Penetration Models vs. Response Models

Let’s say you want to predict the probability of your customers’ response to one or more promotions – an upcoming direct mail or telemarketing effort, or even the next series of campaigns over a particular period or season.

The usual method is to go back to the promotions you wish to include in your modeling history and hope someone remembered to take a snapshot of the database or customer file when the names were selected for the campaigns in question. The logic being, you want to know what the customers looked like in terms of their behavior and promotion history at the time they were targeted.

For example, a typically important behavior variable is the number of days since the last transaction; a key promotion variable is the number of days since the last telemarketing contact.

There’s nothing wrong with this method. It’s the standard procedure. But what if no one kept those data snapshots each time the database or portions of it were used for a promotion? Are you out of the modeling business, at least for the next six months to a year – the time it would take to build up your modeling files?

An Alternative

Probably not, thanks to penetration modeling. This is what some call “look-alike” modeling, a term that’s best avoided because it sounds much the same as profiling.

By profiling we mean the practice of defining and finding your best customers, then overlaying their files with demographic data to build profiles.

After this exercise is over, you can find prospects from your database or elsewhere who look like your best customers and promote to them. This frequently fails, however, because similar descriptive profiles don’t ensure comparable response or performance. For example, while all of your customers may be green-eyed men, not all green-eyed men are – or want to be – your customers.

Penetration models, on the other hand, are not descriptive. They are, by definition, predictive, and can be used as a good starting point for developing an effective customer contact strategy.

In Practice

Here’s how they work. Suppose you sell multiple products and you’ve already segmented your customer database into a handful of life-stage or lifestyle segments. Now you’re trying to decide which products to offer to members of each segment.

Yes, research – and not product availability – should drive this decision. But the reality is, more often than not, you find that something needs to be done immediately. Then what?

The first step might be to measure the penetration of each product within each segment, and rank the products in terms of their penetration rate.

Let’s assume that the penetration rates range from 2% to 10%. From a technical modeling perspective, penetration rate is equivalent to response rate. And, as in response modeling, where the question is “Can we use logistic regression to assign each prospect a probability of response?” in penetration modeling the question becomes “Can we assign each prospect a probability of already owning the product in question?”

Once again, we will use logistic regression to score and assign probabilities of ownership. Success can be gauged by using the same decile analysis technique used to evaluate response models.

Let’s assume we’re modeling a product with a 5% penetration. If our model is good, the penetration rate among those in the top decile should be four or five times the penetration rate of those in the bottom decile. If the model is terrible, the penetration rate in each decile will be around 5%.

The next step requires an assumption that customers with a high probability of owning the product (but who don’t own it yet) will be much more likely to respond to a promotion than customers who have a low probability of owning the product.

This assumption is almost always true. Of course, after you do your promotion to high-probability owners, you will then be in a position to model and refine your targeting. Your response model will likely include variables that didn’t appear in the penetration models – especially variables that have to do with recency of purchase or promotion history.