Aggregated ad metrics are observational unless the exposure was randomized. Use only authorized first-party data, keep client datasets isolated, account for sample size and conversion/return lag, and reconcile platform reporting to contribution economics. A pattern is a candidate for a controlled, reversible test—not a formula.
Creative DNA
Ad Metrics
The Two Dimensions
Creative attributes meet ad network metrics. This is where qualitative meets quantitative.
This turns "this ad had a result" into "this pattern may be worth testing." It organizes evidence without promising that a formula will repeat.
Until now, you've been building two separate bodies of knowledge. Dimension 1 is qualitative — the creative attributes you've extracted from AI vision analysis. Dimension 2 is quantitative — the hard numbers from your ad accounts. The breakthrough happens when you cross them.
Dimension 1: Creative Attributes (Qualitative)
These are the elements you've been tagging since Part 4 and analyzing with AI since Part 5:
- Hook type — question, transformation, social proof, scarcity, curiosity, shock
- Visual style — lifestyle, product-only, UGC, studio, flat lay, motion graphics
- Copy angle — pain-point, aspiration, benefit-led, feature-led, story
- Emotional trigger — fear of missing out, aspiration, trust, urgency, belonging
- Format — static, carousel, video 15s, video 30s, video 60s, Stories
- CTA type — Shop Now, Learn More, limited-time, soft CTA, no CTA
These come from your AI vision analysis (Part 5) and manual tagging.
Dimension 2: Ad Network Metrics (Quantitative)
These are the performance numbers pulled directly from Meta Ads Manager, TikTok Ads, or any other platform:
- CTR — click-through rate (attention signal)
- CPC — cost per click (efficiency signal)
- CPM — cost per 1,000 impressions (reach cost)
- ROAS — platform-attributed revenue per ad spend; not profit or incremental return
- CPA — cost per acquisition (customer cost)
- Frequency — average times shown per person (fatigue indicator)
- Conversion rate — percentage who purchase after clicking
- AOV — average order value (customer quality signal)
The Power: CROSSING Them
Most brands ask: "Which ad has the best ROAS?" That's a Dimension 2 question — useful but limited to a single creative.
Cross-dimensional analysis asks: "Which hook type is associated with higher lag-adjusted outcomes among comparable creatives?" That is an observational pattern to validate, not a scaling instruction.
Single-dimensional: "Ad #47 has 4.1x ROAS" — great, but when that ad dies, the insight dies with it.
Cross-dimensional: "In this illustrative sample, transformation hooks have higher attributed ROAS across 12 creatives" — a hypothesis whose uncertainty and confounders must be tested.
Building the Cross-Analysis Matrix
Rows are creative attributes. Columns are performance metrics. The intersections reveal everything.
This matrix makes associations visible. It does not map every design choice directly to revenue; audience, offer, placement, season, spend, and measurement can confound the result.
How to Structure the Data
Think of it as a spreadsheet. Each row is a creative attribute category (hook type, visual style, format). Each column is a performance metric (ROAS, CPA, AOV). Each cell is the average performance of all creatives that share that attribute.
How to Collect This Data
- Tag each creative in a spreadsheet with its attributes: hook type, visual style, copy angle, emotional trigger, format, CTA type (use your Part 4 framework)
- Export performance data from your ad accounts — Meta Ads Manager has a built-in export, or use the API
- Merge the two datasets — each row should contain both the creative tags AND the performance metrics
- Pivot — create pivot tables that show average performance by each attribute category
There is no universal minimum number of creatives. Set the required sample from baseline rate, minimum detectable effect, variance, allocation, and the metric's conversion/return lag. Report uncertainty and avoid pooling unlike objectives, audiences, placements, or time periods simply to increase row count.
Decisions From Cross-Data
Real insights that change real strategy. Each one backed by your own numbers.
Each of these insights can change your entire creative strategy. Combined, they compound into a significant competitive advantage that grows stronger with every creative you run.
These illustrative statements show how to phrase a hypothesis. They are not client results or default decisions; each needs adequate samples, uncertainty, comparable conditions, and a follow-up test:
Each observation above is a decision hypothesis. Confidence comes from a suitable design, adequate sample, uncertainty interval, delayed outcomes, and replication—not from the matrix alone.
Multiple replicated insights can improve a decision model, while interactions also increase complexity and overfitting risk. Validate combinations prospectively instead of treating layered correlations as a competitive guarantee.
Monitoring Possible Creative Fatigue
Use time-series evidence to investigate deterioration before making a reversible change.
Maintaining replacement candidates can reduce response time, but no schedule guarantees uninterrupted performance. Diagnose the cause before attributing a decline to creative fatigue.
Most brands detect fatigue after it happens. CTR drops. Frequency spikes. ROAS craters. By then, you've already wasted budget and lost momentum.
Cross-dimensional data can help you flag a possible change. It cannot identify fatigue—or its future date—without ruling out auction, offer, audience, tracking, inventory, and landing-page changes.
The Monitoring Framework
Use several contextual variables when investigating a change:
- Audience size — how many people can this ad reach?
- Daily spend — how fast are you burning through that audience?
- Historical frequency curves — at what frequency did past creatives start declining?
Practical Approach
- Track frequency curves of past winners — at what frequency did CTR start declining? This is your historical fatigue point.
- Use history as context, stratified by objective, audience, placement, offer, and format. Do not assume a new creative shares an old threshold.
- Monitor exposure and outcomes together — compare frequency with CTR, conversion quality, margin, auction conditions, and measurement health.
- Set an investigation alert from your own baseline and false-alert tolerance. A human decides whether to hold, test a replacement, or fix another cause.
Track response curves separately by format and context. Do not impose universal day ranges: lifespan varies with audience, spend, placement, objective, offer, season, and creative quality.
Practical Setup
Five steps. One spreadsheet. 30 minutes. The ROI is massive.
Start with a small pilot and measure the actual setup and maintenance effort. The matrix is useful only if tags are verified, data is authorized, and its decisions improve a pre-defined business outcome.
Recommended Tools
| Approach | Tools | Best For |
|---|---|---|
| Simple | Google Sheets + Gemini/Claude for analysis | Most brands, easy to start |
| Intermediate | Google Sheets + Apps Script automation | Brands running 50+ creatives/month |
| Advanced | Direct API + Python/AI pipeline | Agencies, 100+ creatives/month |
A maintained matrix can preserve institutional learning. It does not get smarter automatically: audit tag quality, permissions, missing data, drift, and whether its recommendations replicate in controlled tests.
Built by @itsmazinzaki — AVAMARTECH