The situation
A large plant knew its energy bill was too high and did not know which part of it was recoverable. Thermal and electrical systems had been extended over years, each change locally sensible.
The constraint
An audit that names twenty inefficiencies and ranks none of them does not get implemented. The plant could not stop producing, so every proposed intervention had to survive the question of what it would do to output while it was being installed.
What I decided
I ranked interventions by what could actually be executed against a production schedule, not by theoretical saving. Where a measure looked attractive but depended on an assumption I could not verify from the available metering, I said so rather than folding it into the headline number.
What shipped
An audit of the thermal and electrical systems with a roadmap of interventions, sequenced so each one could be committed to independently.
What it produced
Implemented measures cutting annual energy cost by roughly 15%, with a matching reduction in the site's carbon footprint.