The new trend in platform advertising is to let the machines do it all.
That’s true for Google, it’s true for Meta, it’s true for Amazon as of this week.
On Thursday, Amazon DSP rolled out a system-wide upgrade of its machine learning and prediction algorithms, Neal Richter, the group’s director of data science engineering, told AdExchanger.
According to Richter, the upgrade will include data fed by large-scale language models (hmm… delicious jargon) that build on previous machine learning-based techniques developed for the Amazon advertising business. It contains new features that I used. Already used to extend the addressability of online advertising, these models will now also support predictive analytics and campaign planning.
The upgrade has been in development for a year and a half and is the first major overhaul of Amazon’s forecasting algorithms at the time.
But new machine learning advertising products no Amazon’s answer to Performance Max, Google’s black box AI buying tool, or Meta Advantage+ shopping campaigns.
According to Richter, other walled garden platforms with machine learning-based advertising products “withdraw control and transparency” from advertisers.
Amazon’s machine learning upgrades will become core DSP capabilities for all advertisers. In other words, product enhancements do not pull back campaign transparency or analytical reporting.
The ability to incorporate predictive transformation analytics into machine learning algorithms is a huge advantage for Amazon. Google’s PMax, to which Richter nods without naming names, and Meta’s Advantage+ both attempt the same trick of turning ad impressions into sales. However, these products include significant reductions in campaign management. For example, advertisers do not select media, bids, or creatives, and there is no transparent attribution. This is because Google and Meta maintain their own first party data and need to use it in combination. someone else’s purchase data.
Amazon DSP does not combine purchase data with anyone. Exactly Amazon.
“The backbone of our performance is how we actually measure ROAS and CPA in stores,” says Richter. “So to speak, it is directly connected to the cash register.”
And you don’t have to wait another 18 months for Amazon DSP to release its next machine learning-related news.
More announcements are expected this year, Richter said. “I consider this my first gun.”
