Reference
Every metric Cortexa surfaces, with its marketer-friendly display name, a plain-English definition, and the technical name used in our API.
The Vanity Rule
High engagement on its own is the most common metric for which marketers over-optimize. Cortexa surfaces this rule directly on every results screen — the engagement number renders a stoplight badge so you know when it's meaningful and when it's noise.
Of viewers who see the ad, the share who click.
Range · 0–1 (rendered as %)
How we compute · Predicted clicks / predicted impressions; calibrated against the demographic cell.
When it matters · Matters most for direct-response. Ignore for pure brand campaigns.
Predicted cost to reach 1,000 people.
Range · USD
How we compute · Platform-specific bid model conditioned on objective + audience cell.
When it matters · Matters for reach-bid campaigns. Ignore when optimizing for conversions.
Predicted revenue per $1 of ad spend.
Range · × (multiplier; 1.0 = break-even)
How we compute · Downstream-conversion-rate × predicted basket value / predicted CPM.
When it matters · The north star for e-commerce. Ignore for awareness.
Of clickers, the share who complete the goal you set.
Range · 0–1 (rendered as %)
How we compute · Predicted goal completions / predicted clicks within the attribution window.
When it matters · Matters once traffic is qualified. Ignore if your landing page is the bottleneck.
Average times a single viewer sees the ad.
Range · ≥ 0 (count)
How we compute · Predicted impressions / predicted reach over the campaign window.
When it matters · Watch when above 3 — fatigue risk. Ignore at low budgets.
A composite of fixation duration + sustained look + post-impression recall, calibrated to eye-tracking ground truth.
Range · 0–1 (calibrated; 0.5 ≈ industry median for paid social video)
How we compute · Brain-probe v13 features → trained linear head against IAS-style attention labels.
When it matters · Matters for video-heavy formats. Ignore for static placements below the fold.
Average milliseconds a viewer spends with the ad before scrolling away.
Range · ms (typical paid-social range: 800–4000)
How we compute · Hold-pillar score → linear regression against measured platform dwell data.
When it matters · Matters in feed. Ignore in pre-roll (which is forced viewing).
Share of impressions where the viewer watches past the 3-second mark.
Range · 0–1 (rendered as %); benchmark: 0.25 = solid on TikTok, 0.10 = struggling
How we compute · Hook-pillar score → calibrated to in-product retention curves on KuaiRand.
When it matters · Always matters for paid social video — platform algorithms use this to decide distribution.
Share of feed impressions where the viewer stops scrolling long enough to count as a view.
Range · 0–1 (rendered as %); platform views: Meta 2-3s, TikTok 6s
How we compute · First-frame attention prediction × feed-format conditioning.
When it matters · Matters for Reels / TikTok / Shorts. Ignore for search and display.
Likes + comments + shares + saves per impression.
Range · 0–1 (rendered as %)
How we compute · Sum of predicted social actions / predicted impressions; gated by the Vanity Rule.
When it matters · Conditional. Only worth maximizing IF Click-Through is high AND Downstream Conversion is low — otherwise it is a vanity signal that does not move revenue. Cortexa surfaces this rule directly on the result screen.
Predicted increase in unaided brand recall + favorability among ad-exposed viewers.
Range · 0–1 (calibrated to survey-based brand-lift studies; 0.15+ is meaningful)
How we compute · Brand-lift pillar score × creative-brand-asset matcher.
When it matters · Matters for awareness and consideration. Ignore for performance / direct-response.
Share of viewers who reach 25 / 50 / 75 / 100% of the video.
Range · Four 0–1 values (rendered as %)
How we compute · Hold-pillar curve sampled at the four duration quantiles.
When it matters · 75% and above is the gold-standard marker for video ads. Ignore if your asset is under 6 seconds.
Of viewers who saw the ad, the share who convert within the attribution window.
Range · 0–1 (rendered as %); 7-day post-impression attribution window
How we compute · Algorithmic-fit × hold pillars, calibrated to platform conversion APIs where connected.
When it matters · Always matters when conversion is the goal. The most honest single metric we surface.
Share of impressions actually rendered on-screen long enough to count (MRC standard).
Range · 0–1 (rendered as %); MRC: 50% pixels for 1s+ display, 2s+ video
How we compute · Static format-conditional baseline; not from the brain-pattern model.
When it matters · Matters for programmatic display. Ignore for owned-channel content where viewability is implicit.
Viewers who saw the ad (without clicking) and converted within the attribution window.
Range · 0–1 (rendered as %)
How we compute · Downstream-CVR × (1 − CTR), conditioned on attribution-window length.
When it matters · Matters for awareness with measurable lift. Ignore on platforms with poor cross-device attribution.
Every variant gets a harmonic mean of four pillars. Harmonic — not arithmetic — so a single failing pillar drags the score down. This is what keeps Cortexa from rewarding creatives that only win on a single axis.
How effectively the creative captures attention in the first 1–3 seconds.
How effectively the creative retains attention through the body.
How well the creative aligns with the distribution behavior of the chosen platform.
Predicted lift in recognition and favorability attributable to the creative itself.