Constructing behavioral profiles from consumer online browsing activities is challenging: first, individual consumer-level records are massive and call for scalable high performance processing algorithms; second, advertising networks only observe consumer’s browsing activities on the sites participating in the network, potentially missing site categories not covered by the network. The latter issue can lead to a biased view of the consumer’s profile and to suboptimal advertising targeting. We present a method that augments individual-level ad network data with anonymized third-party data to improve consumer profile recovery and correct for potential biases. The approach is scalable and easily parallelized, improving almost linearly in the number of CPUs. Using economic simulation, we illustrate the potential gains the proposed model may offer to a firm when used in individual-level targeting of display ads.