Towards a marketing attribution model combining Markov chains and regression models with machine learning

 
DATATECH BLACK@0.5x.png

SPEAKER

Gavin C Bowick

AFFILIATION

Digital Marketing Executive - Paid Performance - Crerar Hotels Group Ltd.


Abstract: Causal inference remains a challenge for marketers who want to know whether a particular campaign was responsible for driving sales or enquiries. Since the advent of online advertising, marketers have much more information about how their adverts were consumed and the results of their activity. Modern online tracking allows advertisers to see exactly how many times their ad was viewed and clicked on, and the user's activity following that interaction. As marketing budgets continue to move from traditional media to digital channels, it is becoming increasingly important that campaign effectiveness is measured effectively. This measurement is called attribution modelling and, even with the huge amount of data available, it presents significant challenges. For example, the most common online analytics platform defaults to a model where all the value is assigned to the last non-direct click, but with long, complex customer journeys, is this correct.

More recently, we have used the ChannelAttribution package in R to model the contributions of various online marketing channels probabilistically using Markov chains. However, this model does not consider the effects of offline 'traditional' campaigns such as print or radio. It also cannot determine the most effective use of budget, for example in cases where paid ads could have been replaced with an organic search click. We have been using regression methods to model the effects of offline activities, and have performed small, randomised experiments to attempt to better estimate the effects of various marketing activities. We are now working on integrating the outputs of these models using a machine learning model. Work is currently at an early stage, but the component element models are generating useable insights that are informing our campaign decision making.


Bio: Gavin Bowick has a background in research science in the field of emerging infectious diseases. His introduction to data science was building machine learning models to predict haemorrhagic fever virus disease severity. Following a transition to the commercial sector and several years' business experience - predominantly in marketing, purchasing and supply chain management - he now works in the hospitality industry, applying data science to improve return on investment on marketing campaigns and develop statistical models to assist in pricing and hotel occupancy forecasting.


PosterSteven ScottPoster