Targeting By Incrementality
In an age when elections are often decided by small margins, even minute changes in preferences among the electorate can be significant. Consequently, the accuracy and sophistication of forecasting within such campaigns would make many marketers envious.
Of course, campaigning and traditional marketing are not exactly the same. For example, attack campaigns often work in politics, while businesses have to be content with "price promotion" -- no one really owns up to doing it, we only do it because the other guy is doing it, and it's a great way to boost sales in the short term.
In addition to social media, attack ads and the like, it is important to recognize that today's political campaigns are integrated, extremely dynamic, and incorporate both mass media (branding, PR), as well as targeted marketing. And, of course, they are thoroughly multi-channel. In many cases, they use more sophisticated planning, optimization and execution than many marketing campaigns in our industry.
Such similarities should remind us marketers of something important, especially those of us who are interested in maximizing the incremental return of targeted marketing.
Targeting by incremental value has becoming the new norm in CRM and data-driven marketing. As a result, many marketers have developed metrics that seek to measure the "incremental" effect of marketing.
Similarly to political marketing, there are customers who are already loyal and will purchase on their own, some who will probably never respond, and "swing" customers who can be influenced to behave differently. It is more desirable to target those consumers who truly need the stimulus of another communication in order to purchase. This has led retailers, banks, and others to develop incremental value models that rank-order customers based not on the baseline value, but on the additional lift the focal campaign will generate.
But are marketers applying the resulting model scores correctly?
Simply using incremental rankings to optimize campaigns isn't necessarily wrong, but such an approach may be inappropriate if there is a threshold factor, like the Electoral College votes in presidential campaigns. What if the customer is only likely to buy once some hidden, underlying threshold is overcome? Consider the following example:
Let's assume we have two customers, A and B, that we have scored for incrementality. Our model shows that Customer A has a baseline probability of purchasing of 0.3 and a probability of 0.45 if mailed, giving us an incremental response score of 0.15. Meanwhile, Customer B's baseline is 0.45 and their response score is 0.51, yielding an incremental score of 0.06.
Using the incremental response scores alone, we would select Customer A over Customer B, since Customer A's incremental score is higher. But what if there is a threshold, let's say of 0.50 for the mail response score that governs the actual likelihood to respond? In that event, we would choose Customer B, since that extra communication is more likely to push him over the hurdle.
Unlike presidential campaigns, the threshold in consumer marketing is generally not observed directly. But there are ways to estimate such latent thresholds among customers and to then use this information in addition to the incremental scores. For example, in a recent project using an incremental model, using the incremental score alone to rank-order produced an incremental response of 4% among the top-scoring segment.
By calculating the underlying threshold, which was about 40% in this case, we reordered by selecting customers whose baseline score is below 40%, but the sum of baseline and incremental score is above 40%. This produced a top-scoring segment in which incremental response was now 12%.