Outlier detection at scale: monitoring millions of ad-campaign time-series from Juergen Dietz

With over 12k advertisers and a network of over 17k publishers, Criteo connects a lot of different dots together as part of its day to day business. Keeping tabs on glitches that can affect client performance is a big challenge. We have recently started building a near-real-time monitoring system in R and Shiny that flags potential outliers to account teams and facilitates rapid resolution. We present here our first experiences with the new tool and our roadmap for the future.

Bio: After completing a PhD in Particle Physics, Juergen moved into Data Science at a mathematical consultancy firm where he focused on developing bespoke predictive models. Through his time there he learned to really appreciate the usefulness of R and it’s range of applications. Having recently joined Criteo as a data scientist, he spends most of his day delivering insights to Criteo’s top UK clients and the rest of the time evangelising on the benefits of R.