Upwave opened a closed beta for Predictive Brand Lift, a measurement product that estimates how much a campaign moved brand awareness or consideration using Bayesian modeling trained on its own historical study data, rather than running a new consumer survey for every campaign.
The pitch is about the campaigns that never got measured at all. Traditional brand lift requires exposing a survey to both a treated and a control group, which only works at meaningful statistical confidence once a campaign clears a fairly high impression threshold. Upwave says its model, trained on results across more than 3,000 brands, can estimate what a survey would have shown for smaller campaigns that never reached that threshold. “We’re opening the door to a new generation of brand lift measurement,” said Chris Kelly, CEO of Upwave. “Historically, various constraints meant advertisers could only measure a fraction of their campaigns.”
The original insight here is what this does to measurement incentives, not just measurement coverage. When brand lift required a live survey, marketers had a practical reason to concentrate spend into fewer, larger campaigns that could justify the measurement cost. A predictive model removes that constraint, which means the real test of Predictive Brand Lift is not whether it agrees with survey based lift on the campaigns both can already measure, the same rigor question already facing the broader measurement stack as AI reshapes how consumers discover products. It is whether marketers trust a model’s estimate enough to greenlight smaller, more fragmented campaigns on the strength of it alone.
That trust question is not new to vendor supplied measurement. It is the same one raised by reach figures that a media owner reports about its own inventory before independent verification catches up. A model built on Upwave’s own historical panel is Upwave grading its own campaigns until an outside body validates the approach, which is a reasonable place to start a beta and a real limitation to track as it scales.
Source: PPC Land
