Publicis acquired AdgeAI. The announcement: "empowering brands to make only what works." Their website claims 94% accuracy and the ability to produce ad scripts with higher confidence. The ex-NASA computer vision and computational neuroscience are impressive and real.
The question brands should ask: where does the training data come from?
The architecture of the prediction
To predict that a new creative will perform well, the model needs historical examples of what good performance looks like. At the scale and speed this operates, the only available source of that data is platform APIs. Meta's attribution system. Google's conversion tracking.
The follow-up question: if the model trains on platform-reported attribution, and the prediction is later validated against the same platform's reporting, is that independent validation?
It is not. It is a confidence loop. The same data that taught the model what works will later confirm the model was right.
Why the loop is not a small problem
On platforms where attribution overstatement has been documented at 30% or more, that loop amplifies the distortion. You may be optimising creative for what the platform can claim credit for, not what drives incremental sales.
Independent measurement of creative performance requires a measurement layer that sits outside the platforms being measured. That is the only way to break the loop.
When the entity validating your creative performance is the same entity that trained the prediction model, what exactly is being validated?
#OwnYourIntelligence
