Personalized ads: how valuable compared to generic ads?

Exactly how valuable are personalized ads? 

Intuitively we think personalized ads have to be of great value: much more likely to convert than random or generic ads. But personalized ads are also something that can creep people out. And, personalizing advertising can just be flat out wrong, like when we google something for a parent or a spouse and floral dresses follow us around the internet for the next 3 months. 

(This is totally NOT a personal story.)

So are personalized ads better than generic ones? And if so, by how much?

Personalized ads: gold-standard research

Fortunately, we can answer this question with gold-standard research from just last year that I just ran into at a conference a couple of weeks ago. The study by Malika Korganbekova of Northwestern University’s Kellogg School of Management was massive: 2 years, data on 30 million e-commerce customers, experiments seen by 9 million consumers, 320,000 products, and an unknown number — but probably in the tens of millions at minimum — of ad impressions.

The results?

Personalized ads:

• Lead to 10% lower post-purchase product returns
• Increase repeat purchase probability by 2.3%
• Boost the likelihood that small sellers’ products get featured by 15%

On the flip side, generic ads:

• Decrease consumer welfare by 30% (costing them more money)
• Reduce small and niche sellers’ revenue by 8.6%

Interestingly, sellers earn up to 87% more revenue from personalized impressions compared to bestseller rankings. In other words, in an e-commerce scenario, having your product show up in a bestseller list might sound great, but it’s actually far worse than showing up in a personalized list of products tailored to a website visitor or app user.

The study

In the study Korganbekova tracked consumer interactions on the Wayfair platform, including viewing, clicking, scrolling, and purchasing actions. She and her co-author Cole Zuber used pixel-level data to observe which products each consumer viewed and clicked on over multiple sessions. Then she evaluated how different privacy policies (such as those found in the Safari and Chrome browsers) affected the personalization algorithms. And she created counterfactual scenarios by retraining personalization algorithms with fragmented data to simulate privacy restrictions.

Those privacy restrictions, like Safari’s deletion of first-party cookies after 7 days of inactivity, lead to a nearly 50% reduction in prediction accuracy, the study says. And that accuracy loss translates directly to less seller revenue.

Google’s Chrome isn’t as privacy-focused as Apple’s Safari, of course, but the Chrome policy that blocks cross-website tracking impacts people who have clicked on an ad — which in Wayfair’s case was 26% of the traffic — had “qualitatively similar” effects.

Korganbekova finds that these kinds of privacy features have 2 problematic impacts that were probably unforeseen by either regulators or policy makers at big tech companies:

1. They impact poor people more
   “Privacy policies disproportionately affect consumers who are more price responsive and have high search costs.”
2. They impact smaller companies more
   “The personalization algorithm tends to switch to emphasizing the popular products due to lack of data on smaller sellers.”

In other words, there’s a winner-take-all impact. The big get bigger, while the smaller brands have a harder — read: more expensive — path to market.

Personalized ads without identifiers?

Interestingly, there’s a path to recapturing some of that lost data for personalization, and it’ll sound fairly familiar to most marketers: probabilistic tracking. 

Korganbekova says that by “using IP address information as well as consumers’ detailed behavioral data to probabilistically recognize consumers even when the exact user identity is unknown,” marketers can recover as much as 73% of revenue lost to privacy policies … even the smaller brands that have very limited data.

This study, of course, is done on the web and therefore focused on Safari’s ITP (Intelligent Tracking Prevention), which blocks third-party cookies by default and first-party cookies if not refreshed within 7 days. Firefox has similar privacy protections, but Google’s Chrome, which was scheduled to lose the third-party cookie in 2024, has delayed that deletion.

The concept, however, of personalization boosting conversion is likely generalizable to mobile and mobile ads in the context of ATT on iOS and Privacy Sandbox, if it ever comes on Android.

And a better solution …

While marketers and platforms can improve their odds of matching the right ad for the right product with the right consumer via probabilistic means, there’s a more sure-fire solution that works every time and isn’t blocked by any browser or mobile operating system.

And that, of course is logging in.

“Notably, when consumers voluntarily log in during each visit, the platform can recognize them, without relying on cookies,” the study says. 

But there’s an issue:

“Our data reveal that approximately 37% of consumers choose to log in.”

I suspect that number might be higher in mobile games, where up to 70% of users play with friends or family, according to a 2020 report by Limelight Networks. Playing with others they know, of course, suggests that they log in to enable that functionality.

Whatever the actual number is, nothing beats opted-in, first-party. Of course, it’s mostly of use on-platform and still would require probabilistic extrapolations if marketers wanted to target those users off-platform.

Ultimately, the main point here is that personalized ads boost advertising conversions, in addition to helping smaller brands compete. And that’s credibly backed up by research on millions of people, not just a few A/B tests here or there.

Which is pretty good news for marketers.