Estimating incrementality – whether an ad makes a difference for an action – is central to understanding the efficiency of online advertising, in particular in the presence of sophisticated ad servers. We have developed sophisticated statistical tools to perform experiments based on a geographical decomposition of advertising markets, which allow us to estimate incrementality even for off-line actions (store visits, purchases). In this session Christoph will discuss the background of incrementality vs. attribution, how geo experiments work and how they are set up, in particular the determination of sufficiently matched market regions, and the regression scheme used to evaluate their outcome.