<aside>

Summary

This project aimed to use ML modelling to solve a cost optimisation problem in the area of member acquisition direct mail marketing at the Brooklyn Museum.

</aside>

<aside>

Methods, Skills, Tools

• Data Cleaning and Wrangling.
• Machine Learning.
• Data Storytelling and communication to stakeholders. </aside>

<aside>

Result

The model successfully demonstrated that a data driven approach to direct mail marketing could save the museum significant sums. Several strategies were eventually presented.

</aside>

Direct-mail marketing comes at significant expense: each envelope costs about $1.50 to print, stuff, and post, yet historical acquisition rates are below 1%. With mailing lists routinely topping 40,000 names on two campaigns per year that equates to roughly $120k per anum.

Leadership therefore needed a data-driven way to decide who actually merits a piece of mail using behavioural breadcrumbs found in CRM data.

To achieve this I built a machine-learning pipeline that re-aligned marketing economics with patron propensity. In order to deal appropriately with class imbalance, I used a calibrated XGBoost model, which would output a hierarchical ranking of ‘propensity to convert’. Using this I could devise potential cost saving strategies.

Key questions:

1. "Is our current data which pertains to constituents in the mailing list prospect pools predictive of membership conversion?”
2. “What were the attributes of constituents that would explain these predictions?”

The data in question pertained to two previous campaigns for which the museum had reliable data, having recently undergone data migration to its first comprehensive CRM.

I initially used thousands of Monte Carlo simulated correlations to analyze which features had signal. The distribution of the correlations are reflected in the charts below.