Predict Who Your Best Prospects Are: How Predictive Modeling Can Increase Your Direct Marketing Response

Applying Science to the Art of Direct Marketing

People are creatures of habit, making them predictable.

We all know the truth of this intuitive statement, but the challenge is to predict consumer response accurately and consistently. Using the measurement sciences to increase predictive capability is available to wise practitioners. Applying analytical thinking to marketing efforts in order to find patterns of response (or profitability, or frequency of purchase, etc.) is an often misunderstood art. However, its basic simplicity is also intuitive: our behaviors are consistently associated with other aspects of our lives.

It’s those other aspects that a comprehensive marketing database reveals about consumers. One that includes hundreds of demographic, lifestyle, financial, and purchase characteristics. And through statistical pattern detection we find the many aspects of people’s lives that point to their demand for specific products. These aspects may be referred to as “predictors.” Unlike simplistic selection criteria, multivariate approaches (e.g. regression models, segmentation systems, etc.) consider many variables to impact response at once. From the combination of all the significant predictors of response, we can construct a “scorecard” prediction of how responsive each prospect may be.

It sounds simple, we know. But in reality, there are many chances for things to go wrong. Not all predictive models are created equally well. Working with an experienced analyst enables your firm to move rapidly up the learning curve and take the right steps. Input data must be correct and complete. Interpretation must strictly follow the evidence. Application to a prospect list must be tested properly, and a robust "learning cycle" must be in place to gather and make use of the evidence from past campaigns. And finally, the trust placed in an analyst must be complemented by collaboration and reciprocated by dependable, measurable performance.

The Logic of Regression

Fundamental to the process is the logic of regression. Simply stated, the approach reveals the basic, observable drivers of consumer behavior. An example is the FICO score that we all have. (We’ll avoid anything that looks like a formula in this simplified example.)

a)   Which consumer characteristics are statistically significant predictors of the desired behavior?

            A prediction of likelihood to default on a loan uses several different measures that are significant predictors of default: length of credit experience, number of accounts in good stead, number of times 120+ days late among them.

b)   Do those predictors have a positive or negative effect on the desired behavior?

            More credit experience is better than less, more good accounts are better, and fewer 120+ late accounts all add up to a better FICO—and prediction of good payment.

c)   How much more important is any one of the predictors than any other?

            The ‘weights’ that regression assigns each aspect of the model tell us that length of experience is twice as important as the number of good accounts. . .which is slightly more important than number of 120+ late payments in calculating our “score.”

These elements of regression are not decided by an analyst, but by objective rules of statistical significance that are 100+ years old: proven accurate countless times. The statistical algorithms we reveal are applied to every consumer in our national database, making selection of the most likely responders easy and effective.

A Few Tips:

1)    Ensure that if you are starting with your in-house data it is updated, clean, and of high quality. Work with a vendor who can validate and update your in-house mailing list using Change of Address services, can eliminate duplicate records, and append data elements at a high match rate.

2)    Work with a vendor who has a comprehensive consumer database with many multi-dimensional characteristics. Some combination of these characteristics will become predictors for your predictive model.

3)    Insist on a professional who has deep experience building predictive data models - and has seen the results of his/her work play out in the marketplace. There are times when you’ll come across people with the title, but not the experience or skill set you need. An analyst who has been directly involved in marketing campaigns or project management would bring a broader perspective on customer acquisition than someone who has only statistical knowledge.

4)    Ensure results are tracked, measured, and consistent with your strategic imperative. The metrics you should use to measure success include, but are not limited to, Response Rate, Conversion, Cost-Per-Acquisition (CPA), or Cost-Per-Revenue-Dollar (CPRD). If you’re focusing on getting leads for your sales team, Response Rate may suffice. If, however, you want to optimize the profitability of your campaigns, CPRD would produce a better result. You will get what you ask for so make sure you ask for what you want! You should get to the point where you know exactly how many customers and how much revenue you’ll gain from each direct mail campaign.

5)    Optimize your feedback: prior responses are used to improve upon the next model. You should develop and implement a sort of Learning Cycle for your direct mail program. That’s the promise of applying predictive modeling to direct marketing—consistent, predictable improvement of your overall campaign performance.

Predictive modeling—and the broader discipline of evidence-based decision systems—delivers the full potential of targeted direct mail to those who embrace the guidance of a skilled and experienced practitioner. At DTS, we work in concert with our clients to help them target with greater precision and increase ROI. Contact us if you want to take your direct marketing to unprecedented levels of success.