Galaxy-galaxy (g-g) lensing represents an ideal technique to constrain the dark matter distribution on galaxy scales. The required accuracy in the signal can be achieved only by stacking many foreground galaxies and averaging the ellipticity of the resulting background galaxies. Unfortunately, the stacking procedure complicates any astrophysical interpretation. In order to extract information from the composite g-g lensing signal, a reliable model of the way galaxies populate dark matter haloes is required. We use a realistic description of the halo occupation statistics based on the conditional luminosity function. It provides a statistical prescription for the number of galaxies with a given luminosity living in dark matter haloes of a given mass. Being ``a priori'' constrained by the luminosity dependence of the galaxy clustering, it can be used for predicting the g-g lensing signal without any additional tuning. Our model allows a thorough understanding of the different terms contributing to the signal. We carefully explore the effect of the assumptions entering the model. Our theoretical predictions are in very good agreement with SDSS data. Furthermore, we use the sensitivity of this technique to the underlying cosmological model to  study the feasibility of a joint analysis of galaxy clustering and g-g lensing as a novel technique to constrain the values of cosmological parameters such as $\Omega_{\rm m}$ and $\sigma_{8}$. We show that uncertainties in the model do not significantly affect the model predictions. We conclude that a combined analysis of galaxy clustering and g-g lensing can be used as a novel, complementary and competitive technique to constrain cosmological parameters.