How to Expertly Evaluate Your Campaign Results
Expert evaluation of catalog mailings was problematic in the past, but mostly just for those few catalogers who had retail stores. That’s not the case anymore.
Now virtually every catalog company is multichannel as customers increasingly use the Internet to place orders. The result often is a haphazard online collection of key codes.
In this article, we’ll examine the problems with traditional campaign analysis and how you can use a matchback between orders and mail files to substantially overcome these issues.
The Problems With Traditional Factored Allocations
While unattributable orders usually have been in the range of 10 percent to 20 percent, now unattributable orders are frequently more than 50 percent due to customers’ failure to provide a key code — even when asked — while placing orders on catalog Web sites. With 57 percent of shoppers presently crossing channels and Web sales trending upwards each year, traditional means of evaluating results through key-code analysis has become increasingly problematic.
Traditional methods to determine factored allocation of unknown orders are flawed for several reasons. First, as a general rule, statistical validity of results per segment evaluated requires a minimum response of 100 orders. So if you expect a 1 percent response to a mailing, a list segment size of 10,000 names is required. That’s why test orders of lists generally are 5,000 (2 percent expected response) to 10,000 names (1 percent expected response).
But those 100 orders must be real orders, not factored for unknowns or projected response. So if half of your incoming orders are of unknown origin, the segment size for statistical significance at a 1 percent response level actually is 20,000 names. Such segment or test list quantities for small mailers generally are risky or impossible.
Second, unallocated orders normally are factored against known response. For example, if 20 percent of response is unknown, the known response to each segment is increased by 20 percent across all segments. Unfortunately, unknown response never is random or proportional across all lists or segments.