Skip to main content
RoastIQBuyerLensHugoPricingBlogAbout
Book a demoSign inStart free →
Case studies · 9 real ads · updated 18 May 2026

Twelve ads. Twelve verdicts.
One rubric.

Nine real ads from nine global brands, pulled from official public transparency tools, scored through the same RoastIQ pipeline that runs on every customer upload. No cherry-picked wins, no curated REBUILDs. The real verdict mix landed at 1 SCALE · 8 SHARPEN · 0 REBUILD, global category leaders running through public transparency tools rarely produce a REBUILD on the opening pattern, and the gallery shows that honestly.

12 ads scored 5 categories 3 platforms 2 verdict bands ρ +0.31 held-out OOS
Oussama Nakhil portrait
Oussama Nakhil · Founder, SaliencyLab, Multiple years buyer-side: NielsenIQ insights, then L'Oréal Groupe in global marketing insights.
Every case below was scored with the same pipeline available to every paying account. No bespoke runs, no manual tuning.

How we built this gallery.

Every ad on this page comes from an official public transparency tool, Meta Ad Library, TikTok Creative Center, TikTok Ad Library, or Google Ads Transparency Center. We do not scrape. We do not display the creative in a context the brand would object to. We selected twelve from a benchmark pool of 2,047 to span five categories, three platforms, and across the verdict bands the pipeline actually produced.

Each card shows the composite score, the five frozen KPIs as mini-bars, the verdict, a one-sentence diagnosis, and a one-sentence recommendation. The diagnoses describe the opening pattern, not the brand. A REBUILD verdict on one ad does not mean the brand has a quality problem, it means that particular opening pattern is predicted to lose more attention than the platform's median.

Honesty pact in one paragraph: RoastIQ predicts engagement (likes, shares, comments) and click intent (CTR percentile) on TikTok, plus view counts on YouTube. Held-out OOS Spearman ρ +0.30 to +0.32 (2026-05-05). It does not predict sales, ROAS, attributed conversion, or in-market brand recall. The numbers below are model predictions, not measurements.

Scores generated by the RoastIQ pipeline on 2026-05-18 against model_version gemini-2.5-flash / benchmark_pool_version scoring-runner-v1.
All 12 Food & Beverage Fashion Beauty Automotive Entertainment Travel SCALE · 4 SHARPEN · 8 REBUILD · 0
CeraVe TikTok ad opening frame TikTok · UK 58
CeraVe · Beauty

Educational POV, late brand cameo

SHARPEN
Beat the Skip60
Get Noticed65
Brand Impact50
Sell Prop.60
Build Brand55

Diagnosis: Educational hook with medium-paced opening; the brand is not yet visible in the first three seconds. Brand Impact lands at the rule's lower edge, the diagnostic moment for this opening pattern is the gap between the educational claim and the product arriving on screen.

Keep the educational POV. Bring the pack into the first 1.5 seconds (held by the talent, not as a sting), Brand Impact moves Moderate → Strong without changing the script.

TikTok Ad LibrarySource: public
Read full case study
L'Oréal Paris Meta ad opening frame Meta Reels · US 60
L'Oréal Paris · Beauty

Trend-led studio open, product visibility deferred

SHARPEN
Beat the Skip65
Get Noticed60
Brand Impact60
Sell Prop.55
Build Brand60

Diagnosis: Trend hook with dynamic motion, but the product is not visible in the first three seconds and the saliency layer flags visual ambiguity. Sell Proposition is the softest KPI, the claim the ad is making is not yet specific in the opening window.

Don't rebuild, sharpen. Bring the product into frame 1 (dressing-table angle, hand-held) without losing the trend hook. Sell Proposition and Brand Impact rise together.

Meta Ad LibrarySource: public
Read full case study
Coca-Cola Meta ad opening frame Meta Feed · US 58
Coca-Cola · Food & Beverage

Emotional-story open, slow opening motion

SHARPEN
Beat the Skip60
Get Noticed65
Brand Impact60
Sell Prop.40
Build Brand65

Diagnosis: Emotional-story hook with a slow opening; brand not yet in frame at three seconds. Sell Proposition lands at 40, the lowest KPI on any card here, because the establishing pacing was tuned for long-form, not the Meta Feed window.

Cut the establishing seconds. Open on the bottle reveal; cut backwards into the emotional beat. Sell Proposition and Beat the Skip both move up; Brand Impact stays where it is.

Meta Ad LibrarySource: public
Read full case study
BMW YouTube ad opening frame YouTube · DE 55
BMW · Automotive

Customer-testimonial cold-open, slow motion

SHARPEN
Beat the Skip50
Get Noticed55
Brand Impact60
Sell Prop.60
Build Brand50

Diagnosis: Customer-testimonial hook with slow opening motion and no brand visible in the first three seconds. The trust-led value proposition is doing the work, but the opening pace burns the attention window before the brand earns it back.

Keep the testimonial. Cut a hard insert of the badge or grille at the 1.5-second mark. Build Brand and Beat the Skip both shift up without re-shooting.

YouTube · Google Ads TransparencySource: public
Read full case study
Booking.com YouTube ad opening frame YouTube · AE 75
Booking.com · Travel

Benefit-hook open, app visible at 3s

SCALE
Beat the Skip85
Get Noticed75
Brand Impact80
Sell Prop.70
Build Brand70

Diagnosis: Benefit-led hook with the brand and app interface present in the first three seconds. Beat the Skip is the top KPI on the card, the price or pin-style benefit appears before the skip decision. All five KPIs clear the Strong threshold.

Hold this structure as a template across other geos. To lift Sell Proposition further, swap the destination montage for a single price-vs-search-result comparison.

YouTube · Google Ads TransparencySource: public
Read full case study
Nivea YouTube ad opening frame YouTube · DE 67
Nivea · Beauty

Product-showcase open, static opening motion

SHARPEN
Beat the Skip75
Get Noticed70
Brand Impact70
Sell Prop.70
Build Brand55

Diagnosis: Product-showcase hook with brand visible in the first three seconds, but the opening motion is static, so Beat the Skip carries on brand familiarity rather than visual energy. Build Brand sits at the rule's lower edge because the asset variety is narrow.

Re-cut, don't re-shoot. Add a face mid-expression or a problem-state beat as frame 1; let the pack shot land at 1.5 seconds. Build Brand and Get Noticed both lift.

YouTube · Google Ads TransparencySource: public
Read full case study
BMW Meta ad opening frame Meta Feed · DE 67
BMW · Automotive

Lifestyle-escape open, single brand pulse

SHARPEN
Beat the Skip75
Get Noticed70
Brand Impact65
Sell Prop.55
Build Brand75

Diagnosis: Lifestyle-escape hook with slow opening motion; brand is present in the first three seconds, but only once. Sell Proposition is the softest KPI, the specific spec or model the ad is selling is implied, not on-screen, in the opening window.

Add one spec callout (range, model name, or trim) in the 1.5–3-second window. Sell Proposition lifts into the Strong band without touching the cinematography.

Meta Ad LibrarySource: public
Read full case study
Nike YouTube ad opening frame YouTube · US 67
Nike · Fashion / Sport

Problem-solution open, text-heavy first window

SHARPEN
Beat the Skip75
Get Noticed70
Brand Impact70
Sell Prop.65
Build Brand60

Diagnosis: Problem-solution hook with dynamic motion and brand visible in the first three seconds, but the saliency layer flags text dominance, the opening is reading like a slide, not a film. Build Brand sits at the rule's edge because the visual brand cue arrives after the text.

Keep the structure. Replace the on-screen text with a face-first beat from the existing footage and let the type land at 1.5–2 seconds. Get Noticed and Beat the Skip both rise.

YouTube · Google Ads TransparencySource: public
Read full case study
Netflix Meta ad opening frame Meta · US 67
Netflix · Entertainment

Emotional-story open, brand integrated throughout

SHARPEN
Beat the Skip75
Get Noticed70
Brand Impact60
Sell Prop.65
Build Brand70

Diagnosis: Emotional-story hook with dynamic motion and brand visible in the first three seconds. Brand Impact lands at the rule's lower edge, the Netflix mark is implied via tone and grade, more than shown. Card 04 has the same brand getting Brand Impact higher on YouTube with the same kind of opening.

Same trailer footage, different anchor. Lead with the most dramatic single shot from frame 1; bring the title card to 1.5 seconds. Brand Impact and Sell Proposition both move up.

Meta Ad LibrarySource: public
Read full case study

The card grammar, decoded.

// Composite score

The number in the corner.

Weighted sum of the five KPIs (25 / 20 / 20 / 20 / 15). Not an average. Falls into one of three bands. The whole pipeline is designed to land here in under 90 seconds.

// KPI bar colors

Traffic-light, not vibes.

STRONG ≥ 70, Pattern is working, replicate it elsewhere.

MODERATE 45–69, Sharpen, don't rebuild. Specific edits beat starting over.

WEAK < 45, Two or more reds force a REBUILD verdict regardless of composite.

// Verdict

Three states. No fourth.

SCALE, composite ≥ 70 AND no KPI < 55. Buy media.
SHARPEN, composite 55–69. Fix the weak KPI; do not start over.
REBUILD, composite < 55 OR two+ KPIs < 45. Brief is wrong.

Four questions we get asked.

How are these ads picked?
From the 2,047 ads in the SaliencyLab benchmark, we selected twelve to span five categories, three platforms (TikTok, Meta, YouTube), and the verdict bands the pipeline produced for them. The real mix is 1 SCALE · 8 SHARPEN · 0 REBUILD, REBUILDs are rarer on top-of-funnel work from category leaders sourced through public transparency tools, and we did not hand-pick ads to force a specific mix.
Are the scores reproducible?
Yes. Every ad shown here was scored through the same RoastIQ pipeline that runs on /app/analyze, using the same five-KPI rubric and the same composite weights (25 / 20 / 20 / 20 / 15). The same upload through the same pipeline returns the same score.
Did you contact the brands? Is this fair?
No, we did not contact the brands. These ads are sourced exclusively from official public transparency tools, Meta Ad Library, TikTok Creative Center / Ad Library, and Google Ads Transparency Center. We do not scrape. The diagnoses describe the opening pattern, not the brand's overall quality. A REBUILD verdict on one ad does not mean the brand is failing, it means that specific opening pattern is predicted to lose more attention than the platform's median, on the platform shown.
Why these brands and not smaller ones?
Global category leaders are featured because they (a) run high enough volume that public transparency tools surface multiple variants, (b) are recognizable enough that the diagnosis lands without category explanation, and (c) can fairly absorb a REBUILD verdict on one opening pattern. Smaller brands are scored privately for paying customers, not published in a public gallery.