Data Procurement and Propensity Modeling for Direct Mail Marketing

A regional home energy company relied on direct mail marketing for customer acquisition, which comprised a significant portion of the total cost per sale. The campaigns were generating a lot of business, but management knew there was room for improvement. Extrasensory¹ was engaged to overhaul and optimize the direct mail process from end to end, reducing unnecessary mailings, streamlining campaign design, and modernizing marketing technology.

Motivation

The client was successfully acquiring customers through direct mail channels, but the marketing function had grown rapidly and mailings were poorly controlled. The prospect list had grown by combining lists from different sources over the years and contained a large and unknown number of non-trivial duplicates. Individual records were missing attributes necessary for campaign qualification, resulting in damage to brand reputation and customer frustration. Campaigns themselves were launched and evaluated in an intuitive, ad-hoc manner making performance hard to quantify.

Approach

Extrasensory's first step was to understand the complex relationship between business lines, products and services, and campaign qualification: Which customers were eligible for which products and services based on the attributes of their home. This allowed us to develop a plan to fill gaps, remove duplicates, and optimize campaigns as follows:

• Establish a mailing database designed to handle ambiguity in address records.
• Disambiguate all existing prospect records against an entity model by mapping likely duplicates to a source-of-truth entity using off-the-shelf machine learning.
• Supplement the mailing database with records from municipal governments and other online sources, merging their fields into the entities.
• Train a propensity model using years of mailing and response data.
• Design a campaign management process for launch, evaluation, and tracking of mailing campaigns.

Results

Extrasensory automated the online retrieval and integration of data for 3 million properties from 200+ web sources and covering 100+ fields. This database was used as the foundation for machine learning propensity models which increased response rates by 2x, cutting marketing costs 50%. Streamlining of campaign tracking and evaluation allowed the marketing function to respond to changing conditions or fluctuations in days instead of weeks. As a bonus, the mailing database was able to serve as a business intelligence source for other functions, providing an industry-leading view of the customer in fine-grain and in aggregate.