In conditions of poor soil fertility and increasing importance of global value chain, agricultural extension projects have been one of the main channel to increase farmer's production and income. In this literature, prices received by farmers for agricultural goods are usually assumed to be homogeneous. We dispute this over-simplification: prices and production levels in developing countries are often jointly determined. The analysis relies on a successful extension program in the Peruvian highlands, where the main income source is the dairy sector characterized by a highly segmented market. We propose a simple theoretical model to explore how the discontinuity in price induces non-linear return to investment and diverging incentives. The econometric analysis confirms the model's propositions: producers that were not included in the formal market at baseline, but close to it, have more intensively innovated. This investment leads to a higher price increase than other producers. The effects are shown to resist to falsification tests, mechanisms are discussed and positive externalities are found within communities. Hence we show that innovation in the context of a segmented market leads to heterogeneous impacts and non-trivial income effects. Contrarily to the expected disequalizing effects of innovation adoption, it induces scope for unexpected social mobility. Acknowledgement : We thank Jean-Philippe Platteau, Catherine Guirkinger, Jean-Marie Baland, Franc{c}ois Bourguignon, Michael Grimm, Jo Swinnen, James Fenske and Douglas Gollin for very helpful comments and suggestions