Improved methods for quantifying energy savings in buildings need to be supported by empirical measures rather than modeled estimates of future annual energy demand. This paper uses power temperature gradient (PTG, W/K), or the slope of power demand in response to changes in external air temperature; first, to categorise dwelling energy performance from daily energy data (when 0-15 °C outside); second, to investigate variations in 24-h profiles of delivered power. Estimates of PTG were obtained from 567 UK dwellings with 118,000 days of gas and electricity data. From a multivariable regression model, PTG was predicted by dwelling characteristics (number of bedrooms, number of floors, dwelling type, and dwelling age category (all p < 0.001)) but not by number of occupants. When dwellings were grouped into quintiles of PTG, mean PTG had threefold increase from the first to fifth quintile (188 to 563 W/K, respectively). This was reflected in 24-h profiles of delivered power (30 min intervals): at 0°C, each 100 W/K decline in PTG corresponded to ∼2.5 kW decline in mean morning and evening peak power. Using PTG to estimate reductions in peak power as equivalent 'negawatts' reframes potential benefits of energy efficiency retrofits and for grid resilience.