When 300 Ad Variants Went Out Without a Design System

When 300 Ad Variants Went Out Without a Design System

A meal kit company gave its growth team AI tools for generating ad images and copy. Within a month, about 300 ad variants were running across social channels, far more than the brand team could ever have designed by hand.

 

Then the brand team looked closely. Four slightly different versions of the logo were in circulation. Several ads used a green the company had never chosen. One headline promised “50% off forever,” when the actual offer was half off the first box. Customer service spent two weeks handling complaints and issuing credits.

 

The tools hadn’t failed. Nobody had given them rules.

Advertisment

A Design System Is a Set of Decisions

People sometimes describe a design system as a style guide. In practice, it’s a collection of decisions made once so nobody has to make them again.

 

Colors defined as exact values. A type scale with set sizes for headlines, body text, and captions. Spacing rules. Then reusable components built from those pieces. For ads, that means templates with a locked zone for the logo, safe areas for each placement so text doesn’t get cropped on a phone, and an approved style of photography. For internal tools, it means buttons, forms, tables, and alerts that look and behave the same everywhere.

 

The payoff comes from consistency. A customer recognizes the brand in a split second of scrolling. An employee using six different internal tools knows that a red button always means something destructive.

Rules Machines Can Read

Most brand guidelines live in a polished PDF written for human designers. AI tools can’t do much with a sentence like “our green should feel fresh and approachable.”

 

They can work with exact color values, font names, a defined type scale, and template files with fixed layers. The meal kit company converted its brand book into a short, structured specification and a set of locked templates. AI tools now generate the image and the headline. The template controls the logo, colors, layout, and legal text.

 

The off-brand green disappeared overnight.

Advertisment

Where Governance Breaks

AI-generated creative raises governance questions most marketing teams haven’t faced. Looking at the five failure modes of AI governance through a design lens explains most of what went wrong at the meal kit company.

 

Brand rules were written only for people, so the tools ignored them. Nobody reviewed ads before launch because volume made review feel impossible. No one owned the prompts and templates, so each growth marketer kept their own versions. There was no record of which assets were AI-generated, which complicated questions about image rights later. And offer claims went out unchecked, which is how a first-box discount became “forever.”

 

The fixes were modest. One person owns the template library and approved prompts. Every ad passes an automated check for logo version, colors, and required disclaimers, plus a human check of any price or offer language. Every asset gets a label recording how it was made.

Measuring Which Variants Work

Producing hundreds of variants only helps if you learn from them.

 

The meal kit company tags every ad by template, image style, and headline type, then compares results in its ads reporting tool. Because the design system holds everything else constant, differences in performance point to the variable that actually changed. Recipe close-ups outperformed family dinner scenes by a wide margin on cost per new subscriber. Question headlines underperformed statements.

 

Without consistent tags, an ads reporting tool can only tell you which individual ads won. With them, it tells you which design decisions won, and those lessons carry into the next batch of work.

 

Templates that underperform for two months straight get retired. The library stays small enough to manage.

Internal Tools Deserve the Same Care

Operations teams building their own tools with no-code and AI builders face the same drift. One team’s approval app uses green for “reject.” Another’s inventory tool puts the delete button where every other tool puts save.

 

A shared component library for internal builders fixes most of it: standard buttons with clear danger states, confirmation steps for anything destructive, and consistent table and form layouts. Builders move faster, and employees stop making mistakes caused by unfamiliar screens.

Advertisment

Half Off, Just Once

The meal kit company still runs a few hundred variants a month. The brand team reviews templates and offer language rather than every ad, and customer complaints about misleading offers have stopped. The tools never needed to slow down. They needed the decisions written down where they could follow them.

Advertisment

Pin it for later!

When 300 Ad Variants Went Out Without a Design System 3

If you found this post useful you might like to read these post about Graphic Design Inspiration.

Advertisment

If you like this post share it on your social media!

Share on facebook
Share on twitter
Share on pinterest
Share on vk
Share on telegram
Share on whatsapp
Share on linkedin

You Might Be Interested On These Articles

Advertisment

Latest Post