We contribute to the literature on how program design affects program performance among vulnerable groups by studying the effects of varying the subsidy level and program procedures in an energy efficiency assistance program targeting low-income households in Germany. Eligible households receive, upon enrolment, a voucher to subsidize refrigerator replacement. The voucher is redeemed against cash following replacement. Observing the decisions of 77,305 eligible households, our RDD design exploits two quasi-exogenous temporal discontinuities in voucher value and program procedures. We find that a switch from automatic to elective enrolment and more rigid voucher terms reduces the number of vouchers in circulation, but raises the replacement rate among eligible households, the key performance metric, by 4 to 10 percentage points, consistent with psychological theories of goal setting and time management. A subsidy increase of 50 Euro raises replacement rates by 9 to 16 percentage points. The effect of procedural changes is equivalent to an additional 34 Euro in subsidy. Back-of-the-envelope calculations highlight that low-cost changes in procedures that target the behavioral responses of low-income households represent plausible areas of unexploited economies in program design and merit systematic investigation.