Putting a meter on every light fixture is neither practical nor affordable. So we built — and patented, jointly with the client — algorithms that infer when each lighting load is on, turning a single circuit-level signal into verifiable, billable savings.
The client is a US energy-efficiency building retrofit company running a scalable measurement-and-verification system for small businesses. They needed a way to serve a large pool of SMEs who valued lighting retrofits but lacked the down payment to invest. Full Stack Energy developed a precise, cost-effective disaggregation system — patent-approved (US 10,734,809) and built jointly with the client — that calculates the probability of a given fixture being on or off through the day and uses machine learning to predict run-times where sub-metering isn't feasible. Lighting usage and savings are then presented in a shared-savings bill, generated on demand.
The client runs a scalable M&V system for small businesses and wanted to open it up to a large pool of SMEs who were interested in the value of a lighting retrofit but didn’t have the initial down payment to invest. The shared-savings answer — pay from the savings, nothing up front — depends on being able to bill accurately for the savings each month.
The obvious way to measure lighting savings is to sub-meter every fixture. That’s neither practical nor cost-effective across a large pool of small sites. So the measurement problem became an inference problem: how much was each lighting load actually on, given only what a circuit-level meter can see?
The client needed a precise, cost-effective system that could answer that well enough to put on a bill — and to give both customers and utilities confidence in the savings claimed in every M&V project.
The system was envisaged as a way of calculating the probability of a fixture being on or off at a point in the day, then using machine learning to predict run-times where sub-metering isn’t feasible — turning measurement into billable savings.
You don't need a meter on every light. You need to know how long each one was on — and that's a probability problem machine learning is very good at.
Lead Engineer, Full Stack Energy
Using the disaggregation algorithms, customer lighting usage and savings are calculated and presented in a bill that can be generated at any time — giving customers and utilities confidence in the savings claimed in every M&V project, and making the shared-savings model viable for SMEs who couldn’t otherwise fund a retrofit.
The machine learning leverages historical data from circuits and sites across the country, so the system doesn’t just measure — it recognises recurring occurrences, raises alerts, auto-classifies conditions and identifies trends. And because the approach was patent-approved (US 10,734,809), it’s a defensible piece of IP, not just a clever script.
The framing that made this work was treating it as probability, not measurement. We weren’t trying to recover a perfect ground truth for every fixture — we were calculating how likely each load was to be on, accurately enough to bill. That shift is what made a single circuit signal good enough to underwrite real money.
Getting to a granted patent with the client tells you this wasn’t an off-the-shelf disaggregation library. The combination of domain knowledge — what lighting loads actually do — and the maths is what made it both novel and reliable enough for a utility to stake a savings claim on.
Load disaggregation infers how long individual loads were on from a single circuit-level signal, instead of metering each fixture. Putting a meter on every light is neither practical nor affordable across many small sites, so the patented system (US 10,734,809) calculates the probability of each fixture being on through the day and turns that into verifiable, billable savings.
The approach treats it as probability, not perfect measurement — calculating how likely each load is to be on, accurately enough to put on a bill. Machine learning predicts run-times using historical data from circuits and sites across the country, which is what makes a single circuit signal good enough to underwrite real money.
The disaggregation method is patent-approved (US 10,734,809), developed jointly with the client. The combination of domain knowledge about how lighting loads behave and the underlying maths is what made it novel and reliable enough for a utility to stake a savings claim on — not an off-the-shelf library.
Aggregate data often contains the answer you need — if you can infer it. We've patented one way of doing exactly that. Let's talk.