The new hot trend in platform advertising is to put the machines in charge of everything.
It’s true of Google, true of Meta and, even more so as of this week, it’s true of Amazon.
On Thursday, the Amazon DSP rolled out a system-wide upgrade to its machine-learning and predictive algorithms, Neal Richter, the group’s director of data science engineering, told AdExchanger.
Richter said the upgrades include new capabilities with data fed by large language models (Mmmm … that’s some yummy jargon), which are in turn built on prior machine learning-based tech developed for the Amazon ad business. These models were already in use to expand online advertising addressability, but will now also support predictive analytics and campaign planning.
The upgrades were in development for a year and a half, and this is the first major overhaul of Amazon’s predictive algorithm in that time.
But the new machine learning ad products are not Amazon’s answer to Performance Max, Google’s black box AI buying tools, or Meta Advantage+ Shopping Campaigns.
Other walled garden platforms with machine learning-based ad products “withdraw control and transparency” for the advertiser, Richter said.
Amazon’s machine learning upgrade will now be a core function of the DSP for all advertisers, which means the product enhancements will not pull back on campaign transparency or analytics reporting.
The ability to bake predictive conversion analytics into its machine learning algorithm is a big advantage for Amazon. Google’s PMax and Meta’s Advantage+, which Richter is nodding at without naming, are both attempting the same trick, which is to connect ad impressions to sales. But those products include drastic reductions to campaign controls. Advertisers, for example, don’t choose their media, bid prices or creative, and also have no transparent attribution. This is because Google and Meta must preserve their own first-party data while combining it with someone else’s purchase data.
The Amazon DSP isn’t combining purchase data with anyone. It’s simply Amazon.
“The backbone of our performance is how we measure ROAS and CPA, in real terms, on the store,” Richter said. “We have a direct connection with the cash register, so to speak.”
And don’t expect another 18-month wait before the Amazon DSP puts out its next piece of machine learning-related news.
More announcements will be coming this year, Richter said. “I think of this as just the starting gun.”
