According to the U.S. Mint, it is estimated that people have more than $7.7 billion in coins sitting in their homes. On average, these coins total $30 to $50 per household—about twice what the average consumer believes them to be. The perceived inconvenience associated with large amounts of coins has transformed Coinstar into the largest coin-to-cash conversion business in the market today. Coinstar machines count shoppers' accumulated coins and dispense a voucher that can be exchanged in the store for cash or groceries. It's a priority for Coinstar to identify new and profitable locations on an on-going basis as demand for Coinstar's services continues to grow.
Coinstar's decision-making process was limited to educated guesses based on demographic data used within their Geographic Information System (GIS). GIS enables Coinstar to manipulate, analyze and visualize geographic aspects of data, most frequently in the form of a map. However, due to the large number of variables involved and the depth of analysis that was required, a more comprehensive statistical package was needed to digest their large dataset. Although more than 400 key variables were available to Coinstar, they only analyzed and used four variables due to resourcing and the limitations involved in working with Microsoft Excel.
By combining SPSS Inc.'s data integration and analysis capabilities with various demographic and geography-based variables, Angelo Taylor, Coinstar's GIS research manager, and his market research colleagues were able to identify prospective installation locations with the best potential for growth and profitability.
Since its inception in 1991, Coinstar has provided supermarket patrons with an easy and convenient way to turn their cumbersome loose change into cash. Coinstar takes the inconvenience out of the coin conversion process, counting up to 600 coins per minute and dispensing a voucher that can be redeemed for cash or groceries. With more than 10,000 locations in the U.S., Canada and the United Kingdom, each collecting a service charge for every transaction, Coinstar has grown to become a publicly traded company with more than 500 employees.
One of Coinstar's key concerns surrounds the law of diminishing returns—saturating an area with so many machines that it impacts profitability. Previously, Coinstar based decisions for machine locations primarily on intuition, general knowledge about the potential account and educated guesses with GIS. Although more than 400 key GIS variables were available to Coinstar to be analyzed using Microsoft® Excel, the program was not well suited.
“We had information about each store’s vicinity and local demographics for various broad census categories, such as annual household income, household make-up, population, etc., and the performance of each machine,” said Taylor. “But we didn’t have a comprehensive enough statistical platform to turn that data into information that could really benefit our decision-making process.”
Taylor wanted a statistical system that would help determine high correlations between the numerous variables available to them and the performance of their machine locations. He had previously used PASW Statistics in college, and felt its ease of use and extensive analysis capabilities were an appropriate match for Coinstar’s needs.
“I’ve always found it intuitive and easy to use,” Taylor said. “If you’re used to using Microsoft Windows programs, you can get up and running fairly quickly with the application, and you can appreciate its breadth of analytical functionality.”
Using the statistics gathered from PASW Statistics, we’ve been able to analyze and model machine performance across the country. In this way, we were able to forecast how the machines would perform in any given location.
Geographic Information System (GIS) Research Manager
With GIS information readily available, Taylor and his associates created regression models with PASW Statistics to determine the most profitable locations for Coinstar machines. These regression models allowed Taylor and his team to perform more rigorous and complex financial modeling, offering a more ophisticated sense of how well the machines would perform in any given area.
“Using the statistics gathered from PASW Statistics, we can analyze and model machine performance across the country,” said Taylor. “In this way, we can forecast how the machines will perform in any given location.”
Before deploying any model, Taylor and his team can use standard PASW Statistics output—including charts, tables, and tests—to demonstrate to management a high correlation between the model’s estimates and historical machine performance. Subsequently, at regularly scheduled meetings, an installation task force examines and prioritizes potential new installations by their forecasted performance and other pertinent business considerations.
“The statistics we gather help us to forecast how the machines will perform in any given location,” Taylor said. “The estimates from the models have served as the key piece of information in go/no-go installation decisions for some time.”
Using PASW Statistics’ mining capabilities, Taylor increased the R-squared (a measure of model performance) to 65 percent, more than 1.6 times better than the prior Excel model’s original 40 percent. Furthermore, the improvements reduced the errors associated with the previous forecasting model by 50 percent. For example, if the old model had a hypothetical average discrepancy of $15,000 when forecast¬ing a machine’s performance, PASW Statistics helped to reduce that discrepancy to $7,500. With more accurate models to forecast future volume, Coinstar now performs increasingly complex financial modeling when choosing both individual and groups of installations. This new approach has helped Coinstar focus on the most profitable locations for future expansion of their network.
can make your organization