not the front end. Ask what happens to a site with inconsistent event tagging. Ask for the audit trail from raw signal to reported number, and whether an analyst can follow it without calling support.
2. Question what the benchmark actually is
A platform that compares your assets to the rest of your own portfolio can tell you which of your sites are worst, but not whether the worst of them is genuinely underperforming or simply normal for that module type in that climate.
That distinction decides where capital goes. Prioritizing against your own average sends money to the wrong sites, which is why utility-scale renewables performance optimization depends on knowing what the rest of the market actually achieves. In storage, a degradation curve means little until you can set it against what comparable chemistries doing comparable work actually deliver.
This is also where standardization becomes more than just a technical detail.
The same IEA Wind work found that the industry's own reliability surveys are not compatible with each other, which makes it nearly impossible to derive a general reliability figure even at a high level of aggregation. That is the problem any benchmark has to solve before it can tell you anything useful, and a reference set built from data that was never standardized produces comparisons that look precise and are not.
Ask for the size of the reference dataset, the technology split, the geographic split and how current it is. Then ask the harder question: how much of that benchmark is the vendor's own operational data, and how much is licensed from a third party?
A licensed dataset can be excellent in its own right and still produce invalid comparisons, because it was measured to definitions that were never applied to your assets. Licenses also lapse and get repriced, where a benchmark built from an operating fleet does not.
3. Test whether it explains variance or only reports it
The first number often noticed is underperformance against forecasts, and it is almost never one single dramatic failure. Gaps accumulate undetected from availability shortfalls, curtailment, degradation, resource variance and contractual definitions that do not match what the meter recorded.
Which means a platform that tells you production came in under budget has told you something you already knew. The board already knows; the question they will ask is why, and a number on its own cannot answer it.
The test is whether the platform can split that variance into attributed loss categories you can act on and defend in a boardroom.
Ask the vendor to run a loss waterfall on a sample of your own data during the evaluation.
The same test applies to renewables production forecasting and risk analytics: a reforecast is only credible if the platform can show which part of the operating history moved the number.
4. Check that it covers all of the assets in one portfolio view
Portfolio reporting usually gets built around whichever assets were easiest to get data from.
Most owners see the assets they operate themselves clearly, with a direct SCADA feed. Minority stakes, co-investments, third-party managed sites and anything added since the reporting was set up usually arrive as a monthly report instead, and those are the ones most likely to surprise you.
Every asset added after the reporting was built has to be brought into it, and a lot is being added. Of the 86 GW of new US utility-scale capacity that was planned for 2026, the EIA expected solar and battery storage to account for 79% between them, with wind at 14%. Storage alone has grown at an average of 70% a year over the last three years there. In Europe, Ember found operating battery capacity passed 10 GW in 2025, more than double the 4 GW from two years earlier, with a pipeline around 40 GW. Portfolios built around wind and solar are taking on assets with different contracts, different failure modes and different data, usually faster than the reporting can catch up.
So ask two things. Does wind and solar asset health monitoring run on the same data model as the vendor's battery storage analytics platform, or on separate products that generate separate reports? Separate reports put your team back to reconciling by hand.
Then the harder one:
What can the platform tell you about an asset where you have no SCADA connection at all?
In renewable infrastructure fund portfolio management those assets can be a significant share of the total, and a platform that only reports on direct feeds will leave them unmonitored.
5. Ask how the answers reach people who never log in
The people who make the decisions rarely open the platform. The investment committee or board, the CFO and the investor relations team read a memo. If the intelligence stops at a dashboard that your asset manager has to translate every month or quarter, you have bought a reporting tool that requires you to take extra steps.
Ask two things:
Can the platform produce board-ready output directly, with the loss attribution and benchmark context already in it? And if your firm already has or will roll out an AI assistant, can it feed that?
Generic models are confident but wrong about production losses, availability and OEM contract terms, because they have no access to your asset data and no renewables context to ground them in.
How Clir answers this checklist
Answering the five questions in order.
The data model. Clir standardizes SCADA, financial and contractual data across OEMs, vintages and technologies into a single model built for renewables rather than adapted from general-purpose business intelligence software. Every reported number traces back through the loss attribution to the raw signal behind it, so an analyst can follow the chain without calling support.
The benchmark. Clir's operational data includes 350+ GW of renewable energy intelligence, including 150+ GW of wind and 100+ GW of solar, plus more than 100 BESS assets, split roughly 40% North America, 40% Europe and 20% rest of the world, drawn from every major OEM in the market. It is our own operational data rather than a licensed third-party feed.
Variance. More than 50 detectors and machine learning models run against Clir's 350+ GW industry intelligence dataset, and the loss waterfall attributes variance to categories you can act on and defend. One owner found a power boost upgrade that had stopped working across several turbines, worth 0.2% of annual energy production once resolved, alongside a serial defect their board report had not mentioned. Another lifted portfolio availability by 2% through faster identification and resolution of underperforming turbines.
Coverage. Wind, solar and BESS run on one data model rather than separate products bolted together, so a mixed portfolio produces one report instead of three. And Clir's Data Agent extracts structured data from monthly, board and service provider reports, so assets with no SCADA connection appear in that same view alongside the ones you have a direct feed from.
Distribution. Clir produces board-ready reporting directly, which took one portfolio manager's report cycle from weeks to hours, and Clir's MCP server lets your team query live portfolio data from the AI assistant they already use.
Software diligence deserves the same rigor as asset diligence. Ask to see the platform with your own data in the evaluation, ask what sits underneath the dashboard and ask what the benchmark is made of. The answers separate the market quickly.
Tell us how your portfolio is put together and where it's giving you issues; we'll show you what Clir’s AI platform does with each one.
.png)
