Cross-correlations between Cosmic Microwave Background (CMB) temperature and y-spectral distortions anisotropies have been previously proposed as a way to measure the local bispectrum parameter f loc.NL in a range of scales inaccessible to either CMB (T , E) bispectra or T -µ correlations. This is useful e.g. to test scale dependence of primordial non-Gaussianity. Unfortunately, the primordial y-T signal is strongly contaminated by the late-time correlation between the Integrated Sachs Wolfe and Sunyaev-Zel'dovich (SZ) effects. Moreover, SZ itself generates a large noise contribution in the y-parameter map. We consider two original ways to address these issues. In order to remove the bias due to the SZ-CMB temperature coupling, while also adding new signal, we include in the analysis the crosscorrelation between y-distortions and CMB polarization. In order to reduce the noise, we propose to clean the y-map by subtracting a SZ template, reconstructed via cross-correlation with external tracers (CMB and galaxy-lensing signals). We combine this SZ template subtraction with the previously adopted solution of directly masking detected clusters. Our final forecasts show that, using y-distortions, a PRISM-like survey can achieve 1σ(f loc. NL ) = 300, while an ideal experiment will achieve 1σ(f loc. NL ) = 130, with improvements of a factor ∼ 3 from adding the y-E signal, and a further 20 − 30% from template cleaning. These forecasts are much worse than current f loc.NL boundaries from Planck, but we stress again that they refer to completely different scales.