PPC optimisation via Thompson sampling @ London Bayesian Mixer

Jordan Palmer, a fellow data scientist at Datasparq, and I presented our recent work on helping a client optimise their bids to the Google Ads pay per click auction. The event was the Bayesian Mixer meetup, hosted by the Bayes Business School of City University, London.

We talked about how we used Thompson sampling and constrained convex optimisation to produce bids that optimally balanced uncertainty against the objective to maximise paid-for sessions, all while falling within financial (ROI) constraints.