Today’s column is written by Chris Peterson, managing partner at R2C Group.
Many technologies promise to deliver the holy grail of advertising: the ability to know exactly how advertising is performing and where it’s working best.
Consumers are tracked as they watch TV in their living rooms, order burgers in drive-throughs and commute on trains home. Colorful charts reveal a deep measurement of every move – the ads they saw, what actions they took, where they browsed, what train they took, what they bought and what drove them to buy it.
All of this apparent precision must be right because, well, it looks so accurate. Right?
As it turns out, most of it is wrong most of the time. Use some of these systems and your advertising becomes like a tire spinning in mud. The faster you optimize your investment using the reporting, the less you move forward.
It’s pretty simple: Digital attribution systems rarely, if ever, measure an actual lift in results over any kind of baseline of existing business. Instead they often operate in their own little attribution echo chambers, counting things as they come and go. These techniques are applied to paid search, paid social, TV, display and just about any other medium. The results may come from third-party ad tech or direct from the source.
Most ad tech ignores the fact that if you turn off all advertising, you will continue to see new customers showing up. For some brands, this can represent 10-40% of total customer acquisition, with no paid search, social or TV required.
This baseline of existing business then becomes the go-to pot of customers that some systems use to subsidize their attribution. People who would have bought from a brand regardless will still click, view and show up within the attribution because the brand’s paid media efforts just happen to show up at the right time.
If brands don’t account for true lift over baseline, then how can they ever know how to manage their media investment?
On the flip side of this ad tech are organizations that focus on a more “top down” approach, which means statistical modeling versus chasing digital bits and pieces. Statistical modeling aimed at advertising fundamentally attempts to measure advertising’s ability to generate lift over baseline first and go from there.
The problem with statistical modeling is twofold.
First, it just doesn’t seem as nifty as counting digital bits. Second, it can seem pretty black-box. You give some folks a bunch of data and they go away and don’t return until they have something that explains the advertising. What exactly are they up to? When you look at clicks and IP addresses, it’s not hard to have an intuitive sense for what the company is doing. With statistical modeling, it can feel like an act of faith.
The good news is that any strong approach to modeling has a clear expression of the strength of the model in terms of confidence. You can look at R-squared and mean absolute percentage error (MAPE) to have a fairly high degree of confidence. If your R-squared result is high and your MAPE is low, you can feel pretty good about it. That’s doesn’t make it perfect, but at the very least you are looking at what channels are contributing, over and above what a business already generates, which is a starting place.
The other way to gain greater faith in these approaches is to dive into specific examples of how the approach can yield very different results from chasing digital bits. Here’s one that’s very clear: A colleague showed me an email campaign promoting expensive coats for a national retailer that looked like a success when the retailer evaluated the clicks and followed people through purchase.
But when modeling was applied to the same campaign – looking at the baseline of current coat activity through the course of the campaign – it became clear that most of the email campaign’s buyers would have bought anyway. The email campaign simply inserted itself between the buyers and the coats at the right time.
In the world of advertising attribution, everything is somewhat right and wrong. There are no magic wands. But, if brands want to get closer to the truth, at the very least they need a way to account for people who will purchase from them regardless of any media they may buy, and then go from there.
Follow R2C Group (@r2cgroup) and AdExchanger (@adexchanger) on Twitter.
This post was syndicated from Ad Exchanger.