When ‘Accuracy’ Is Misleading: A Primer On Understanding Identity-Based Advertising

August 22, 2018

Media planners aspire to put the right message in front of the right audience in the right place at the right time. The roadblock to make good on that promise is the fragmentation of the media landscape. We’ve all heard about this ad nauseum and live it through our endless devices. The issue with fragmentation is not that there’s a lot of opportunities to reach our target, it’s that our vantage point is shattered. We view consumers in silos. Our picture of them is based on the identification mechanism for each Internet access point. In other words, we view them as either cookies or UDIDs (unique device identifiers) depending on the device on which they’re accessing the Internet. Browser-based Internet access is cookie-based; mobile devices are generally UDID based. Walled gardens like Facebook or Google add an additional layer of complication because they don’t share their view of the consumer with outside parties. Don’t forget about connected TVs or gaming consoles.

To understand the implications of this, think about how many devices you have. I have 5. I use multiple browsers on both of my computers and since I can be identified by a cookie and UDID on my phone that means I have 10 tracking mechanisms associated to me. If an advertiser is trying to reach me, I effectively look like 10 different people. If the advertiser uses an average monthly frequency cap of 30 per person, I could conceivably receive 300 advertisements. I can’t think of a more effective way to waste a client’s money than that.

Enter identity-based advertising. Its champions aim to solve that issue by using a master identification profile for each person they want to message. In other words, they identify you regardless of which device you’re using. A lot of people are creeped out by this. Don’t be. It’s all anonymous. It’s not you, it’s profile 123409wer09lk. The underlying technology that makes it possible is known as a cross-device graph. The graph uses a mix of deterministic and probabilistic matching to tie all of those piecemeal identities into one master profile.

From the advertiser standpoint, that opens up a world of opportunity, including the ability to sequential message consumers regardless of the devices they’re on. However, before you invest your advertising spend on them let’s take a look at what you should be looking for to evaluate these companies.

A lot of those companies will tout a high level of accuracy as a part of their sales pitch. Often, you’ll hear rates as high as 97 percent accurate. That sounds pretty great, right? Don’t get too excited. It’s misleading and only gives you a sliver of the picture. Accuracy is a measure of how often the cross-device graph correctly identifies whether a device belongs to an individual. In other words, if every single device “company A” encounters are classified as not belonging to any person, then it would be 100 percent accurate. That’s a pretty useless CDG.

In order to understand how accurate a device graph is, you need to dig way back into your memory from stats class to pull out a confusion matrix. A confusion matrix, for those of you who don’t remember, is a series of measurements including misclassification rate, true positive rate, false positive rates, etc. The value of that is it informs you of the efficacy of the graph by unveiling how often the device graph is wrong, how precise is it, how often devices are actually matched to a master profile, etc.

Doing that will give you a better idea whether the sales pitch is smoke and mirrors. Don’t expect the results to be perfect. They won’t be, but it will give you a more representative picture of how well the device graph works compared to “97 percent accuracy.”