Using multi-armed bandit experimentation to optimise multichannel digital marketing campaigns

Abstract

Multichannel campaign optimisation using a machine-learning technique called multi-armed bandit experimentation is a powerful, though nascent, alternative to traditional campaign attribution approaches for maximising return on marketing investment. The technique works by treating the various attributes of a digital marketing campaign as combinatorial treatments in an ongoing controlled experiment and continuously optimising delivery towards the combinations that deliver the best results. Performing true multichannel optimisation requires significant investment in experiment design and maturity in data, marketing automation technology and organisational alignment. However, despite these challenges, organisations can start to move towards a full optimisationdriven approach for their digital marketing by identifying campaigns within their existing channels which could benefit from this technique and using those campaigns to establish the appropriate processes and technical capabilities, before scaling out efforts more broadly across multiple channels. This paper provides an overview of this new campaign optimisation approach, including some existing in-market solutions that use it, and examines some of the factors and prerequisites that organisations will need to consider in implementing such a technique.