Measurement and Verification (M&V) is often required for energy efficiency or demand side management projects in buildings, to demonstrate that savings were in fact achieved. For projects where sampling has to be done, these costs can be the most significant driver of the overall M&V project cost, especially in multi-year (longitudinal) projects. This study presents a method for calculating efficient combined metering and survey sample designs for longitudinal M&V of retrofit projects. In this paper, a building lighting retrofit case study is considered. A Dynamic Linear Model (DLM) with Bayesian forecasting is used. The Bayesian component of the model determines the sample size-weighted uncertainty bounds on multi-year metering studies, with results from previous years incorporated into the overall calculation to reduce forecast uncertainty. The DLM is compared to previous meter sampling methods, and an investigation into the robustness of efficient sampling plans is also conducted. The Mellin Transform Moment Calculation method is then used to combine the DLM with a Dynamic Generalised Linear Model describing the uncertainty in survey results for the longitudinal monitoring of lamp population decay. A genetic algorithm is employed to optimise the combined sampling design. Besides the reliable uncertainty quantification features of the method, results show a reduction in sampling costs of 40% for simple random sampling, and approximately 26.6% for stratified sampling, as compared to realistic benchmark methods.