AI models and machine learning have reshaped ad campaign measurement.
It’s now commonplace for advertisers to set up campaigns with a budget and a cost-per-acquisition goal on all the walled gardens and on third-party programmatic platforms, with AI modeling its own results.
The concept of campaign analytics has mostly folded and been replaced by “insights” based on AI-modeled platform data. For example, campaign analytics might include 945 advertising IDs that were targeted in the campaign and on which sites. That analytics reporting can be exported and added to a brand’s own CRM, even matched to specific profiles.
But insights reporting includes details like whether a campaign over-indexed to certain cities or demographics, such as young women. It might combine multiple first-party data sets, so Ticketmaster and McDonald’s, say, might analyze a group of people who attend certain games or concerts and eat regularly at the burger chain. It’s all very useful, but the insights can never leave the platform, so the data can’t be attached to a CRM profile or actual known customers.
Since AI solutions rely on attribution and analytics to fuel their insights, it’s time to push back on the narrative that AI is usurping campaign measurement. Instead, think of attribution as the auditor and delegator for AI models in advertising.
At least according to Paula Despins, VP of measurement for Amazon Ads, Amazon Marketing Cloud and ads trust, who penned a blog post on the topic and also recently spoke to AdExchanger.
There is no denying there’s a shift underway that puts AI behind the wheel, so to speak, for campaign targeting and optimization, and even generating ad creative. There has also been a shift from campaign “reports” to AI “insights.”
“When I talk to advertisers, they want a level of insight and transparency that helps them still have important levels of control,” Despins told AdExchanger.
The AI modeling of data science analytics within the Amazon Ads platform is “simplifying very rapidly,” she said. And the insights tab is populated with more useful information at a faster clip than ever before. These reports “would have taken a data scientist or a high-skilled analyst previously,” she added, but they “are now accessible to everybody.”
The rise of AI modeling and the “insights” tab have also spurred new metrics for marketers.
Amazon Marketing Cloud (AMC) released a metric, “long-term sales,” which creates a cohort of new-to-brand customers that were gained in a particular campaign. The metric estimates a dollar value associated with converting those new customers for the next year.
That modeling is possible because an advertiser can use AMC to examine prior years’ worth of sales and ad campaigns. The advertiser can then forecast how much a customer who was acquired through one type of ad, at a certain cost, will spend in the next 12 months.
The type of outcome can be adjusted as well. For example, a brand could create a cohort of potential customers who added something to a cart or visited a product detail page but didn’t buy. It could then model how much that type of person might spend on the brand in the next 12 months.
However, Amazon can’t allow that data to be exported and thus its shoppers to be retargeted willy-nilly.
To gauge whether reality catches up with the long-term sales projection, Amazon also has a new benchmark: “accumulated sales.” This metric tracks the actual purchases from these cohorts of new customers (or near customers). So the accumulated sales should catch up to the long-term sales over the course of a year.
Despins said more specialized metrics will be added to insights modeling, such as a metric for automotive brands that want to forecast rates of dealership visits or signups for test drives.
Amazon isn’t alone in expanding attribution-based insights modeling. The grocery chain Albertsons, for instance, began adding lifetime value forecasts to its campaign insights. But those types of forecasts, even when cited confidently by an AI model with a specific dollar value attached, are still guesses. The modeling is more like a storytelling metric to support the marketing org, rather than a true deterministic measurement.
In other words, the CFO will believe it when the numbers show up on the bottom line, not in the insights tab of a marketing dashboard.
So, rather than letting AI modeling take over completely, Despins said that marketers should think of their campaign measurement strategy as a critical part of their overall AI management and oversight.
“[Measurement] acts as an important guidepost both for the AI itself,” she said, “but also for the marketer who is engaging with AI to build their business.”
