ML Strategies for Marketing Budget Optimization

Developing a marketing budget is crucial for your business. It ensures you allocate the right amount of money to marketing activities. Indeed, it's easy to underspend, in which case you're losing potential sales or overspending, in which case you're losing capital that could be deployed in other areas of your business. Imagine a perfect world where you would spend the optimal amount on marketing activities:

This article explores an approach undertaken for a client looking to optimize their online and offline marketing budget. The client faced a common problem: overspending could lead to diminishing returns while underspending might miss potential revenue opportunities.

Data Gathering for Marketing Budget

Our initial step was to create a comprehensive dataset capturing various facets of the client's marketing efforts and external market factors. We compiled monthly data for the past three years, including spend on various channels like Google Ads, Facebook Ads, and more. Additionally, we considered internal metrics such as newsletter subscriptions and external variables, including search volumes for specific keywords, inflation rates, and unemployment figures. To complete this dataset we also included the target variable: sales.

Machine Learning Approach

To understand the intricate relationship between marketing spend and sales, we employed a range of machine learning models, experimenting with different lag periods and ensuring robustness through strategies like avoiding overfitting and using holdout sets. The solution came with a simple yet effective random forest regressor, which showed impressive accuracy in predicting sales.

Key Insights and Features

Interestingly, the analysis revealed that search volumes for certain keywords and Google Ads spending were the most influential factors. A critical discovery was the interaction between Google Ads spend and specific keyword searches. This interaction term ended up being the most significant feature, providing pivotal insights into when to increase Google Ads spending in response to rising demand for certain keywords.

Programmatic Solutions and Sales Forecasting

Building on these insights, we proposed as a future step a programmatic solution that will automatically adjusts Google Ads spending in response to fluctuating keyword search volumes. This approach will not only help optimize marketing budgets in a timely manner, but also help save precious time for the marketing team.

Technical Infrastructure

For this project, we utilized various technical tools. Data was sourced using APIs, web scraping techniques, and extracting customer data from a PostgreSQL database. We utilized Fivetran to funnel this data into a data warehouse, enabling us to work efficiently with structured data at scale. The machine learning models were developed using Python and popular libraries like Statsmodels, XGBoost, and Scikit-learn.

Conclusion

The project delivered substantial benefits to the client. By leveraging machine learning to analyze and predict the relationship between marketing efforts and sales, we empowered the client to make data-driven decisions. This optimized their marketing spend, improved budget efficiency, and enhanced sales forecasting capabilities. The success of this project demonstrate the transformative potential of machine learning in marketing strategy and budget optimization.