We present a new method aiming at addressing the well-known issue of ill-conditioned correlation matrices for Euclidean time correlators relevant for lattice calculations. We propose to transform the data through a linear and invertible regulated Laplace filter, and show that it systematically reduces the correlations in the data. Since this method is based on a modification of the data rather than a direct regularisation of the correlation matrix, it introduces no additional systematic errors or modifications to the interpretation of goodness-of-fit tests. We discuss how the filter regulator can be optimised to maximise the improvement on a specific correlator and time range. We also show that further improvement can be obtained by combining the filter with a downsampling step.