Optimizing Marketing Efforts by Identifying Proxy Metrics with Machine Learning

Company X has a strong history in the marketing space but continues to operate on several long-held assumptions, one being on how to optimize while a campaign is in-flight. Several business partners of Company X base campaign performance on incremental return on ad spend (iROAS), but not all metrics in the calculation are available for in-flight optimizations, namely sales lift. Instead, Company X has assumed that return on ad spend (ROAS) is the best metric for optimizing decisions in place of the missing key performance indicator (KPI). This project addressed the optimization metric assumption utilizing data science techniques. First, implementing correlation analysis revealed relationships between sales lift and metrics available for in-flight optimization decisions. Second, developing and evaluating various machine learning models to predict sales lift explored proxy optimization metrics. Random forest was the best predictive model based on MAE and analyzed for the importance of features. Percentage of new buyers (13 weeks and 13 months) variables were found to be critical for predicting sales lift but had an unexpected negative correlation, creating implications for reporting and strategy. Correlation analysis and machine learning models validated Company X’s assumption that ROAS is the best proxy metric for optimizations of campaigns with a goal of sales lift or iROAS. This confirmation provides shareable results to build trust with business partners further. This project opens the door for future research to improve marketing practices by applying data science.