1,000× cost reduction on ad creatives. 858 compliant images in two weeks.

Meridian Claims runs paid acquisition campaigns across Meta for financial claims — the kind of advertising where a single compliance error (a missing disclaimer, a prohibited word in a headline) can get an ad account shut down.

Their creative production was manual: a designer or agency produced 5–10 new images per week at £50–150 each. Video was outsourced at £200–500 per clip. Compliance review happened after the creative was finished — meaning rework was frequent, expensive, and slow.

The bottleneck wasn't strategy. The bottleneck was production. They knew which angles worked (social proof, authority, urgency) but couldn't test fast enough to find winners. By the time a winning ad set was identified, the creative was already fatiguing.

At 200 images per month across multiple brands, the agency model cost £120–200k per year in creative production alone.

Claude generates structured ad copy, image prompts, and video specifications from a single input — a JSON file defining the client, persona, and creative angle.

Nano Banana (Google Gemini) generates the images. Each image is automatically scored by a separate Gemini instance on a 10-point scale. Anything below 7/10 is regenerated without human intervention.

A Python compositor overlays compliant disclaimers and CTAs onto every approved image — the mandatory regulatory text is baked into the creative, not added manually in the ad platform.

For video, fal.ai runs Kling, Veo, and Sora models. Each video is scored the same way — below 6.5/10 triggers a regen.

Everything lands in Supabase: structured storage, taxonomy, searchable by angle, persona, and brand. A dashboard gives the marketing team a two-stage approval workflow — marketing review first, then compliance review — before anything reaches the ad platform.

The first production run delivered 858 images across 97 category folders in two weeks. Total compute cost: roughly £50. The equivalent agency cost at £50 per image: £42,900.

More importantly, the pipeline changed how the team works. Instead of waiting a week for 10 new creatives, they now run weekly iteration cycles: launch → identify winners → generate 50 variations of the winning angle → launch again. The speed of testing went from monthly to weekly.

Compliance errors caught before launch instead of after. The compositor enforces regulatory text on every image — no human has to remember to add the disclaimer. No ad account suspensions since the pipeline went live.

The pipeline is reusable across brands. The same system produces creatives for three separate claims brands, each with its own compliance rules, tone, and visual identity. Adding a new brand is a config file, not a rebuild.

Estimated annual saving versus the agency model: £120–200k.