Don’t Forget the Human Element of Data Analysis

In hot pursuit of actionable insights, data mining analysts and modelers are often held in the highest esteem by marketing organizations. They have magic at their disposal that will transform unwieldy data into new insights about our customers, and 75% of our circulation into 100% of our response.

But no amount of analytic muscle or high-tech software can guarantee that actionable insights materialize into valuable business initiatives without a great deal of human intervention. It’s here, in the hand-off between the art and science of advanced analytics and the reality of program execution, that we need to acknowledge there is a skill separate from database science that enables great marketing.

Here’s some real-life examples that back this up:

Customer Profiling: A purveyor of telephone equipment and headsets commissions a customer profiling analysis to uncover its best customers. It turns out their best buyers have a hugely disproportionate interest in motorcycles —what a discovery. As visions of Harley co-marketing programs danced in their heads, a second look revealed the data append vendor accidentally inserted a ‘1’ in the column marked as the ” interest in motorcycles” field.

Testing and Personalization: An enterprise software company tests and re-tests formats for its direct mail seminar invitation. After multiple quarters of statistically valid results, the company rolls out with a personalized #10 letter package. In a perfect storm of errors, the personalization printer inadvertently swaps the title field for the last name field in the input record. Even more unlikely, he truncates the title field (now last name) to only four characters. The champion package with a personalized letter now reads, ” Dear Mr. Anal.” What’s worse—Mr. Anal, aka John Smith, Analyst, XYZ Corporation, is a personal friend of the vice president of marketing. That particular letter garners a 100% response rate.

Scoring Models: A credit card marketer asks its database vendor to build a scoring model to identify the most likely responders to an insurance offer included with a credit card statement. The logistic regression model looks good and is expected to lift response substantially in the top three deciles. After disappointing results, it’s discovered the most likely responders were assigned to the lowest deciles and the least likely responders were assigned to the highest deciles. The model worked perfectly—but customers were targeted in reverse order.

When we hear about the power of genetic algorithms and the promise of marketing automation, we might want to reflect on the value of human intervention, business sense, common sense and plain old experience. An experienced vertically integrated staff of marketers who care every step of the way from model to mail might just be the secret sauce that could be more valuable than any analytic technique or software package will ever be.

Ellen Sato is vice president of strategic services at Haggin Marketing.