Following an Upfronts dominated by conversations about outcomes and optimization (and EDO’s metrics), it’s no surprise that many in our space are following our lead in prioritizing outcomes-driven optimization. Imitation is flattery, right?
As AI commoditizes mere data observations and enables autonomous optimization, marketers and media companies are demanding to buy and sell Convergent TV ads with the same accountability, performance, and ease they expect from digital. But while AI is absolutely transforming how we measure and optimize TV advertising, there’s one thing it can’t do: turn bad data into investment-grade insights.
That’s because no matter how powerful our algorithms are, a machine learning model is only as good as the data that powers it. If a predictive model is trained on synthetic data – data that doesn’t reflect current reality and actually predict future results – machine learning won’t do anything besides multiply the amount of bad intel you’re getting. In other words: garbage in, garbage out.
Nowhere is this aphorism more applicable than to the raft of companies that are using AI to optimize TV advertising based on survey data — real and even simulated. Or at least that’s what they’re trying to do.
Cheap and cheerful survey data from a few hundred survey respondents should be suspect to begin with. We all lie to ourselves, and especially to surveys. Synthetic survey responses just serve to magnify and amplify those biases. If you really want to know how a consumer is going to respond to an ad, you need to watch what they do — not what they say.
Our memories are imperfect. Our mood changes throughout the day. Our stated intentions don’t always align with our actions. And when asked why we picked one option over another, we often tell a story that sounds logical, even if it isn’t what drove our decision. Sometimes, we simply choose the survey response we think the interviewer will like best.
And all of this assumes survey respondents are even trying to be honest, as opposed to racing through the questions to earn points toward a gift card. It also assumes surveys are designed without bias, rather than tilted to tell clients what they want to hear and win repeat business. A faster, cheaper solution for collecting and analyzing faulty data doesn’t get you any closer to smarter decision-making. It just gives you more reasons to make bad decisions.
And yet, it seems like there’s a breathless new story each week about some new innovation using AI to mimic human survey respondents. Because these companies’ so-called “digital twins” are always available, founders claim their services are faster and cheaper than traditional market research. But at what cost?
Even traditional market research — with responses from real human beings who buy real products! — is woefully flawed when it comes to predicting how people will actually respond to a new product or novel ad creative in the wild. Focus groups in the early 2000s would have asked for bigger keyboards on their Blackberries (which is what RIM actually gave them) instead of touch screens (which is what Steve Jobs insisted on – screw the research!).
Rather than using AI to photocopy faulty data, Vertical AI allows us to analyze scaled intent signals created through real, observed human behavior — things like brand searches, LLM chats, and app usage — that are proven to predict future sales. The output of this analysis is investment-grade data that serves as a reliable source of truth across the Lumascape. From there, our increasingly AI-driven models enable us to mine for actionable recommendations faster than any team of human experts ever could working alone.
This enables decision makers to discern which media placements, creative elements, and strategic choices are driving results, and which ones ought to be abandoned. And now, AI agents can automate this process, turning recommendations and adjustments into our final deliverable. The result? A cycle of continuous optimization based on real, in-market testing, delivering greater consumer engagement and stronger performance with each passing campaign.
At EDO, we know settling for imprecision can be expensive. Spotty data that leads marketers astray can cost brands millions in lost revenue. That’s why we focus on real consumer behavior, the only true signal of future sales.
Watch what people do, not what they say.