← Notes · Talk given May 26, 2026 · Windpower Data and Digital Innovation Forum, Amsterdam
Bridging the gap to decision-ready metrics in renewables
Most owners have invested in a shared data platform. The gap that remains sits between having the data and being able to answer a management question quickly, clearly and in a way the team trusts enough to act on.
Three questions that should be easy
Asset management asks the same things every week. What is down right now, since when, and why? What did we lose last month, and what caused it? What is our availability, and which definition is that?
Stephen Dictor opened the talk with an estimate: in many portfolios the first question takes 30 to 45 minutes to answer, and at the end the team is still not sure the answer is right. These look like reporting questions. They are trust questions: will the person reading the number stake a decision on it?
Why they are hard
Down is not one state. When a site loses communications, the gap may be recoverable from a secondary source, need a backfill once communications return, or stay unexplained until someone visits the site, which turns a five-minute question into a three-day investigation. The raw telemetry looks the same in all three.
A loss is only as good as the expected output behind it. To estimate lost MWh, each asset needs a model of what it would have produced if it had been running. A modeled power curve, a neighboring turbine and a long-term P50 estimate give different answers, and the choice can move the loss by several percentage points of annual energy. If the method is implicit, the loss number is an opinion that happens to have decimal places.
Availability has several valid answers. For the same asset and month, operations quotes time-based availability, a lender may require a production-based view, and an OEM warranty applies its own contractual definition and exclusions. All three can be right and still differ by several points. This is definition drift, and it compounds with every acquisition made under different assumptions.
Missing meaning, not missing data
All three questions depend on interpretation, classification and logic, and a platform supplies none of them. Those decisions belong to the owner. When nobody makes them explicitly, every question becomes a debate instead of a lookup.
- WorkflowsThe morning review, availability reports and investor questions.
- Owner layerStatus logic and KPI rules, defined, versioned and owned by the operator.
- PlatformIngestion, storage, dashboards and APIs.
What changes when the owner builds it
The morning review. Picture a site with four assets at run state zero, all identical in the raw data. What separates them is classification, and it can be done during the 7:15 a.m. call or before it.
Without status logic
- 7:15 a.m.Four assets show run state zero
- 7:15–7:35 a.m.Which are outages, comms gaps or curtailment? Who owns each?
- 7:35 a.m.Decisions start, with the facts still partly open
With status logic
- OvernightEach event is classified as it arrives
- 7:15 a.m.Action queue ready: three with an action and an owner, one under investigation
- From 7:15 a.m.The call is spent on decisions
Availability. With agreed, versioned definitions, operations, the lender and the warranty team each get their view, and all three numbers are correct at the same time. The question becomes which number applies and why. Carve-outs are declared in advance, and a newly acquired asset with different contract terms joins the same structure rather than another spreadsheet.
Coverage. Once the logic is owned, the next question is where it is strong enough to trust today. A coverage map answers site by site: how much data arrived, which checks failed, which rules version applies and what limits are known.
| Site | Data received, last 30 days | Failed checks | Status rules | Known limits | Can support today |
|---|---|---|---|---|---|
| A | 99.8% | None | v3, approved | None | Lender and investor reports |
| B | 99.6% | None | v3, approved | Curtailment taken from SCADA flags only | Internal reporting |
| C | 96.2% | None | v3, approved | Two inverters report intermittently | Internal reporting, with a stated gap |
| D | 81.5% | Met mast disagrees with nacelle wind | v2, update pending | Intermittent comms since March | Not yet: fix the gaps first |
And AI?
The cost of building a model has fallen from months to minutes. That makes the work before it, defining the question and preparing the data, more important, because shortcuts are now easier to take and harder to spot. AI helps most when the question is well defined and the logic is sound: summarizing work orders, triaging event piles, pre-screening anomalies. Defining the question remains the owner's job.
The order matters
The talk closed with four steps, each depending on the one before.
- Access. Signals are visible and data is queryable.
- Logic. Status rules are owned and KPI definitions are versioned.
- Readiness. Coverage is visible and trust boundaries are declared.
- Acceleration. Automation and AI run on governed logic.
Accelerate before the logic is sound and you get faster wrong answers.
Renewable Asset Partners works with owners and operators to define and build the owner layer: the status logic, loss accounting and availability definitions behind the three questions.