Score the ad,
before you spend.
Upload one ad. Get a scored verdict in ninety seconds, benchmarked against tens of thousands of public outcome ads.
Operators who lived the creative testing problem from the inside.
Operators already running pre-spend diagnostics with SaliencyLab
Paid-media leads and GTM leaders across Estée Lauder brands, independent agencies, and enterprise SaaS teams.
“I manage paid media for Erborian across France, the UK, the US and Europe. Being able to check creative effectiveness in minutes, in English, in French, with the right regional benchmarks, lets me optimise campaigns daily instead of monthly.”
“The first diagnostic I've used that flags what's wrong before the data comes in, when I can still change the cut.”
“I've run GTM for SaaS platforms scaling to 30K customers. Creative testing is the loneliest line in the marketing budget. Either you skip it or you pay Kantar. SaliencyLab is the first option actually proportional to how modern teams operate.”
“I've built products on YC's playbook: validate cheap, ship fast. SaliencyLab is how I validate creative without briefs or wait times.”
Score, decide, ship: in 90 seconds.
RoastIQ scores first. BuyerLens explains the why. Four steps, one project layer.
Upload one ad. Any format.
Video or image. MP4, MOV, JPG, PNG, up to 60 seconds. No SDK, no integration, no media-buying-platform setup. Just drag, drop, done.
≤ 5 sec to uploadMore on RoastIQ →RoastIQ scores it scene by scene.
RoastIQ reads the visuals, transcribes the audio, classifies 35+ creative attributes per scene, and benchmarks the result against 1,200+ real ads. Every detection is timestamped and traceable.
90 seconds end-to-endMore on RoastIQ →Skip
Noticed
Impact
Prop
Brand
5 KPI scores + a single verdict.
Scale, Sharpen, or Rebuild. Each verdict is tied to specific creative attributes (the hook timing, the brand's visual prominence, the sell proposition pacing) so you know what to change, not just whether to ship.
Held-out OOS ρ +0.40More on RoastIQ →When the score still leaves the ‘why’ unanswered, open BuyerLens.
Synthetic users built from your category, market, and target persona explain their resistance, in their own voice. Use the answers to shape the next edit, not just guess at it.
Optional · Opens from any RoastIQ resultMore on BuyerLens →How five signals become one verdict.
No black-box brand-equity index. Every weight is held-out cross-validated against public outcome data: five scored questions, one composite, one verdict.
Skip
Does the hook land before the scroll reflex does, or does the brand show up after the thumb has already moved on?
Noticed
Visual contrast and an opening hook strong enough to survive next to everything else in the feed.
Impact
Logo surface area and mark timing: does the brand cue register before the viewer decides to scroll?
Proposition
An explicit offer and a closing frame that gives the viewer a reason to act now, not just a good feeling.
Brand
Tone, codes, and distinctive assets that still read as you three campaigns from now, not just this one cut.
77 lands in Scale: composite ≥ 70 and no single KPI below 55.
In the founder's words.
Pre-spend creative testing has been broken for thirty years. The $56B market-research industry still runs on consumer surveys.
- Small samples
- Slow turnaround
- No methodology disclosure
- Multimodal AI
- Scored against real public outcomes
Four ways that's different:
1Speed over surveys.→ 
Survey-based testing takes four to eight weeks and reaches roughly 150 respondents per cell. RoastIQ scores your ad in about ninety seconds, calibrated against a benchmark cohort of 2,000 real ads.
2Open, not a black box.→ Legacy panels never disclose methodology and report concurrent validity, never held-out. Every SaliencyLab report shows the benchmark size behind the verdict, and our validation code is public with a peer-review track at Scientific Data.
3Built for the feed, not the boardroom.→ Survey-based testing was built for the TV era, one ad at a time, anchored to recall, what people say they remember. SaliencyLab is anchored to 42 million behavioral signals extracted from public ads, what the feed actually rewards.
4Evidence you can check.→ Held-out five-fold cross-validation puts us at ρ +0.33 to +0.40 across TikTok CTR and YouTube views — on the cohort those weights were tuned on, brand-clustered 95% intervals [+0.32, +0.53] and [+0.37, +0.58] exclude zero. Not vibes.
Start free. Pay when you scale.
First analysis in under ninety seconds. No card required. Same scoring engine across every tier. We do not gate accuracy by plan.
For curiosity, brief reviews, and early stage creative checks.
- 3 analyses per month
- Full 5 KPI breakdown plus verdict
- Platform native scoring
- Public benchmark comparison
- Methodology disclosure on every report
For brand teams iterating on creative every week. 300 analyses a month plus BuyerLens.
- 300 analyses / month (30 / day)
- BuyerLens synthetic buyer panel
- Brand drift tracking over time
- Private benchmark cohort
- Priority compute · sub 60 second scoring
- Team workspace · 5 seats included
For agencies and global brands needing private benchmarks, API access, SOC 2.
- Everything in Pro
- Private benchmark cohorts
- API access plus webhooks
- Dedicated CSM and onboarding
- SOC 2 Type II plus DPA
- Volume pricing for high throughput teams
Start free with 3 analyses per month · No card required · Cancel anytime · GDPR ready · SOC 2 Type II in progress
Stop guessing.
Start simulating.
Thirty minutes. We score one of your real ads live and walk through the verdict, heatmap and synthetic buyer reaction. You leave with the report, even if you do not sign up.
Oussama Nakhil
Founder at SaliencyLab. Previously L'Oréal Groupe, NielsenIQ, HEC Paris, Grenoble EM.
SaliencyLab · Demo
Does the score predict sales or ROAS?
No. RoastIQ predicts relative engagement and click-intent on TikTok and YouTube. It is calibrated against the public outcomes of 1,200+ real ads (held-out out-of-sample Spearman ρ +0.33 to +0.40).
We do not claim sales lift, ROAS, attributed conversion, or brand recall. The score is a pre-spend gate, not a substitute for in-market measurement. That honesty is part of the contract.
How is this different from Kantar Link AI or other survey-based testing?
No survey panel. No three-week wait. No proprietary black box.
SaliencyLab is calibrated against 1,200+ public-outcome ads with measured engagement and click-intent data. Methodology is open, validation code is reproducible, and every report shows the benchmark cohort size for your category. Ninety seconds instead of three weeks, at a fraction of the cost.
How accurate is it, really?
Our platform-native attribute model, measured against public outcomes: TikTok CTR (ρ +0.33 held-out, n=691), YouTube views (ρ +0.40 held-out, n=397).
On the cohort the weights were tuned on those rise to ρ +0.43 [+0.32, +0.53] and ρ +0.49 [+0.37, +0.58] — 95% intervals, brand-clustered. Lead with the held-out numbers: they are what to expect on an ad the model has not seen. We publish no TikTok engagement figure, because held-out evidence there is no better than our own baseline.
Pool-wide quintile lift is 6.5×. Ads in the top predicted quintile beat the bottom quintile by that multiple on real engagement. Scaling curves show ρ still growing as the cohort expands. Meta is directional defaults only; we mark it as such on every Meta-Feed score.
Is my creative private?
Your creative stays in your workspace. We do not use customer ads to train scoring models. Calibration runs on the public benchmark cohort only.
Storage is purged on request. GDPR-ready. SOC 2 Type II in progress. DPA available on the Enterprise tier.
Does it work for static images, or only video?
Both. MP4, MOV, JPG, PNG, WebP up to 60s.
Static images skip the audio and video-saliency stages and finish in about thirty seconds. Long-form video beyond 60s is on the roadmap.
What if my category isn't in the benchmark yet?
Every report shows the benchmark cohort size next to the verdict. If your category has fewer than 100 ads in the pool, we flag it as “Building” with the current count and growth rate, never hidden as “Unbenchmarked.”
If you need a private brand benchmark, the Pro tier builds one across your own uploads so you can track drift over time even where the public pool is still growing.
What we're building and saying.
DigiTalk: How AI Can Save Creatives' Time with Oussama Nakhil
Oussama joins the DigiTalk podcast to talk through how AI-based creative diagnostics change the pre-spend workflow, and where RoastIQ fits as that shift plays out.
