Improve ROI of Direct Mail Campaigns by Understanding True Lifty

Lift is one of the most frequently used terms in the world of database marketing. The statisticians and modelers use Lift to gauge the effectiveness of predictive response models; and the direct mail marketers use Lift to measure the performance of direct mail campaigns. However, although both parties use the same word Lift, the definitions are actually quite different. A thorough understanding of the lift not only will eliminate confusions between a modeler and a marketer, but also will help modelers to improve response modeling and thus to generate higher ROI from direct mail campaigns.

Your Campaign Results Might Be Inflated

For a statistician or a modeler, lift is a measure of the effectiveness of a predictive model calculated as the ratio between the results obtained from model and the results without using the model.

Say you have 100,000 names in your database, you mail to everybody, and the overall response rate is 2%. If you randomly mail to 30% of your customers, you will get 600 responders based on the 2% overall response.

Now if you used the predictive response model to select the top 30% customers, out of these 30,000 mailed customers, 1,500 responded; the response rate is 5%.

Therefore, the Lift generated by using the model as oppose to not using the model is an astonishing 150%! (5% model – 2% random) / 2% random = 1.50 = 150%).

So in essence, the lift is calculated by comparing the best segments to the hold-out response of the overall average.

If this number is used for the purpose of evaluating the effectiveness of the models, that is fine. However, the problem is that many companies are still using it as the results of direct mail campaigns, thus inflating the actual incremental gain.

The True Lift

Direct mail marketers measure the lift of campaigns by comparing the results between the like customers with a special emphasis on incremental sales and margin. To do so, marketers create mail groups and control groups. The customers in the mail groups and the control groups are virtually the same except one will receive the treatment, the other won’t. Therefore, assuming everything else holds equal, the incremental sales or incremental margin can be credited to the direct mail efforts. Obviously, this is a more stringent method and it reflects the True Lift of direct mail campaigns.

To calculate the True Lift, you need to know six numbers: the total mailing cost, the number of mail customers, the average margin/customer of the mail, the average response rate of the mail customers, the average margin/customer of the control, and the average response rate of the control group.

Below is the formula that calculates the Lift of direct mail campaigns:

True Campaign Lift = Total Incremental Margin - Total Cost (including creatives, prints, postages, etc.)

Where Total Incremental Margin = Number of Mail Customers *(Average Response of Mail * Average Margin of Mail - Average Response of control * Average Margin of Control), and Total Cost = Number of Mail Customers * Cost per Customer

The formula can be rewritten as

Lift =Number of Mail Customers * (Average Response of Mail * Average Margin of Mail - Average Response of control * Average Margin of Control - Average Cost per Customer).

From the formula above you can see that in order to achieve higher incremental margin (the true lift), marketers need to find customers that will be truly motivated by the direct mail offers. They will either have to spend more (higher margin) or/and will be more responsive (higher response rate) than the control, or both. In other words, marketers would expect the modelers not only to find who is likely to respond, but more importantly, from these likely responders, to identify ones such that the incremental difference between the mail (treatment) and control is maximized.

The Response Model Is Not Designed to Achieve the True Lift.

Unfortunately the predictive response model alone cannot accomplish this goal. Many experienced marketers noticed that while a good predictive model can do a decent job selecting likely responders, it fails to identify those who have decided to take actions regardless of whether they receive the campaign contact or not. Using response model alone to generate mailing lists may also cause a couple of problems for marketers:
1. Lower profit margin. Some of the customers, especially the most frequent customers will be shopping anyway. Offering them dollar off or other discounts will hurt bottom line.
2. File fatigue. Since most of the likely responders will be selected by the response models, they will be mailed many times, which may cause file fatigue among these customers. Even worse, you may train your best customers into promotional customers.

Challenges for Modelers

The response modeling is an effective tool to identify the most likely responders. However, it is not designed to accomplish marketer’s business objective which is to maximize the incremental margin through the efforts of direct mail campaigns. When working with direct mail marketers, statisticians and modelers need to understand the meaning of the Lift from marketers’ standpoint of view. The challenges facing modelers are how to tweak their statistical models towards identifying customers that truly require incentives and stimulations so as to maximize the incremental value of the campaigns instead of simply identifying the optimal number of likely responders.