What You’ll Never See Using Channel-Level Attribution
Attribution modeling is, at its core, an analysis of the co-occurrence of events. This makes intuitive sense – if each purchase was only affected by one channel, there would be nothing to attribute! From your own everyday experience, you’ve been exposed to emails, catalogs, display ads, social media, tv… with each playing a different role in influencing your purchases.
Measuring attribution at the customer level exposes information that channel-level analysis will never see. That has always been our major hang-up regarding Marketing Mix Models (MMM). MMM were state of the art when they were invented. These models were pioneered by CPG companies as soon as they gained access to syndicated bar code scanner data. The techniques correlate response to signals of increases and decreases in marketing. However, the major shortcoming of these models is the underlying data.
By collecting and analyzing all data at the marketing channel level, you will never see co-occurrence. You won't know if a specific person who bought (or didn't buy) was marketed to once, twice, or a dozen times. In statistical terms, all of the marketing events are considered to be independent. This could not be much further from reality. We know that a second marketing touch is less powerful than the first. The events are not additive.
For example, if Mike received 10 emails, 12 display ads, and a television ad, a Marketing Mix Model would show a retailer that those 33 events happened, but not that Mike experienced that combination. The retailer would only know that those events happened in Mike’s geography – along with thousands of other marketing events.
In short, if you are brand marketer of consumer products that doesn’t have access to customer-level data, you can’t measure co-occurrence and you should stick with Marketing Mix Models. But if your spend is on direct marketing and digital, and you know who purchases (like multi-channel retail, travel, auto, etc.), then you really should be thinking of retiring your MMM for better, richer, bigger data and the methodologies to analyze marketing effectiveness.