We establish the probabilistic well-posedness of the nonlinear Schrödinger equation on the 2d sphere S 2 . The initial data are distributed according to Gaussian measures with typical regularity H s (S 2 ), for s > 0. This level of regularity goes significantly beyond existing deterministic results, in a regime where the flow map cannot be extended uniformly continuously. Cauchy problem for the cubic NLS is locally well-posed in the Sobolev space H s , s > 1 2 . This result is of interest because the classical methods, based only on Sobolev embeddings, give the well-posedness under the much stronger restriction s > 1 (one half of the dimension). In the case of the sphere S 2 the restriction s > 1 2 was relaxed to s > 1 4 in [8]. As shown in [6] the s > 1 4 restriction is in a sense optimal because this is the limit of the semi-linear well-posedness methods. The goal of this paper is to show that, in the case of the 2d sphere S 2 , one can go beyond the s > 1 4 threshold by randomizing the initial data in a Sobolev space of low regularity. This is in the spirit of the program initiated in [11], which aims to study dispersive