Everything Patternix decodes, monitors, and predicts.
A complete reference for every capability in the platform — signal definitions, model outputs, delivery states, integrations, and what's coming next. If you need to explain it to an engineer, a media buyer, or your CMO, it's here.
Every ad decoded the same way, every time.
When a creative enters Patternix, this is what happens to it.
00:07 / 00:219:16 · 1080phm_cp302_v3.mp4- L1buyingactive
- L2lifecyclein test· spend below threshold0.94
- L3andromedaaudience search· cluster unstable0.81
Four categories. Every field documented.
Every signal is a named, typed field — not a proprietary score. You can filter by it, search by it, and brief from it. Below is the current field set; new signals ship with model-card entries on the same cycle.
formatUGC talking-head, static image, screen-record, meme overlay, animated.
product_shotTimecode range where the product or logo appears on screen.
paletteDominant colour family: warm-neutral, high-contrast, pastel, oversaturated.
pacingCut rate: slow (>3 s/cut), medium, fast-cut (<1.5 s/cut), erratic.
visual_complexityScene-change frequency and on-screen element density.
ai_probProbability that the visual track is AI-generated (0.0 – 1.0).
orientation9:16 portrait, 16:9 landscape, 1:1 square, 4:5.
resolutionNative resolution at upload: 1080p, 720p, 4K, etc.
text_overlayWhether burned-in captions or headline text are present.
brand_logo_visibleWhether the brand logo or wordmark appears at any point.
L1 buying ·
L2 lifecycle ·
L3 Andromeda.
Every creative gets three independent status readings, updated daily: what the ad account says, what the trajectory model says, and what Meta's retrieval algorithm is doing. Each layer has its own confidence score and reason code.
Full Lifecycle Monitor docsDay-0 score. Day-3 verdict. Day-14 grade.
The Forecast Engine generates a probability-to-scale score on every new creative the moment it's decoded. The score refines as delivery data arrives, settles into a scale / wait / cut verdict by day 3, and is graded against reality at day 14.
Full Forecast Engine docsWhat the patterns across your whole library reveal.
Winner neighbors
Every winner's five nearest semantic neighbors — similar hooks, similar voices, similar formats. Shows you what the algorithm already thinks are duplicates before suppression kicks in.
Concept families
Ads grouped by their assigned concept code, then ranked by winner rate within each family. Instantly see which concepts have legs and which are one-hit wonders.
Period-shift detection
Win-rate and fatigue-velocity trends by signal value, compared period-over-period. Catches when 'urgent female voice' stops outperforming or 'fast-cut pacing' starts fatiguing faster.
Plugs in. Pipes out.
Built to pass your security review.
Your data stays yours
Creative assets are processed in your isolated environment and never used to train models for other customers. Embeddings are stored in a dedicated vector store — your index, your data.
Public model card, every version
Every Forecast Engine release ships with documented AUC, training volume, calibration curves, and known failure modes. Accuracy is published before marketing is written.
Enterprise security
SOC 2 Type II in process. SSO/SAML enforced org-wide. Full audit log queryable via API. Data residency in EU or US, contractually guaranteed.
Where the platform is going.
Dates are targets, not promises. Ship order may change; the model card policy does not.
- Lifecycle Monitor GA
- Forecast Engine beta
- Triple Whale + Meta API
- Creative Intelligence Suite
- Multi-brand workspaces
- Semantic search
- TikTok Ads connector
- Visual clustering GA
- Pattern Library
- Google Ads connector
- Automated brief generation
- Custom taxonomy API