We build a percolation model to study diffusion of technological innovations in a network of
social relationships. The phase transition of percolation is a result of imperfect information, and
strongly a ects market effciency, the demand curve and consumer welfare. Different network
topologies are implemented, such as regular lattices, small-world networks and Poisson random
networks. The most important factor for percolation is the increasing number of neighbours
of successive orders. The average path length has a moderate e ect, while clustering is unim-
portant. The most effcient structures are Poisson random graphs and 2-dimensional lattices:
random graphs present a lower percolation threshold , while 2-dimensional lattices have higher
adoption rates above the threshold. This means that random graphs are more effcient at low
levels of product quality (high levels of price), while a 2-dimensional lattice is more effcient
when the market is well in the diffusion regime. We also study endogenous learning in di usion,
allowing the quality of a product to increase (the price to decrease) with the number of adopters.
Learning shifts the critical transition threshold of percolation. This translates into an positive
e ect on demand and welfare.