6 Oct 25 
How brands produce hundreds of on-brand ad adaptations with a small design team: the system, the roles, and the guardrails, learned from automating creative at genuinely absurd scale.
Let us tell you about the moment creative automation stopped being theoretical for us.
We were working on Amazon Easy, the assisted e-commerce program that grew to 45,000+ stores across 4,500+ pin codes. Every store needed localized marketing material. Do the arithmetic on that with a traditional design workflow: even at a heroic ten adaptations per designer per day, you’d need a design department the size of a small town. The brief was impossible by hand. That was the point.
What made it possible wasn’t a magic tool. It was a system, and the same system, scaled down, is how a ten-person brand today can produce a hundred ad variants a week with two designers. Here’s the anatomy.
The Core Idea
Separate the Decisions From the Production
Every ad you’ve ever shipped contains two kinds of work glued together: decisions (what’s the message, what’s the visual concept, what does the brand allow) and production (resizing, swapping headlines, localizing, exporting 14 formats). Traditional workflows pay designer-hours for both. Creative automation splits them: humans make decisions once, machines execute them a thousand times.
That split has a name in practice: the template. Not “template” as in generic-website-theme, but a designed system with locked and variable layers.
The System
The Four Layers of a Creative Automation Pipeline
✅ 1. The design system (what never changes)
Logo placement, typography, color values, spacing rules, the compositional grid. This layer is designed once, by your best designer, and then locked. It’s the reason variant #847 still looks like your brand. Skipping this step is how automation produces on-brand garbage at record speed. Our AI creative checklist makes the same point about generative work: anchor to the brand or drift to the average.
✅ 2. The variable layer (what changes per variant)
Headlines, product shots, offers, locales, formats. Define these as named slots with rules: headline max 42 characters, product image on transparent background, offer text from an approved list. The discipline of naming what varies is itself clarifying. Most teams discover they have five real variables, not fifty.
✅ 3. The production engine (what does the work)
This ranges from humble to fancy, and humble wins more often than you’d think. A well-built Figma component set with variants. Batch-rendering scripts. Feed-based tools that merge spreadsheet rows into templates. Generative AI for background and imagery variation. The tool matters far less than the layer separation: we’ve seen spreadsheet-plus-script pipelines outperform expensive platforms because the template thinking was better.
✅ 4. The QA gate (what keeps you honest)
Automation fails in bulk: one broken rule produces 200 broken assets, quietly. Every batch gets a human pass, and not a rubber stamp. Check the edge cases: the longest headline, the darkest product shot, the smallest format. The reviewer’s job isn’t admiring the 90 percent that’s fine; it’s hunting the 10 percent that isn’t.
The Payoff
What Changes When Production Is Free
The obvious win is cost. The interesting win is behavioral: when a variant costs minutes instead of days, you stop rationing creativity.
Media buyers finally get enough creative to fight ad fatigue properly, instead of running one hero ad into the ground. You can afford to speak to segments too small to justify custom design before: the campaign for one city, one audience, one moment. And testing becomes real, because ten genuinely different concepts beat one concept in ten colours.
There’s a quieter benefit too: your designers get better jobs. Nobody went to design school to resize banners. When the production grind moves to machines, the humans move to concepting, art direction, and the system itself. Retention improves. So does the work.
Where to Start (Without the 45,000 Stores)
Start embarrassingly small: take your single best-performing ad, rebuild it as a proper template with locked and variable layers, and produce 20 variants for your next campaign. Measure. The template will be wrong in instructive ways; fix it and do 50 next month. The Amazon Easy pipeline wasn’t designed in a war room on day one. It evolved from exactly this loop, batch after batch, rule after rule.
The scale is optional. The system is not.
Curious what a creative automation pipeline would look like for your brand and your volume? This is genuinely our favourite conversation.