Circulation: Use Regression Analysis to Optimize Catalog Circ and Sales Data

The Internet certainly has changed since 1993 when I posted my first businesses, which were catalog-based, on the Web. I sold industrial liners for corrosion and environmental protection, and I had a mail-order pond and landscape supply catalog. Each targeted a different clientele, and I learned that marketing strategies have to be directed specifically at target markets.

You must determine the effectiveness of any promotional sales program, because it will add to your overhead allocations. Statistical analysis is one such powerful tool. And regression analysis in particular is at the top of my list. Any tool that allows you to determine if your catalogs are being sent to the optimal locations and the best prospect base is something that shouldn’t be ignored.

Regression analysis has become much simpler to do since it first appeared a few years ago.

Although it sounds intimidating, regression analysis simply allows you to quantify the relationship between two variables, such as the comparison between discounts offered and a sales increase. It enables you to predict what a sales increase would be with a given discount offer and be statistically correct more than 95 percent of the time. If the correlation is too small or even non-existent, you then can look for other factors that may be affecting your bottom line.

Say you have customers on both the U.S. East and West coasts, but you’ve come to suspect that discount coupons offered in one area don’t generate the same sales boost found in the other. However, in our example the average sales increases in the two areas were very close, 8.84 percent and 9.17 percent, and we want to determine if it actually was the discount offer that was the driver.

To discern how purchases fluctuate with discounts, the amounts of the actual discounts in both areas will vary, and this becomes the independent variable. We’ll offer discounts from 5 percent to 10 percent in monthly catalog mailings inserted as discount coupons for a year. The dependent variable is the percent increase in sales, as this variable is entirely dependent on the discount offered (or it should be; if it’s not, then something else is affecting sales).

We want to know if sales increases on the East Coast match increases on the West Coast. Before we can justify spending time and money to discover if differences are due to things such as demographics or customers’ personal income distribution, we first must discern if there’s a statistical difference in sales increases. Gather the data from the coupons returned from the two localities and chart them with their corresponding sales increases. The more data points gathered, the better regression analysis works. Use the graphing capabilities of your spreadsheet software to create XY scatter diagrams to give a better visual illustration.

When graphing East Coast sales increases: If a diagonal line greater than a 30 degree slope can be drawn with the same number of points above and below it, and equidistant from the line (with the points still being fairly close to the line), this indicates there’s a good correlation between the two variables.

But, if on the West Coast chart you find it would be more difficult to force a line through the points while being close to any of them, your discount offer may not be boosting sales as you suspected.

Next, highlight the appropriate columns in the number charts and run the linear regression program according to the routine used by your spreadsheet program. (Using my favorite spreadsheet, this process takes only four or so mouseclicks.) Analyze the data that’s generated.

¥ The regression R-squared tells you how well your two variables matched -- a simple cause and effect. R-squared ranges from 0.0 to 1.0, and a value of more than 0.7 is considered a satisfactory correlation.

¥ The X-coefficient tells you how much the dependent variable changes for each unit increase in the independent variable. This number also is used to predict future results when used with the predictor formula and added to the constant.

¥ The standard error of coefficient shows how close the prediction will be.

¥ The T-statistic tells you if the error is large or small. Any number greater than 2 is a good indication of a reliable predictor.

Back to our regression analysis of the discount coupons offered over the course of a year: R-squared on the East Coast is greater than 0.7, so we can assume there’s a good relationship between discounts and new sales. The West Coast comes in at less than 0.7, so we know there’s no correlation; for whatever reason, discounts offered have less impact on generating new sales.

The T-statistic is more than 2 for the East Coast but not for the West Coast, which means any information gathered from coupons from the West Coast can’t be used to determine potential sales increases with future discounts offered. Unknown forces are driving West Coast customers to spend money differently, and discounts have a smaller impact on their purchases.

So the East Coast regression analysis predicts a 5.9 percent increase in sales for every 8 percent discount offered. But on the West Coast, there will be only a 2.1 percent increase in sales for the same amount of discount, which doesn’t make sense at all. Remember, the sales increase averages are almost the same -- 8.89 percent vs. 9.17 percent. What’s wrong?

We’re not optimizing sales on the West Coast. We can’t predict with certainty how customers there will spend. On the other hand, since the increases were fairly similar in both regions (even though the West Coast has no statistical match with the discounts), if we could find out how to improve sales there, then double-digit sales increases should be easily reachable even with minimal discount offerings.

The differences in spending between the two regions could be explained by, for example, customers’ ages, gender, wage brackets, ethnicity or number of catalog mailings by direct competitors. Upon further study, we can determine if we would want to do the following to increase sales on the West Coast:

¥ Raise the discount offer for West Coast customers.

¥ Lower the initial cost on the West Coast.

¥ Change marketing and advertising strategies on the West Coast.

¥ Target different consumers on the West Coast.

¥ Abandon West Coast mailings to the competition.

These are some of the benefits of using regression analysis. It gives you the data you need to determine, in this scenario, how discounts affect sales. It indicates where you’re strong and where you may have a problem. Minimizing marketing and advertising overheads while maximizing sales and profits, after all, is the name of the direct marketing game.

Orest Protch is an Alberta, Canada-based merchant and freelance writer. Contact him at: (780) 539-0040 or via e-mail at: oprotch@telusplanet.net.