Performance Insights & Causal Attribution
Master the Impact Heatmap, TreeSHAP marginal lift calculations, Retention Drop-off curves, and the Element Performance Matrix.
Once ad creatives are analyzed and Meta performance metrics are ingested, Performance Insights (/impact-heatmap, /timeline-dna, /element-performance) answer the ultimate growth question: “Which exact creative traits drive our ROAS, and what should we produce next?”
1. Impact Heatmap (/impact-heatmap)
The Impact Heatmap isolates which visual and auditory elements correlate causally with performance gains vs. ad fatigue drag.
graph TD
Step1["1. Select Lookback Window<br/>(Last 7d, 14d, 30d, or Custom)"] --> Step2["2. Select Cohort Scope<br/>(All Campaigns, Top Prospecting, Retargeting)"]
Step2 --> Step3["3. Choose Target KPI<br/>(ROAS, Hook Rate, Hold Rate, CPA, CTR)"]
Step3 --> Step4["4. Set Decision Mode<br/>(Exploratory vs. Decision-Grade Gating)"]
Step4 --> Results["🏆 Run Heatmap & Generate Recommendations"]The 4-Step Setup Accordion
- Time Period: Choose 7, 14, or 30 days. Shorter windows capture immediate fatigue trends; 30-day windows provide deep statistical significance.
- Cohort Scope: Compare like with like (e.g. isolate Cold Prospecting Video Ads from Bottom-of-Funnel Retargeting Stills).
- Target Metric: Select your primary optimization KPI (
ROAS,Hook Rate,Hold Rate,CPA,CPC). - Intent Mode:
- Exploratory: Broad exploratory analysis surfacing emergent patterns.
- Decision-Grade: Strict Bayesian shrinkage filtering (p-value under 0.05 and minimum 10 creative occurrences) ensuring bulletproof confidence before allocating large production budgets.
2. Interpreting TreeSHAP Marginal Lift
AdHelix utilizes game-theoretic TreeSHAP (SHapley Additive exPlanations) to isolate the marginal causal contribution of an individual element, removing confounding variables like ad spend skew or seasonal baseline shifts.
| Classification | Lift Threshold | Strategic Action |
|---|---|---|
| 🏆 Brand Winners | +10.0% or higher (Vibrant Green) | Double down & scale: Replicate this talent, hook style, or lighting in your next batch of ad iterations. |
| ⚖️ Neutral Baseline | -10.0% to +10.0% (Muted Slate) | Maintain: The element supports the narrative without significantly altering conversion probability. |
| ⚠️ Brand Hazards | -10.0% or lower (Amber / Coral Red) | Pause or iterate: This element causes rapid audience drop-off or inflates CPA. Eliminate it from future creative briefs. |
3. Retention Diagnostics & Orienting Reflexes (/timeline-dna)
Meta ads are won or lost in the first 3 seconds. The Retention Diagnostics engine correlates viewer drop-off with micro-level scene events:
- 3-Second Hook Rate Benchmark: Dynamically calculated as:
Hook Rate (%) = (3-Second Video Plays / Total Impressions) × 100 - Retention Decay Curve Overlay: Visualizes exact millisecond drop-off slopes aligned with scene cut boundaries.
- Orienting Reflex Spikes: Highlights sudden retention upticks (+9% to +15%) triggered by pattern interrupts (e.g. sudden sound effect, macro product texture glide, rapid camera zoom, or provocative question).
4. Element Performance Matrix (/element-performance)
The Element Matrix aggregates detected entities across your entire historical ad account:
- Talent Leaderboard: Compare creator faces and demographics side-by-side to identify your highest-ROAS talent partnerships.
- Hook Archetypes: Compare UGC ASMR Swatch vs. Founder Explainer vs. Split-Screen Comparison across blended ad spend.
- Audio Cadence Matrix: Analyze whether high-BPM upbeat music or spoken conversational voiceover drives lower customer acquisition costs (CAC).
5. Creative Playbook & Ad Sequencer
- Creative Playbook (
/playbook): Aggregates top-performing winning patterns into an auto-generated Brand Production Brief ready to send to creator agencies and video editors. - Ad Sequencer (
/sequencer): Drag-and-drop recipe builder allowing strategists to simulate recombining winning Hooks (0–3s), Bodies (3–15s), and CTAs (15s+) to forecast expected ROAS lifts before rendering new video edits.