In order to achieve no net loss of biodiversity, the mitigation hierarchy (avoid, reduce and offset) must bring gains at least equivalent to the losses associated with development projects. This can be achieved through ecological restoration actions that aim to accelerate the recovery of ecosystems. In order to size offsetting measures, an estimation of the gains must be done prior to the realization of the projects. This is a crucial step which remains very uncertain, in particular due to the lack of feedback on the success of ecological restoration actions. In this article, we suggest ways to improve this estimation. For this, we compare estimations of gains provided by restoration measures on the island of Kembs (Alsace, France), using an equivalence evaluation method (ECOVAL), and monitoring data collected just after restoration actions. Several analyses are carried out and concern (i) the accuracy of the estimations and the link with the degree of uncertainty attributed to the estimations of the gains (low, medium or high) and (ii) the sources of the differences observed between the estimated values and the observed values. Overall, half of the estimations were correct in the short term, but a majority is due to the short time frame to observe a change in the indicators. Good estimations are found for all degrees of uncertainty and almost all taxa and natural habitats assessed. Errors in the estimations are mainly due to contingencies during the works and unforeseen natural changes in the environment. In addition, the greater the uncertainty, the greater the differences between the estimated value and the outcome of the monitoring. Our results underline the importance of setting up regular monitoring over several years (and especially in the long term) to verify that biodiversity is changing as estimated and to rectify offsets measures if necessary with adaptive management. The main lever for improving the estimation resides in the capitalization of feedback, paying particular attention to the homogenization between the monitoring data and those of the initial state.