With IP Obfuscation Imminent, Contextual Is Not The Identity Substitute You’re Looking For

Apple’s iCloud Private Relay IP address obfuscation feature has been available for some time now, but it hasn’t been a game-changer for the broader IP ecosystem because of its limited availability – it’s functionally a premium feature because it’s only available to iCloud+ subscribers. Meanwhile the team behind Google’s Chrome web browser has been working on its own version under the umbrella of its Privacy Sandbox set of features, and the release is poised to be both much more impactful than Private Relay and much more imminent than the deprecation of the cookie. 

We’ll get the background out of the way quickly here: Most marketers have become reliant on the ability to target individuals or households belonging to a specific first-party list or third-party segment. But doing that requires that impression opportunities contain information that allows the buyer to identify the user on the other end of that impression. A lot of non-cookie targeting solutions lean heavily on IP addresses. Both cookies and IP addresses are now becoming unreliable in web and mobile app environments. 

Most ad targeting features – even entire companies – are based on the assumption that the only good way to do identity-less targeting is with content-contextual mechanisms that place an ad for your company alongside content relevant to your industry. But you and most advertisers like you have long known that’s not actually a viable replacement for your digital ad targeting needs. 

But why isn’t Contextual the identity substitute you’re looking for, and what might be? 

Contextual’s Constraints

Contextual targeting can be useful for brands who meet certain criteria: Their category is very participatory, meaning consumers do a lot of research, learning, or peer-to-peer engagement about the category. Brands serving a specific hobby is one example, where how-to’s and forum sites related to that hobby are great places to advertise. Expensive, research-intensive purchases like cars and appliances are another example, where auto review sites are high-value ad placements. But even then there are serious limits and downsides to contextual targeting. Specifically, scale and cost. 

The total scale you can achieve with content-contextual targeting  is very limited. There’s just not a lot of publishers serving a given research niche. Depending on your industry the contextual scale you can achieve may range from “that’s not quite enough” to “where is everybody?” And while well-targeted ads are going to perform better than poorly targeted ads, your marketing is not going to sell anything if your ads aren’t reaching anybody. 

Then there’s cost. As a natural result of highly limited supply, advertiser demand for impressions alongside industry-relevant contextual content is high, and so are the costs. It’s not just the CPMs that can be high either. Many of the most relevant publishers only sell remnant inventory programmatically, and achieving any real scale often requires making publisher-direct deals. This can mean making larger spend commitments with a single publisher than you may prefer, and requires personnel spending time making that happen manually. 

Interestingly, this cost dynamic applies to identity-capable impression opportunities as well. As the share of media capable of supporting identity-based targeting declines, the CPMs on those impressions are and will continue to be higher than those that don’t support identity-based targeting. 

Data-Driven Targeting Doesn’t Always Require Identity

So if the scale I can achieve with identity-based targeting is declining and the cost of those impressions is increasing, and the scale I can achieve with content-contextual targeting is low and the cost of those impressions is increasing, what’s an advertiser to do? 

The answer lies in the concept of Propensity – the inclination to behave in a particular way. At the end of the day, marketers’ goals require targeting that performs better than run-of-network. And the higher that performance lift, the better it is for the advertiser. Achieving that lift doesn’t require absolutely perfect individual-level mechanisms, and that truth is reflected in the reality that most financial institutions will run their advertising and their Firm Offer of Credit programs (that have specific regulatory requirements) as distinct practices with distinct channels and methodologies. 

The fun (and encouraging!) thing about Propensity is that advertisers can achieve shockingly high propensities with targeting methodologies that do not rely on impression-level identity, and do not rely on the limited scale and higher cost of content-contextual placements. There is a “but” though. This would be too good to be true if there wasn’t a “but”. 

Here the “but” is that building maximum-propensity no-identity targeting tactics actually requires better data than relatively simple targeting practices based on impression-level identity. Better in the sense of both higher scale (the rate at which you think you know something) and higher accuracy (the rate at which you actually know the right thing). 

Data must be higher accuracy because even with the best AI or machine learning, a propensity model will inherently multiply the inaccuracies in a dataset. Think about multiplying fractions. If your first fraction is your accuracy rate and the second fraction is the AI used to build your model, if either of those fractions get even a little low, the resulting fraction is uselessly tiny. Most identity-based ad targeting datasets have been built with the assumption that they only need to be accurate enough to drive the lift marketers are looking for. And because in that business more scale means more money, the decision to increase scale has almost always won out over the decision to increase accuracy. 

To be useful for building maximum-propensity no-identity targeting tactics, a provider’s data also needs to have scale. The “multiplying fractions” dynamic applies here, too. A provider needs to know the facts of both the input and output side of an equation at a high rate in order for a propensity model to even be possible. Fortunately, because of the incentive structure described above, you’re much more likely to find a provider with the necessary scale in the dataset than you are to find a provider with the necessary accuracy. 

Where to start? 

So we learned that the degradation of cookies as a targeting mechanism has been in progress for a while, and that the degradation of  IPs on the web and mobile will be happening more quickly and more completely than cookies. We learned why content-contextual targeting isn’t a real substitute for today’s identity-based targeting tactics because of both scale and cost. We learned that the best replacement uses data and models to reach an audience with a maximum propensity to convert to your desired outcome. And we learned that – because of upside-down incentives – most existing ad targeting datasets lack the accuracy to output propensity models good enough to drive the kind of lift marketers need. 

There will be a thousand different ways to build those propensity models, and you will probably get pitched half of them over the next year. And if you’re not getting pitched no-identity propensity-based targeting yet, it’s genuinely worth your time to get that ball rolling, identify a partner, and get something tested. 

But in your search, surprisingly, the exact type of inputs used by a propensity model doesn’t matter all that much – at least not as much as the quality of the data behind those inputs. So rather than spend your twenty minutes with a potential partner digging into the model’s architecture (and we all know they’ll rather baffle you with that particular bologna than talk about this next part), go deep on their data. Ask the hard questions that make them stutter and dance around the answer. One thing we recommend looking for is a company whose data is used for things like identity validation or fraud prevention in industries like finance or healthcare, and not just used for ad targeting. 

But it’s important to start now. You’ll benefit greatly from testing and refining your identity-less targeting practice when identity is still available on the measurement and attribution side. You were taught how to drive during the daytime for a reason. Get started on this before the sun goes down.