Learning Engine
Every campaign outcome becomes a written, confidence-scored learning. These are injected into every future agent run — this page is why month six beats month one.
9learnings in this workspace's store
For single-headline requests under tight word counts, pairing one number with one named formulation flaw fits within 8 words without sacrificing specificity — testable by whether compressed weakness+number headlines maintain their CTR advantage versus longer versions.
← Live run — Quill
When a product brief lacks a differentiator, splitting headline angles across benefit/spec/lifestyle surfaces the client's implicit positioning preference — testable by whether their pick correlates with the differentiator they later name.
← Live run — Atlas
Headline requests without a stated channel default best to one benefit-led, one spec-led, and one lifestyle-led option — testable by whether clients pick across angles rather than within one.
← Live run — Atlas
For ingredient-conscious skincare audiences, headlines that name a known formulation weakness (e.g., vitamin C oxidation) and quantify the fix outperform headlines that only quantify the benefit.
← Live run — Quill
For sub-$25k monthly budgets, concentrating on two channels beats spreading across four — every split test confirmed it.
← Quarterly portfolio review
Win-back emails with a plain-text look from a founder's name beat designed templates on revenue per send by 64%.
← Win-back Email Sequence
Comparison pages targeting '[competitor] alternative' keywords convert 2.4× better than feature pages for B2B SaaS.
← Comparison Pages — Organic Cluster
Shifting 20% of Meta budget to 6–8pm dayparting lifted ROAS from 3.9× to 4.7× for coffee retargeting.
← Founder Story — Video Retargeting
Headlines that lead with a specific number outperform benefit-led headlines by ~31% CTR for DTC skincare audiences.
← Spring Launch — Performance Max