When Does Reactivation Modeling Make Sense?

The ability to ask the right questions is a critical factor for effective analytical work. This is the key point I made in a recent article, "Asking the Right Questions in Data Analysis" (Chief Marketer, March 11, 2011). I also noted that asking the right questions typically does not require an advanced degree in statistics. And that, conversely, having an advanced degree in statistics does not guarantee that the right questions will be asked. For example:

I referenced a discussion group on the best ways to build a Retail attrition model. Specifically, the stated goal was to predict which customers are likely to defect, and when. However, none of the data miners had thought to ask, "Does it even make sense to try to build an attrition model in a Retail environment?"

My contrarian perspective was that attrition models typically do not make good sense in verticals such as Retail in which non-contractual relationships with customers are the norm. In these sorts of industries, we typically never know for sure if a customer has defected, much less the exact moment that the defection took place.

The Application to Reactivation Modeling

Let's apply the same sort of perspective to reactivation modeling. I received a phone call recently from a company that has hundreds of thousands of inactive customers. As is often the case, the inactive file is much larger than the number of actives. Therefore, the caller wanted to hire an analytical firm to build statistics-based reactivation models. His reasoning was that if he could reactivate just a few percent of his inactives he would have a big impact on his top and bottom line.

The caller wanted to know all sorts of things about Wheaton Group, including the number of reactivation models we had constructed, the statistical techniques we employed, the success we had achieved, and the cost of our services.

How would you have responded? A thoughtful analytical professional would first explore whether, with the circumstances at hand, reactivation models even make sense. The following is how the line of questioning might proceed:

Questions to Ask

FIRST, determine the industry in which the company operates, and whether contractual or non-contractual customer relationships are the norm. This plays a major role in defining when, and how, customers move from active to inactive status.

It turned out that the caller worked for a catalog/e-commerce company. Therefore, the norm is non-contractual relationships. As we just established with attrition, non-contractual relationships mean that we typically never know for sure when - or even if - a customer has slipped into inactive status. Instead, the likelihood of being inactive is probabilistic, and is contingent upon a number of well-known direct marketing factors such as time since the last purchase ("Recency").

SECOND, find out how the company defines an inactive customer. The caller's answer was fairly typical. Customers are deemed to be inactive when it no longer makes financial sense to contact them by direct mail. This is important because it provides insight into the number of interactions that take place between a company and its customers before they fall to a status of inactive.

It turned out that the catalog/e-commerce company would mail customers as many as 40 straight times without a response, across quite a few years, before deeming them to be inactive. Therefore, these customers have established quite a precedent of not being interested in the company's products!

This is very different from industries such as Telecommunications in which contractual relationships are the norm, and customers can go from active to inactive status at just about any time. And, if we are smart enough to ask during the time of the cancellation, we also know the reason. Therefore, the circumstances naturally lend themselves in contractual industries to determining which of the canceled customers to try to reactivate, and how.

THIRD, ask what the company plans to do to squeeze out orders from customers who have failed to respond 40 straight times. The caller's answer was fairly typical. He talked about sending a last-chance offer with a significant discount to the subset of inactives that the modeling exercise deems to be financially-viable.

FOURTH, ask these questions:

1) How many follow-up orders should you expect from people who - after not ordering 40 straight times - have to be "bribed" with a significant discount in order to reactivate?

2) Will you have to keep bribing them to generate these follow-up orders?

3) Are there better ways to spend your time and money?

4) Even if you decide that it makes sense to build reactivation models, is there something else you can do to enhance the reactivation process? Perhaps the company should try to understand why these people have failed to respond 40 straight times. There might are all sorts of reasons for inactivity, including:

They were unhappy with their previous order. Their needs changed. They discovered that they can purchase the merchandise less expensively somewhere else. Or, they have moved, and therefore are not even receiving the promotions (although the self-ordering functionality of e-commerce muddies this one quite a bit).

Each of these has ramifications on the strategies and tactics that define the reactivation process!

Closing Thoughts

Effective reactivation modeling - just like any kind of data mining and analysis - requires a lot more than just technical knowledge. It requires a business person's perspective, and sufficient experience, to ask the right questions. Often, asking the right questions alters the analytical approach that is taken. Sometimes, the right questions even result in the cancellation of a project that, at first, seemed to make perfect sense. Regardless, the effect of asking the right questions is to enhance the leverage to be gained from reactivation models as well as all other forms of analytics.