Agentic Workflows
Ad Diversity at Scale
Company
ConsumerAffairs
My Role
Associate Creative Director
Tools
Weave Claude CoWork
Timeline
Q2 2026
Description
Automating static ad creation at scale
Context
At ConsumerAffairs, I built a custom agentic pipeline inside Weave that transformed how we produce paid social creative. The result: we went from producing 16 static ads per month to 240 — a 15x increase in output without a proportional increase in headcount or production time.
In 2024, Meta rolled out its Andromeda ad delivery update — previous systems rewarded a handful of proven top performers, Andromeda's machine learning infrastructure required a much broader creative pool to find signal. For our team, the mandate was clear: producing 16 ads a month wasn't a resource constraint — it was a competitive liability. We needed to flood the system with high-quality, diverse creative to give Meta's algorithm what it needed to optimize, or we'd continue watching CAC climb while ROAS eroded.
Process
Built with assistance from Claude Co-work, our agentic workflow runs in five structured stages inside Figma Weave:
Data intake. The pipeline simultaneously ingests Category Configs, Motion's top-performing ad data, Motion AI Insights, Demographic Information, and Evergreen Messaging Themes to establish the creative parameters for each run.
Copy optimization. Weave's AI layer stress-tests combinations of Headlines, Sub-headlines, and CTAs against each config — selecting only the pairings most likely to drive conversion for that audience and offer.
Template matching. Optimized copy combinations are mapped against a library of winning ad templates already proven to perform in Meta's auction environment.
Grid generation. The pipeline assembles a grid of twenty unique creative executions per run — distinct enough to avoid internal auction competition, consistent enough to maintain brand integrity.
Review and deploy. A human creative lead does a final pass for brand consistency before assets are exported and trafficked — keeping judgment in the loop without making it the bottleneck.
Solution
Ad output scaled by 1,400%, giving Meta's algorithm the creative volume it needed to find and optimize toward high-intent audiences. Paid Social VMD grew 10x year-over-year, from $1M to $10M, as improved creative diversity directly unlocked greater spend efficiency and delivery scale. ROAS stabilized in the 1.5–2.5x range — consistent performance across a high-volume, rapidly iterating creative pool that would typically introduce variance. And lead volume tripled, a direct reflection of the pipeline's ability to keep fresh, relevant creative in market without the gaps and drop-offs
Key Insights
By engineering diversity at the input level — varying hooks, demographic framing, CTA logic, and visual templates in a structured way — we were able to maintain freshness in the auction without sacrificing the brand consistency or performance signals that make creative trustworthy over time. The pipeline doesn't just produce more ads. It produces ads that are different in the ways that matter to the algorithm, and similar in the ways that matter to the brand.