The situation
A mid-size DTC skincare brand, roughly four years into the market, was running a mature paid acquisition stack. Meta (Facebook Feed, Instagram Feed, Stories) took the majority of the budget, with Google Performance Max and a small Display budget filling in the rest. Monthly ad spend sat in the mid-six figures USD.
The creative operation was lean but high-output: two in-house designers, one media buyer, and one creative strategist shipping 15–20 new variants per month across product lines, offer types, and seasonal themes. They were doing everything right on paper — high creative volume, iterative testing, proper UTM tagging, and a healthy A/B test discipline.
Despite all of that, the blended ROAS had been stuck at 2.1× for six straight weeks. Not catastrophic, but not sustainable either. And with a major spring push approaching, the team needed to figure out why the plateau held — and break it.
The problem
The plateau wasn't one thing. It was a cluster of overlapping issues that compounded:
- Blended ROAS stuck at 2.1× for six straight weeks despite consistent creative volume.
- Shipping 15–20 new variants per month with no systematic pre-launch review — every creative got roughly equal test budget.
- Clear creative fatigue signals: CPMs rising, CTR softening week over week, but no way to identify which variants were the true underperformers until after spend.
- Internal disagreement on which creative direction to double down on. Reviews were stakeholder-opinion-led, not data-led.
- Seasonal push looming — the team needed to stop burning budget on losing creative before a major launch window.
The approach
Instead of overhauling the creative pipeline, the team added one step: a pre-launch attention score on every new variant, with a hard floor on what could ship to ads manager.
Baselined the existing creative library
The team uploaded their last 40 live creatives to GazeIQ and scored them retroactively. The correlation between attention score and CTR was strong — creatives scoring 75+ had an average CTR 1.6× higher than those scoring 60 or below. This gave them internal evidence that the metric meant something for their audience.
Set a 70-point minimum as a launch gate
Any new variant under 70 went back to the designer with the specific GazeIQ recommendation attached (move CTA, increase product focus, simplify hierarchy). Nothing below 70 was allowed into ads manager. Design reviewed, iterated, re-scored.
A/B tested the top 2–3 scoring variants weekly
Instead of running all 15–20 variants hot, only the highest-scoring 2–3 per theme got meaningful test budget. Losers from the pre-score were killed before spend. Winners moved into scaling campaigns within 5 days.
Built attention scoring into the creative brief
Briefs started specifying target sub-scores: CTA visibility ≥ 75, product focus ≥ 70, headline salience ≥ 65. Designers got the rubric before they opened Figma, which shifted creative decisions upstream instead of fighting them in review.
The results
CTR lift in Meta Feed after filtering sub-70 variants
Landing page conversion rate improvement
Blended ROAS after 90 days (up from 2.1× baseline)
Exploratory spend on weak variants
CTR lift showed up first. Within three weeks of routing every new creative through GazeIQ and filtering out sub-70 variants, Meta Feed CTR was already up 34% vs. the prior 8-week baseline. Landing page conversion lifted 22% — partly because the creatives that made it to launch had clearer offer signals, and partly because audience quality improved when weak creatives weren't pulling in low-intent clicks.
ROAS moved more gradually. By day 60, blended ROAS had crossed 2.9×. By day 90, it settled at 3.4× on a rolling basis — and held. Exploratory spend on losing variants dropped 41% because fewer losing variants made it to launch in the first place.
Our creative team stopped arguing about taste and started arguing about scores. That's the single biggest unlock — we finally had a shared language between design and media buying.
— Head of Growth, DTC skincare brand
Quote reconstructed from customer feedback. Individual and brand identities withheld at customer request.
Key takeaways
Pre-scoring didn't kill creative volume — it just redirected design effort away from losing concepts earlier in the process.
The 70-point minimum was arbitrary but effective. The real value was having any objective gate, not the specific threshold.
ROAS lift trailed CTR lift by about 4 weeks. If you only look at ROAS in the first month, you'll underestimate the impact.
The biggest non-metric win: design and media-buying stopped arguing about taste, because they had a shared number to aim at.
Attention scoring doesn't replace live A/B testing — it makes live tests more efficient by ensuring only viable candidates get spend.