An improved model for reducing the cost of long-term monitoring in Clean Development Mechanism (CDM) lighting retrofit projects is proposed. Cost-effective longitudinal sampling designs use the minimum numbers of meters required to report yearly savings at the 90% confidence and 10% relative precision level for duration of the project (up to 10 years) as stipulated by the CDM. Improvements to the existing model include a new non-linear Compact Fluorescent Lamp population decay model based on the Polish Efficient Lighting Project, and a cumulative sampling function modified to weight samples exponentially by recency. An economic model altering the cost function to a net present value calculation is also incorporated. The search space for such sampling models is investigated and found to be discontinuous and stepped, requiring a heuristic for optimisation; in this case the Genetic Algorithm was used. Assuming an exponential smoothing rate of 0.25, an inflation rate of 6.44%, and an interest rate of 10%, results show that sampling should be more evenly distributed over the study duration than is currently considered optimal, and that the proposed improvements in model accuracy increase monitoring costs by 21.4% in present value terms. (C) 2014 Elsevier Ltd. All rights reserved.