Exceptional results.
Optimizing Your Contact Strategy
By: David Shepard
This Article First Appeared in Direct Magazine
Want to make your head hurt? Think about this. You have X million customers, N number of products and Y dollars to spend over some promotion period, say three to six months, and at least three contact channels: direct mail, telemarketing and e-mail.
The smart reply is that why would you want to make your head hurt. So, skip to another article or go for coffee. That’s kind of like what a lot of direct marketers do. They try to keep it simple. This month, they say, I’m going to promote product A, next month I’ll promote product B and so on; I’m going to use Direct Mail or telemarketing; my fiscal plan calls for so many pieces; I’ll sort my customer file on the basis of RFM or some regression or artificial intelligence model, and I’ll mail/call as deeply as my budget calls for. That’s it. Next month I’ll do it all over again.
I’m exaggerating you say. Nobody could be that simple minded in the year 2000. Any direct marketing company worth its salt (whatever that that means) with multiple products would not consciously decide to promote one particular product in one particular month. What they would do is calculate the expected value of promoting each potential product to each potential customer; and not only that, they would calculate this expected value number for all potential channels of distribution, and then allocate their monthly budget so as to maximize their return on their discretionary marketing budget.
That strategy certainly sounds better than the “Product of the Month” strategy, but does it go far enough? Suppose you could actually execute the strategy defined in the above paragraph. And then another month went by and you re-scored your database and again estimated the expected value of promoting each customer for each potential product, using a model that did not explicitly take into account the fall off that occurs when the same product is promoted in two consecutive months (or some other measure of a promotion period).
What would happen, a little more or less, is that the same persons would be matched against the same products and the prior months promotion strategy would be repeated, assuming the same budget and adjusting for seasonality.
So, is this good or bad? If there is no such thing as fatigue, then it’s fine. Assuming your models are as accurate as possible, which is the only assumption you can make. But what if there is fatigue and its mirror image, lets call that the “rest factor”, and while we’re at it we might as well consider, at least intellectually, the third possibility which is the “build up factor”.
To say it another way, if response in one period is affected by promotion activity in prior periods, then your models have to take these effects into account. Into account how? Obviously by affecting your estimates for the next promotion period, but also if “fatigue” and “rest” affects do exist and are relatively strong, the implication is that in order to maximize response or contribution, or whatever measure you wish to optimize, your models must extend over a planning period which is greater than one month.
If fatigue exists and is not dealt with directly the danger is over-promotion and the result is declining response rates. This effect will be most severe if the product line is limited so that your models keep directing the same products to the same people. This is relatively easy to observe and to some extent the only action that needs to be taken is to build in name suppression rules so that your models are prevented from over-promotion.
However, if we can measure the effect of “fatigue” and “rest” and if we can begin to measure the impact of using different channels in different combinations then we can begin to work toward optimizing your promotion or contact strategy over a longer planning period.
What’s required to accomplish this is a vehicle for estimating the expected value of each permitted contact strategy. Then the expected strategies for an individual can be ranked, and within the context of budget and processing constraints, a promotion plan can be developed that will indeed optimize results over a defined promotion cycle.
So, what are the major stumbling blocks to implementing such a process. Not knowing how to quantify “fatigue” and “rest” for starters. Therefore, the place to begin is to develop a testing environment in which reasonable strategies can be tested, not all-possible combinations, but a limited set of reasonable strategies. At the same time you can begin building the algorithms that will be required to handle the optimization process.
As always, I would be interested in hearing from readers that have made major strides in this direction. I'’ be glad to share both positive and negative experiences.