Can you score ad creatives before spending on media?

The direct answer: yes, and it's fast

Yes — Kettio's pre-launch scoring runs your creative through a synthetic audience panel and returns a preference score and written rationale in under 30 seconds, with no media spend required. The scoring model is validated at ρ=0.78 Spearman correlation against University of Washington consumer survey panels (n=160 paired ads) and achieves 70.3% pairwise agreement with real CTR labels on a behavioral panel — well above the 50% coin-flip baseline. This is enough signal to cull losing creatives before any impression is purchased.

Why this question matters

Traditional creative testing requires live traffic. You run two creatives simultaneously, direct real paid impressions at both, wait for one to accumulate a statistically significant lead, kill the loser, and roll budget into the winner. The entire process costs money — the impressions on the losing creative are sunk cost — and it takes time, typically 7–14 days to reach actionable significance at modest spend levels.

Pre-launch scoring inverts the model. Instead of using real audiences in live media to generate preference signal, you use a calibrated model of your target audience to generate preference signal before any media dollar is committed. The tradeoff is fidelity: a synthetic audience model is noisier than a live behavioral A/B test. But it's free, instant, and good enough to prevent the most obvious mistakes — which are responsible for the majority of wasted creative spend.

What the validation numbers actually mean

ρ=0.78 Spearman correlation means that if you rank a set of ad creatives by their Kettio scores, that ranking will correlate strongly with how real human survey respondents rank the same ads by perceived persuasiveness. It's not a 1:1 mapping, but it's a consistent directional signal — creatives that score meaningfully higher in the model tend to be the ones humans prefer.

70.3% pairwise agreement with real CTR labels means: given two creatives, Kettio picks the higher-performing one 70.3% of the time based on actual click-through rate data from a behavioral panel. That 20.3-point margin above random chance is the budget you're protecting. On any given launch decision, you're right roughly 7 times out of 10 instead of 5 times out of 10.

This doesn't make pre-launch scoring a replacement for live testing — it's a filter that makes your live tests better by removing the obvious losers before they touch budget.

What most teams get wrong

The most common mistake is treating pre-launch scoring as a pass/fail gate with a hard cutoff. "Score above 7.5 = launch, below = kill" doesn't capture the nuance of how the model works. The meaningful signal is relative — the score of Creative A versus Creative B from the same campaign, against the same audience. Absolute scores vary by category, visual style, and audience archetype.

The second mistake is expecting the score to predict exact CTR. Pre-launch scoring predicts relative preference within a comparison set. It doesn't tell you "this creative will achieve 2.3% CTR." It tells you "this creative will likely outperform that one with this audience." Use it for ranking and culling, not for forecasting absolute performance metrics.

How Kettio's scoring works

Kettio uses a Semantic Similarity Ranking (SSR) architecture. When you upload a creative, the model builds a written rationale from the perspective of your target audience — describing what they notice, what resonates, what creates friction, and what motivates action. That rationale is then compared via embedding similarity to a set of anchor descriptions calibrated to your goal (purchase intent, click-through rate, brand recall, etc.).

The result is a continuous score that reflects how closely the creative's audience-perceived qualities match the qualities associated with high performance on your goal. The model doesn't rely on visual feature extraction alone — it uses language as an intermediate representation, which makes it sensitive to nuanced creative choices like copy tone, product framing, and benefit clarity that purely visual models miss.

You can use this for any static image, video, carousel, or UGC-style creative. No connected ad account required. See how the platform works for a full walkthrough.

The floor and ceiling of pre-launch scoring

Pre-launch scoring is strongest when the performance gap between creatives is large — when one creative is structurally better than the others (cleaner visual hierarchy, stronger hook, clearer CTA). It's weakest when you're comparing creatives that are nearly identical — the same product, same offer, slightly different copy or color. For fine-grained comparisons within the same stylistic envelope, live behavioral testing is more reliable. For culling a batch of 20 AI-generated variants down to a shortlist of 5, pre-launch scoring is the right tool.