By Bert de Vink ·
The hard part of applying AI to energy infrastructure is rarely building a model. It is showing that the model changes a decision, and by how much. That is why Vink Energy starts with a bounded pilot rather than a product demo.
Vink Energy is pilot-ready: scoped for pilots with clients, without completed pilots to report yet. This article describes how we would test an idea, and what we would and would not claim from the result.
Start with the decision, not the model
A pilot begins by naming the decision it should improve, the area and the period it covers. For Grid Intelligence that might be how well local generation from rooftop solar is estimated. For Grid Build Intelligence it might be how well an LV-grid reinforcement is prepared: routes, constraints, obstacles and differences between plan and reality.
Use today's way of working as the baseline
A model is only useful if it beats what you do now, so the current process is the baseline. The pilot runs the Vink-assisted process on the same area and period, and both are compared with the baseline and, where available, with what actually happened.
What a pilot can measure
- Detection accuracy: are the assets we find in imagery really there?
- Estimation error: how far off are the estimates, for example of local generation?
- Analyst hours and planning time: does the work take less effort?
- Constraints identified: are more of the relevant constraints found before work starts?
- Field visits or rework avoided: does better preparation prevent expensive surprises?
Not every measure applies to every pilot. They are agreed per pilot, before it starts.
State the uncertainty
An estimate without an error margin is hard to use in a planning decision. Where records are incomplete, imagery, spatial and network data can fill the gaps, but each estimate should come with how uncertain it is. That is part of what a pilot checks.
Agree the data and who supplies it
We do not assume access to proprietary network data. Network, infrastructure and customer data are used only where the customer supplies them, under agreed terms. Public spatial information such as BAG, aerial imagery, Earth observation and weather can be combined with them.
End with a decision, including stopping
At the end the comparison leads to one of three outcomes: deploy, refine, or stop. Stopping is a legitimate result. A pilot that shows a model does not change the decision has still saved you a larger investment.
What we will and will not publish
We publish results only when a pilot has been completed and agreed with the client. Until then we describe the approach, as here, and we do not claim production deployments.
← All articles