In this paper we discuss the design and performance of the BBN Call Director product for automatic call routing and the methodology for its deployment. The component technologies for the BBN Call Director are a statistical n-gram speech recognizer and a statistical topic identification system that, together, provide the framework for processing natural language responses from callers. To achieve commercial success, however, a superior deployment process is as important as the technology itself. In order to minimize the operational and financial risk for the call center, BBN has developed a deployment process that provides an integrated methodology for building the business case, optimizing performance, and proving the benefit of natural language call routing. This process is based on quantifying IVR benefit in terms of saved agent labor by analyzing end-to-end recordings of live calls. Routing accuracy is of critical importance because of the direct cost impact of a misrouted call to the call center. Experimental results on real traffic coming into a customer call center indicate that BBN Call Director can reduce the number of misrouted calls by 28%, which translates to a savings of 2 to 4 minutes for each of those calls. For large call centers handling several million calls per year, the corresponding cost savings can be in the millions of dollars.