Deciding where to advertise has become more agonizing than ever. Which publications will produce the best leads? You can't rely strictly on a magazine's demographics, says Gordon Gossage, head of sales and marketing at MathSoft, a Cambridge, Mass., software company that has grown its sales from $1 million to $10 million in five years.

Instead, Gossage came up with his own way of measuring the performance of the 15-plus technical journals in which MathSoft advertises. Once a year he surveys a 10,000-name sample of prior leads to identify which of them turned into sales.

Gossage draws from his database of some 160,000 leads; he takes a sample of phone-in leads and reader-service-card leads from the past three months to a year to get the 10,000 names. He sends each a mailer with a reply card that includes a $1 bill and a short survey on the company's product, Mathcad. Each reply card is coded by the magazine source for tabulation. Nearly 30% responded to the latest survey. Here's what MathSoft learned after tracking the performance of 17 publications:

* Not all leads are equal. Gossage found that a lead turns into a sale 25% of the time at best, and at worst just 6% of the time. Even publications with roughly the same average cost per lead (number of leads generated divided by the cost of ad space) delivered very different rates of sales conversion. "The conversion test shows who's pumping up leads," says Gossage. "Magazines know that most people just look at the number of leads."

* The true cost of advertising. Once Gossage had the conversion rates for all the publications in hand, he was able to calculate the marketing cost per sale (average cost per lead divided by the conversion rate, multiplied by a factor of .68 to account for estimated sales through word of mouth).

* Which titles to keep. After crunching the numbers, Gossage dropped the six publications with a marketing cost per sale of more than $80. "We know that above $80, we can't be profitable." That alone lowered MathSoft's advertising costs as a percentage of sales by 4%.

MathSoft brought back one publication it had dropped, based on readership data supplied by the journal. The title scored high enough to rejoin the lineup. Perhaps most important, Gossage increased the frequency for the four top-producing titles, where the marketing cost per sale ranged from $25 to $48.

Dropping some titles and increasing advertising in others yielded 30% more sales in the second half of 1991, with about the same ad budget. Says Gossage, "It's like cutting your losses and letting your winners run." -- Susan Greco