Emotions That Convert

What it actually answers: which emotions drove ROI

A conversion probability per creative is useful. The thing underneath it is more useful. Because every prediction carries its regional and emotional contributions, you can roll the structured output back up across a campaign and ask the question creative teams have never been able to answer cleanly: which emotions brought the most ROI, for this brand, this campaign, this audience.

Not emotion in the abstract. Anticipation might be the lever that converts your cold prospecting audience while trust is what closes your retargeting pool. The creative that wins on a UK feed might win on a different emotional register than the one that wins in Germany. The pipeline makes that legible because the emotional signal is a typed parameter attached to a measurable outcome, not a strategist's hunch after the campaign is over.

That is the shift worth naming. Advertising now has to connect with people more intimately than it ever has. Feeds are saturated, attention is shorter, and generic creative is invisible. The brands that win are the ones that know which specific emotion moves their specific audience, and right now almost nobody can measure that. This is an attempt to.

Validation: backtesting against Meta hook and hold rates

The open question is whether the predicted neural response tracks real performance. There is academic precedent that it should. Falk and colleagues showed in PNAS that brain activity forecasts aggregate video engagement in an internet attention market, and a body of neuromarketing work has linked neural and attention metrics to ad recall, liking, and view rates. The thesis is that a neural readout carries signal about retention that exists before any spend.

The cleanest place to test this is paid social retention, where Meta already produces two metrics that are effectively behavioral analogues of the neural ones:

• Hook rate = 3-second video views ÷ impressions. Did the opening stop the scroll. The behavioral version of initial attentional capture.

• Hold rate = ThruPlays (15s or completion) ÷ 3-second views. Did the body keep them. The behavioral version of sustained engagement.

The design is a backtest, not a live experiment. Score a library of historical creatives through the pipeline to get region-level activations and a predicted conversion probability for each, then pull the actual hook and hold rates those same creatives earned on Meta and check for correlation. Specifically: do high early-attention activations predict high hook rate, and do sustained-engagement and valuation activations predict high hold rate. Because hook and hold isolate different stages of the same view, they let you test the neural prediction at two points instead of one, which is a sharper validation than a single conversion number.

If the activations track hook and hold, the pipeline is picking up real retention signal from the creative alone, before a cent is spent. If they do not, the backtest tells you exactly where the neural model and real behavior diverge.