Frequently asked questions

Omniscia scores ad creatives before launch. You upload a video ad set and the platform reads visual diversity, hook structure, audio patterns, narrative angles, and format mix, then tells you which creatives are too similar, which categories are weak, and what to fix.

It works across Meta, Google Ads, and TikTok. Every analysis feeds Cortex, the intelligence engine that correlates creative scores with real ROAS, CTR, and CPA from linked campaigns. Helix, the five-layer trend model, reads emerging pattern shifts across the network and compares them with ad-manager data as it arrives. Every analysis, every linked campaign, and every user sharpens the model for everyone.

MP4, MOV, AVI, and WebM. Videos can be any length. Frames are extracted more densely in the first 6 seconds (the hook zone) since that drives most of ad performance. No minimum or maximum duration.

Upload every creative you plan to run together in the same campaign or ad set. The minimum per analysis is 2 files. The maximum depends on plan: 5 on Trial, 15 on Starter, 25 on Pro, 30 on Agency. Research puts the optimal range at 5 to 15 creatives; fewer than 5 limits a platform’s ability to find winners, more than 15 to 20 causes creative dilution. For most advertisers, 8 to 12 diverse creatives is the sweet spot.

Most analyses complete in 1 to 3 minutes depending on the number and length of videos. Frame extraction and AI vision analysis are the slowest steps. All plans get priority processing.

Your ad set is scored across 6 categories: Visual Diversity, Hook Variety, Angle Coverage, Format Mix, Audio Diversity, and Text & CTA Variety. Each category scores 0 to 10. The weighted average gives your overall score. Default weights are 25, 20, 20, 15, 10, 10 percent.

Defaults are a starting point. As users link real campaign data, Cortex learns which categories actually correlate with ROAS and replaces the defaults with empirical weights. Visual fingerprinting uses CLIP 512-dim embeddings to compare creatives at a perceptual level, not just pixel similarity. Your score is also ranked against the network, so you see your percentile position (e.g. “Top 22%”). Scores of 7.0 and above are considered good. Below 5.0 means significant clustering risk.

Effective creatives is the count of truly unique ads in your set after removing near-duplicates. Meta’s Andromeda algorithm assigns a Creative Similarity Score to every ad in your account. When it detects ads that share visual style, hook structure, or audio patterns, it groups them under a single Entity ID. Clustered ads compete for the same audience segment, so you pay multiple times to reach the same people. Ads with a similarity score above 60% trigger delivery suppression. You could launch 20 creatives and see only 5 or 6 get meaningful impressions.

For Meta analyses, Omniscia estimates how many Entity IDs Andromeda is likely to assign your ad set based on hook-fingerprint similarity and visual clustering inside your upload. The Cluster Risk panel shows the estimated Entity ID count, hook fingerprint analysis (Andromeda weights the first 3 seconds disproportionately), branch coverage (which creative dimensions you’ve covered vs. where you have gaps), and a Readiness Score that measures alignment with Andromeda’s delivery requirements.

The similarity matrix compares every pair of videos across visual, audio, and hook attributes. Each cell shows a combined similarity score from 0 (completely different) to 1 (nearly identical), weighted 45% visual, 30% audio, 25% hook.

Pairs above 0.7 are flagged red. Those videos will likely be clustered by ad platforms, competing for the same audience instead of expanding your reach. The matrix also explains why each pair is similar. Click any cell for a detailed side-by-side comparison.

Every Lens analysis can include your landing page URL. Omniscia scrapes the page for offers, urgency signals, trust indicators, social proof, and CTA verbs, then scores how well your ads align across 5 dimensions: headline match, offer alignment, CTA verb alignment, urgency alignment, and trust signal coverage.

If your landing page has a “20% off first order” offer but none of your ads mention it, you’ll know. Each analysis creates a persistent snapshot, so you can track how alignment changes over time. Ad-to-LP alignment is one of the tighter correlations in Cortex data: when the ad and page don’t tell the same story, post-click conversion usually takes the hit.

Each recommendation tells you what’s wrong, which video is affected, how to fix it, and the expected score shift. When Cortex has enough data, recommendations also show the ROAS change observed on linked accounts that fixed the same category. A reference point from real data, not a promise about your result.

Recommendations are ranked using a Thompson Sampling multi-armed bandit. When Cortex data is available, the bandit prioritises fixes with the strongest historical impact on linked campaigns, not just the largest score gap. Start with critical recommendations (red): those are similarity flags that indicate wasted budget. Then warnings (yellow) and suggestions (blue). Re-upload your revised set and compare scores. Aim for 7.0 or above before launching.

Cortex is Omniscia’s intelligence engine. It aggregates data across 4 tiers: platform metrics (ROAS, CTR, CPA from Meta, Google Ads, TikTok), creative analysis outputs (scores, attributes, similarity), ML model outputs (fatigue risk, trend analysis, weight calibration), and external signals (market trends, competitor activity). Each linked campaign contributes roughly 80 data columns to the model.

Cortex retrains weekly. Every user’s anonymised data sharpens the model for everyone on the network. Your percentile benchmarks are computed against the full Cortex dataset. The Cortex dashboard shows every learning event, weight adjustment, and correlation discovery. Data quality safeguards include minimum $50 spend, 7-day minimum duration, outlier trimming, SHA-256 deduplication, and statistical significance filters.

After 30 or more linked campaign-analysis pairs, Cortex switches to personalised weights: 70% your data, 30% global. At that point, scoring, recommendations, and trend analyses are calibrated to your specific audience and creative patterns.

Helix is the 5-layer trend-analysis engine behind Signal. It’s not a generic time-series model. Each layer adds a distinct capability:

1. Quality weighting: prioritises high-reliability data sources.
2. Adaptive smoothing: filters noise without lagging real shifts.
3. Lifecycle curves: tracks how trends mature, peak, and decay.
4. Cortex correlation: cross-references trend output against real ad performance.
5. Macro context: factors in seasonality and market-level patterns.

Helix flags emerging patterns in the network and compares them with ad-manager data as it arrives. Trends where the model’s reading aligns with real Cortex-linked campaign data are marked separately, so you can weight what’s supported by real performance over what’s still pattern-level signal.

Signal is the intelligence pipeline. It monitors 27 curated industry sources (industry reports, platform announcements, market research, competitor moves) and distils them into actionable insights. Each insight is classified into one of 6 types: trend shift, market movement, platform update, competitive intelligence, creative pattern, and audience behaviour.

Multi-source consensus means Signal only surfaces insights confirmed by 2 or more independent sources. Source reliability is tracked over time, so sources that produce consistently accurate signals get weighted higher. Helix powers the trend-analysis layer, and Cortex-validated trends are cross-referenced against real ad performance data so you can separate signal from noise.

Omniscia reads the patterns that usually run ahead of fatigue (frequency creep, CTR decay, CPM inflation, engagement drop-off) and flags at-risk ads early. The model is a Random Forest (20-stump ensemble) trained on historical fatigue events across Meta, Google Ads, and TikTok. On accounts with enough linked data, flagged ads typically show clear fatigue inside 5 to 7 days, giving you room to prepare replacements before performance drops. A pattern read, not a guarantee.

On top of that, survival analysis (Kaplan-Meier curves and Cox proportional hazards) estimates ad longevity: how many days of runway a creative has left before fatigue risk crosses the threshold. Works across Meta, Google Ads, and TikTok. Every fatigue event on the network feeds back into Cortex and sharpens the model for everyone.

When you publish an ad through Launch, Omniscia suggests budget and duration ranges based on your creative scores, historical performance on linked accounts, and Cortex pattern analysis. Budget priors use a Thompson Sampling multi-armed bandit: each bracket carries a probability distribution from observed outcomes, so suggestions are ranked by real historical performance, not rules of thumb.

Each creative needs roughly $50 and 7 days to produce a reliable read. Suggestions are structured around a platform’s learning phase (around 50 conversions per week) and flag clustering before you put budget behind ads that would cannibalise each other. When Cortex data is available, budget suggestions are calibrated to your account’s specific performance patterns.

Forge is the creative direction engine. It generates data-backed briefs from your Lens scores, Signal trends, and linked campaign outcomes, synthesising Cortex insights, audience segments, and Signal trends into actionable creative directives with clear rationale. Rather than generic templates, Forge produces direction grounded in your specific performance data. Forge feeds structured context to the in-product chat assistant, so recommendations stay informed by fresh creative direction. Available on all plans.

Lens is the creative library and ad set composition surface. It holds your analysis history, uploaded creative assets, folders, search, score sorting, exports, and shared report links. Use the Videos tab to inspect individual creatives, and the Ad Sets tab to compare compositions before launch. Available on all plans.

Yes. Scia is the context-aware chat assistant built into every page, available on all plans including the free trial. It reads your current view (the page you’re on, what’s selected, the analysis or campaign in front of you) and your full account history (Lens scores, Forge briefs, Cortex insights, linked campaign data), so its answers are grounded in your specific data, not generic advice. Scia is a feature of the platform, not a separate product. Ask it things like “What should I change in my next ad set?” or “Which of my angles perform best on TikTok?”

Nexus unifies Meta, Google Ads, and TikTok into a single Cortex model. Creative fingerprinting matches the same ad across platforms automatically using perceptual hashing and CLIP 512-dim embeddings. If you run the same creative on Meta and TikTok, Cortex recognises it as one asset and correlates performance from both.

Your scores, fatigue-risk flags, and recommendations draw on cross-platform data. A creative fatiguing on Meta might still have runway on TikTok, and Cortex catches that. One intelligence model, three platforms, no manual mapping.

Yes, and it’s where the platform compounds. Connect Meta, Google Ads, or TikTok and Omniscia correlates your creative scores with real ROAS, CTR, CPC, and spend data. This feeds Cortex, so category weights adapt to what actually moves performance for your audience, recommendations are ranked by historical impact on linked accounts, fatigue-risk detection compares new ads against your own fatigued creatives, and trend analyses calibrate to your account.

Meta, Google Ads, and TikTok account connections are available across plans. After 30 or more linked campaign-analysis pairs you unlock personalised Cortex weights tailored to your specific performance patterns. Connect from Nexus.

Yes. Agency includes full API access (1,000 req/hr) so you can pull scores, reports, fatigue data, trends, and competitor intelligence into your own tools. API keys support read-only and read-write scopes for programmatic analysis submission, campaign management, and chat-assistant queries.

Authenticate with an API key (generated from your Profile page) via the X-API-Key header. Keys support separate read and write scopes, and you can generate up to 10 active keys per account.

Each plan includes monthly credits. Credits are spent on video analysis (10 credits each), image analysis (5 credits each), and Intel deep analysis (10 credits each). Everything else (Forge, Signal, ad set compositions, Launch, the in-product chat assistant) is unlimited. Credits reset each calendar month.

Trial: 500 credits. Starter: 500. Pro: 1,500. Agency: 15,000. If you run out, you can buy top-up packs or overage credits at your plan’s rate, or upgrade for more included credits.

Trial (free, 14-day Starter trial, no card required): 500 credits, 15 ads/set, 5 Intel brands. Full Starter access during the trial: every product unlocked, including Google Ads, TikTok, Slack, Cortex Simulator, Scheduled Reports, and Priority Processing.

Starter ($99/mo): 500 credits/month, 15 ads/set, 5 Intel brands. All features unlocked.

Pro ($299/mo): 1,500 credits/month, 25 ads/set, 12 Intel brands. All features unlocked.

Agency ($1,499/mo): 15,000 credits/month, 30 ads/set, 40 Intel brands, 15 seats ($79 per extra). All features plus enterprise: team sharing, white-label, API, bulk upload, dedicated support, auto top-up.

Annual billing saves 20% across all paid tiers. Full breakdown on the pricing page.

Team collaboration is available on Pro and Agency. Create a company from your Profile page and invite team members with granular roles: Owner (full control), Admin (manage members, approve reports, configure white-label), Editor (run analyses, submit for review, comment), Viewer (read-only access).

Editors can submit reports for approval before they go to clients. Admins and owners review from the Approvals queue. Anyone on the team can comment directly on reports. You can also share individual reports via a public link. Agency includes 15 seats ($79 per additional seat).

Agency users can rebrand Omniscia’s reports with their own identity: brand name, logo, primary colour, footer text. These apply to shared report pages, PDF exports, and custom reports. You can optionally hide all Omniscia branding entirely, so everything looks like it came from your own platform. Built for agencies that share reports with clients.

Yes. Go to Profile > Change Plan to upgrade, downgrade, or cancel at any time. Upgrades take effect immediately: you pay the prorated difference and get a proportional share of the new tier’s credits. Downgrades and cancellations are scheduled for the end of your current billing cycle. Refunds are not available; see our Refunds Policy for why. We recommend the 14-day free trial first, and starting on a lower tier when you do subscribe.

Your uploaded content is encrypted and stored on secure servers, used only for your analysis and playback. Only you (and your team members on a team plan) can view your uploads. No Omniscia staff have access to your raw creative files. Content can be deleted at any time from Lens, and deletion is permanent and immediate. We never use your video or image content for training, and we never share it with third parties.

When you link ad performance data to an analysis, only the AI’s analytical outputs are stored separately to train Cortex: things like classified creative attributes (“visual style: UGC”, “hook type: question”) and aggregated performance metrics. No video frames, audio, images, transcripts, or personal information is ever included. This anonymised data persists after account deletion to preserve model integrity but contains zero identifiable content.

AI providers (Anthropic Claude, Google Gemini, OpenAI Whisper) process your content via zero-retention API calls; they do not store or train on your data. For full details, see Trust & Security and Privacy Policy.

Most tools tell you what happened. Omniscia tells you what’s happening: the creative that’s fatiguing, the audience you’re doubled up on across Meta and TikTok, the score that’s drifting while CPA still looks fine. A closer read of the data you already have.

• Pre-launch creative scoring. Scores ad sets before launch, not after.
• Network-effect intelligence. Every user’s anonymised data sharpens the model for everyone.
• Percentile benchmarks. You see where you rank, not just your raw score.
• Real performance correlation. Cortex learns from actual ROAS, CTR, CPA. Real numbers, not AI opinions.
• Five-layer Helix trend model. A purpose-built pattern engine, not a generic time-series model.
• Cross-platform intelligence. Meta, Google Ads, and TikTok unified in one model. One creative, three platforms, automatic matching.
• Cortex-validated trends. Signal output cross-referenced against real ad data so you can weight what’s supported by real numbers.
• Compounding loop. Analyses train Cortex. Cortex sharpens scoring. Scoring sharpens recommendations. Recommendations sharpen your next analysis.
• Fatigue-risk detection. Surfaces patterns that historically precede fatigue by 5 to 7 days on accounts with enough data.
• Personalised intelligence. After 30 or more linked campaigns, Cortex calibrates to your specific audience and creative patterns (70% your data, 30% global).