Reprinted with kind permission from Nov '95 Research Business Report, Skokie, IL
Tel: (847) 673-6284 E-Mail: rbrrfl@flash.net

In the sophisticated world of business, research modeling is both one of the hottest and most underutilized tools. As an accurate predictor of consumer purchase behavior, it is a vital piece of research for many cpg companies--but it has never been affordable for the overwhelming number of marketers that could put it to work. Even for many companies with budgeted modeling dollars, cost has squashed its widespread use. Until now.

"A lot of companies have paid modelers significant amounts of money to run their model and prepare a report. If you wanted more than that, you had to be prepared to pay more," describes Ross Link, President of Marketing Analytics, Inc. (Chicago, IL). In 1991, Link decided to reduce these costs in order to make models available to more potential users and to expand their overall usage. Link's new company developed his copyrighted Coefficient Generator, a statistical software providing a client with the means to automatically analyze customers' scanner POS data and estimate what drives a brand's sales.

The coefficients are the estimated sales impact of hypothesized sales drivers, including price, advertising, couponing and merchandising. "The Coefficient Generator is used with separate user workstation software that combines coefficients with internal costs and shipments to track and project profitability of marketing programs," explains Link. It can analyze pricing, promotion, ad and marketing mix, plus forecasts and ECR applications.

Link credits Media Marketing Assessment for the growing interest in modeling. MMA, an advertising modeler since 1989, convinced many marketers of modeling's reliability as a sales predictor. "MMA's techniques find large effects from advertising and have been popular," Link explains. "A lot of companies don't want to put large amounts of money into trade promotion, and MMA provided the justification they needed."

Running lots of models frequently for lots of products and over wide geographies are major reasons why clients love Link's software. "The Coefficient Generator automates the process, allowing them to do it in-house and without errors," Link notes. With the absence of people who are both both good modelers and good programmers, Link helps clients with their modeling specifications. "Once I put in the specs, clients don't have trouble updating them," he says.

Link deems his software state of the art due to its high level of automation and sophisticated estimation techniques like "shrinkage estimators." These help the Coefficient Generator mass-produce accurate and intuitively reasonable model estimates with minimal human intervention.

Additionally, the Coefficient Generator has built-in model specifications appropriate for the aggregate data most cpg companies work with. While the Coefficient Generator can run against store level data, many companies only have access to standard ACNielsen and IRI aggregate data (at key account level, market level or sales district level). Due to a phenomenon called aggregation bias, models based on this type of data--if not run properly--can be severely biased. To address this, the Coefficient Generator has built-in model specifications that Link says dramatically reduce aggregation bias problems. (In simple terms, Link says the largest and most common aggregation bias is caused by summarizing data across stores with different merchandising, prices or other key marketing activities. Link claims his built-in Store Group model avoids doing this.)

While modeling accuracy is obviously of paramount importance, cost savings has an obvious impact on the bottom line, which Link's clients often use to sell the system to management. "Whether you do a lot of reports or one national model for a category once a year, my software costs about $40,000," reveals Link. "That's roughly the same as buying one customized study from MMA. So, if you allocate the $40,000 internally across a year's worth of work with the Coefficient Generator, the cost per report drops to one-tenth to one-one hundredth of a single report cost. And there are few software problems.

Link's clients include Kraft Foods, Nestle Frozen Foods, Tropicana Products, Millward Brown International and management consultant Booz Allen & Hamilton. Evidence of the accuracy of the Coefficient Generator comes from renewals on his software. "My clients have the right to license my software for one year, look at my models and tell me they can do it themselves. That has never happened," says Link with a smile.

Prevision Corporation's (Wellesley, MA) 1994 Marketplace Data User Survey featured a special look at purchase behavior modeling. Half the respondents stated they resorted to such modeling on a regular basis in 1994, and almost 75% went on the record as planning to use it during 1995 to prepare for trade promotion, consumer couponing, advertising, marketing mix allocation, sales forecasting for ECR or account-level modeling to support sales. (Information about the Prevision report is available by calling 617-239-9800.) These intentions show the immediate future for this type of modeling is promising.