Why response models are the wrong way to target
Who to target is a crucial question for direct marketers, and response models are a popular but difficult solution. No one wants to send spam or junk mail. David Baker takes on the problem in a recent blog post. Now Baker is a very smart guy. His position at Acxiom gives him great insight into what’s happening in email marketing. When he’s the author of the latest Email Insider post, I read it right away.
This week, David writes about trends he sees as marketers try to adjust to the regular doubling of email volume and the corresponding “decline in response since 2009.” He deplores “the ‘grow your list’ and ‘send it to as many as you can’ mindset,” and the “fragmented, high-cadence communication regimen” of too many companies. He says “Intelligence over Volume should be 2012’s banner.” So far, all of us at Loyalty Builders are in violent agreement with him.
But in his enthusiasm for response modeling as a solution, David has it all wrong. At Loyalty Builders we know that’s a flawed approach.
David believes that response modeling “should yield a positive business result.” His advice: “If you don’t have the resources to do this, I’d find them.” His mistake, and it’s a huge mistake, is assuming that response modeling is the only way to do behavioral targeting . To him they are coupled.
“Why isn’t response modeling and behavioral targeting through email more pervasive in this space?“ he wonders. Well, David, it’s not so much that response modeling demands considerable resources, which it does, but because it doesn’t work that well.
There is a better, simpler way. What every marketer needs are the answers to the three who-what-when questions: Who are the customers that are likely to buy? What are they likely to buy? When is the best time to reach them? With the answers to these questions marketers can build effective, efficient campaigns with very high response rates.
What every marketer needs
Response models are the wrong kind of analytics to answer these questions. Response models — more properly, marketing stimulus-response models — examine how customers will respond (buy or not buy) when treated in the future with a specific stimulus, for example an advertisement or catalog. They suffer from some deep problems. First, they only use the treated customers to build the model, ignoring valuable information from the rest of the population. Second, and more important, they only represent their slice of the customer world at a moment in time. Customers are always in motion, and today’s model doesn’t match tomorrow’s customers. Too many companies try to run the same model forever. Baker rightly recognizes this, saying “Don’t think you can do this once and run off it for years.”
He is also aware of the tremendous commitment in human resources this approach takes, a commitment that most companies can’t make, another strike against response models (and a big reason why it’s not more pervasive).
Companies should use the better way – predictive analytics that look at the past behavior of all the customers to answer the ‘who, what, when’ questions. Rather than building a model once to operate on the data, the better method (as proven by the spectacular results our customers get with our SaaS technology) is to let the data continuously drive and build evolving models. Behavioral targeting is here now and it is definitely yielding higher response rates.
Just don’t try to do it with response models, David.
Thanks to my colleague Bill Vorias for helping me work through Baker’s post and these ideas.