SEBA Students Analyze Catalog Mailing Methodology for Popular Local Hardware Company

A student team composed of Andrea Bernard, Scott Jacobs, Kelly B. Lampman, Denise Pitney, and Justin Salvador, worked with the now Corte Madera-based company to find a way to improve the profitability of catalog mailings for the company’s new Teen Home furnishing line.

In order to make sure that they were getting the best return on their marketing efforts, the group took into consideration Restoration Hardware's current process for mailing expensive catalogs. Second, they worked on an analytics question posed in order to predict who was likely to make a purchase from the Teen Product line.

“In 2014 the company mailed out a 7-pound catalog that featured 3,300 pages and 13 sourcebooks,” said Andrea Bernard. “This created a lot of buzz for the company, both positive and negative.”  The team researched the current catalog process for the company, taking into account what created buzz and what had garnered criticism. They considered whether or not the company should bundle the catalogs together or mail them separately.

The team learned that with the launch of RH Teen in 2015, and with little data to support a specific mailing to target RH Teen buyers, the catalogs had been sent bundled with the company’s Baby & Child source books in order to reduce postage expenses. As the team gathered more data on the buyers of RH Teen, they considered whether or not to continue to bundle RH Teen with Baby & Child, or if instead there was a significant model that could be built to optimize the mailings of the two catalogs.

While developing a model, the team faced a few challenges, but overcame them with the right dedication of time and effort. In the end, they proposed the use of Data Robot, an automated machine learning platform the students had used in order to deploy the predictive model they developed.

"The students modeled the target customers using various machine learning approaches, performed exploratory analysis, and cleaned up data sets for a real-world problem," said Professor Elfiky.

The student team also developed an interactive Shiny App. In order to allow Restoration Hardware to find the optimum return on investment (ROI), the app would allow the company to input varying mailer costs, average transaction amount per client, and Teen purchase probability thresholds in order to explore the subsequent anticipated ROI.

“We utilized both linear and non-linear machine learning algorithms to build a predictive model to help Restoration Hardware glean customer insights to make actionable decisions. With our results we estimated a 15-fold increase in catalog distribution efficiency to RH teen purchasers while decreasing mailer costs,” said Kelly Lampman.

The students tested nearly 30 different models to come to the top performing model, a Python-based model called Extreme Gradient Booster Trees with Early Stopping. The team present their algorithm and model findings to Co-President and Chief Creative and Merchandising officer Eri Chaya, and Tobin Schiller, Vice President of Marketing this fall. 

"Overall this project was a great learning experience for the students as it helped them to tie together learnings from the programming classes, statistics, as well as the Marketing class," said Elfiky.