Case Studies

Optimizing Direct Marketer Mailings

Campbell, D., Erdahl, R., Johnson, D., Bibelnieks, E., Haydock, M., Bullock, M., & Crowder, H. (2001). Optimizing customer mail streams at Fingerhut. Interfaces, 31(01), 77 - 90.


Fingerhut. Fingerhut, a mail-order direct marketer, engaged the help of IBM in decreasing 340 million catalog mailings to its 7 million customers annually, because many of the catalogs were redundant and unproductive. Together, Fingerhut and IBM formed a partnership to meet Fingerhut's business need. Fingerhut supplied the statisticians, direct-marketing expertise, and programming resources, while IBM brought the horizontal-marketing concepts and mathematical-optimization expertise.

Traditionally, Fingerhut followed the standard marketing approach to managing customer contacts, choosing the best customers for each catalog. This vertical marketing process led to cannibalization or saturation between catalogs. The team developed a customer-selection system called mail-stream optimization (MSO), which focuses on the customer, not the catalog. The MSO system selected the most profitable sequence of catalogs for each customer.

Customers were partitioned into 100 micro classes. This was done based on long-term customer value and length of time as a customer. Statistical prediction models were devised to estimate customer value and advertising productivity over the next 12 months.

The advertising-allocation model used risk/return curves, noting the slope changes to make investment decisions. The curves describe the relationship between the advertising spending (risk) and the revenue (net returns) attained through that spending. The challenge was to invest in advertising up to the point where any continued spending yielded little, if any, profit. These risk/return curves were piecewise linear curves.

Matrices representing the similarity of merchandise, content, and features of various catalogs were used to model saturation of catalogs. A series of regression models used merchandise-similarity measures and promotional- and presentation-similarity measures as independent variables. The dependent variable came from Fingerhut's data warehouse.

The MSO System had several phases: Customer prediction, advertising allocation, customer-profit prediction, catalog saturation and its reduction, and then finally optimization. In the optimization phase, Fingerhut assigns the most profitable set of mail streams to customers while enforcing a variety of budget, management, and technical constraints. Optimization at Fingerhut had four steps:

  1. Cluster customers to produce a more easily managed optimization problem.
  2. Solve an optimization problem for each cluster, generating potential mail streams, given advertising constraints.
  3. Solve a global optimization problem to find the best allocation of mail streams across all customer clusters.
  4. Translate the cluster-level solution to the customer level.
The MSO System uses concepts in systems of linear equations, linear programming, linear regression, and organizational matrices. The project paid for itself within the first year, generating an annual profit-gain of $3.5 million. The MSO system is now being applied to prospective and inactive customers. In the future, it will be translated for use in other aspects of the business.

Back to Case Study Listing



Website by:
QuIC Solutions, Inc