David Springate is senior data scientist in PetSmart’s Advanced Analytics Group. PetSmart is the largest specialty pet retailer in the US and is headquartered in Phoenix, Arizona. David’s focus is on understanding and predicting customer behavior using a range of classical statistical and machine learning methodologies on large longitudinal datasets. Prior to PetSmart, he was a principal data scientist at Teradata, a data scientist in the market research technology sector and a statistician at the University of Manchester Medical School. He holds a PhD in Evolutionary and Computational Biology from the University of Manchester in the UK.
It is increasingly important for retailers to understand the journeys of their individual customers, in order to direct effective personalized marketing campaigns and deliver value to their customers. Large retailers, like PetSmart, store huge amounts of transactional, marketing, loyalty, web and online traffic data. The combination of each of these events for a given customer can reveal their journey with the company. The customer journey is the best predictor we have of future behavior at the individual customer level. At PetSmart we are taking a novel approach where we treat the interpretation of the customer journey as an NLP problem, with each customer touchpoint, transaction and interaction forming a symbol in the journey, like a word in a sentence. This allows us to understand the 'language' of the customer journey and use this to predict critical customer metrics such as events to focus marketing on, churn and lifecycle stage. We will share our approach to deal with training and productionizing these models on the huge amounts of data needed using a stack based on Databricks, Pyspark and Keras, with MLflow for model management, Horovod for distributed training and Hyperopt for hyperparameter optimization.