A New Advertising Era Based On Modeling And Measurement

Sponsored post byNeal RichterDirector, Bidding Science and Engineering


This isn’t another article about the death of third-party cookies.

Cookies are false precision and they always have been. Conversations about cookies and what advertisers are going to do when they’re gone are missing the point. It’s a distraction.

Instead, we should be talking about how we can actually engage consumers now and in the future with relevant, useful advertising. Repeatedly or relentlessly showing ads for the shoes shoppers considered but didn’t purchase is a cookie-driven tactic. It isn’t relevant or useful.

Instead, advertisers need modeled measurement and inference-based methodologies that consider shopping interest, contextual signals and purchase signals – none of which require a cookie.

The real programmatic power: insights

Yes, a demand-side platform is a tool advertisers are using to buy media. But advertisers should be thinking about their DSP with an expanded lens. It’s not just about access to inventory.

Can your DSP help you understand whether your ads worked or didn’t? Is it getting “smarter” with every campaign you run, using machine learning and a breadth of signals to make better decisions in the future? How accurate are the signals your DSP has access to? In what ways are they differentiated? Are those signals predictive for the long term or just the next click?

You should be confident in the answers to those questions.

At Amazon Ads, we place a great deal of emphasis on the insights Amazon DSP provides for our advertisers, not just the inventory we can access. When advertisers run campaigns through Amazon DSP, our technology builds models that help us understand shopping patterns and predict ad performance.

Our vision is to make those insights available to advertisers and reusable for future campaigns, using the billions of signals unique to Amazon, with models getting smarter the more they’re used.

After recent enhancements made to Amazon DSP, brands across verticals saw cost per click decrease by 24.7% – without having to take any action – simply because the algorithm started to better “understand” the relative value of an opportunity to show an ad to an audience based on expected ROAS.

It’s up to advertisers to help consumers find the things they need or discover the things they didn’t know they needed. We shouldn’t be distracting them. True insights and predictive modeling, not cookies, make this possible.

Old school meets new school

Before the cookie, much of advertising was about understanding how groups of consumers behaved and the products and messaging that would engage them. Think broad demographics, such as “men ages 18-35.”

At Amazon, we’re dedicated to earning and maintaining our customers’ trust, whether they’re brands using our ad products or consumers shopping on Amazon.com. So, when advertisers use Amazon DSP, they can tap into modeling capabilities and audience insights that are unique to Amazon Ads, combine them with their own signals and leverage insights that can’t be found anywhere else – without compromising that trust.

But the new era of advertising should be reminiscent of those early instances of machine learning recommendation on Amazon: that widget at the bottom of a product page that reads, “shoppers who bought this also bought this.”. While this is not an advertising feature, it’s a good example of the direction ad tech is headed: using aggregate signals that are population-based.

Thanks to our measurement capabilities, brands can actually understand how their ad dollars inform and drive shopping behaviors collectively, proving the impact of their efforts. That’s why our advertising approach is a combination of old-school tactics and new-school advertising. We believe brands should be authentically engaging customers and helping them in their shopping journey, instead of mourning the end of cookies.

Leave a Reply