This ecommerce growth agency hit a scaling wall. They needed to scale ad production from 40 to 1,000 ads per client per month to meet creative testing demand across 30+ clients, but margin pressure made hiring more people impossible. Manual production was too slow and too expensive. They built a custom AI ad generation system to solve it.
Agency Intelligence built the system using GPT-5 for analysis, GPT Image 1.5 for image production, and an interface with human approval. The system turned high-performing ads into hundreds of on-brand variants, scaling production from 40 to 1,000 ads per client per month while cutting cost per asset from $10 to $0.15.
The result: unlocked testing velocity across all clients, drove universal CAC reduction and ROAS improvement, and reclaimed margin without adding headcount. The agency could finally meet client growth demands without destroying unit economics.
The Agency
This is a mid-market ecommerce growth agency serving DTC brands between $2M and $100M in annual revenue. They manage about 30 active clients and deliver financial forecasting, creative production systems, and paid media management focused on profitability metrics like aMER, contribution margin, and CAC control.
Their model is not just ad buying. They run integrated services: strategy, creative production, and conversion optimization. The agency has a strong track record scaling client revenue while maintaining or improving margins, but that performance depends on creative testing depth. Creative production had become the bottleneck.
The Problem: Ad Production Could Not Scale Without Destroying Margins
The agency's performance model required high-velocity creative testing. Dozens to hundreds of ad variants per client per month to optimize CAC and ROAS. Manual production, with designers and copywriters building assets one at a time, could deliver about 40 ads per client per month, but each asset cost roughly $10 in labor and required significant human effort.
To scale testing across 30+ growing clients, they would need to hire more creative staff. But margins were already tight. Adding headcount to match production demand did not pencil. They were stuck: sacrifice testing velocity and client performance, or blow up unit economics by scaling payroll.
- Manual production: ~40 ads/client/month at ~$10/asset
- Needed hundreds of variants to test nuances: headline, CTA, image, promotion, layout
- Thin margins prevented scaling headcount to match demand
- Limited A/B testing depth across 30+ client accounts
- Slower turnaround meant fewer iterations, less optimization
- Creative became the bottleneck preventing clients from scaling spend profitably
The agency could not out-hire the problem. They needed a system that could produce volume, maintain quality, and protect margins all at the same time.
Goals and Success Criteria
The goals were clear: dramatically increase ad volume per client, cut cost per ad to sustainable levels, maintain brand quality across 30+ distinct clients, and enable faster testing to improve performance.
Success looked like 10x or better production increase, cost per ad under $1, 70%+ approval rates, and faster testing cycles that help clients scale profitably.
Constraints
They also needed a simple workflow with human approval before ads went live, an intuitive interface for non-technical team members, and the ability to adapt as ad platforms and client strategies evolved.
The Solution: A Custom Ad Generation System with Human Quality Control
We built a custom AI ad generation app that turns high-performing ads into hundreds of on-brand variants. The system uses AI to analyze what works, generate variations across copy and images, and surface concepts for human review. Approved ads download ready for ad platforms.
The team uploads high-performing ads with performance data. AI analyzes creative elements and generates variations with headline tweaks, CTA swaps, image adjustments, and layout changes. Concepts surface in an approval interface where the team approves or rejects with feedback. Rejections train a knowledge base that improves future output.
- Custom AI ad generation tool
- Ad performance analysis engine
- AI image generation pipeline
- Platform specific layout library
- Human approval interface
- Knowledge base with continuous learning
- API based architecture for model updates
- Training and onboarding for creative team
Implementation Timeline
The Results
Production Volume Increased 25x (40 → 1,000 Ads/Client/Month)
Post-rollout, the agency delivered 1,000 ads per client per month across large ad accounts with multiple campaigns. That volume enabled nuanced A/B testing at a scale that was operationally impossible before.
Cost Per Asset Dropped 98% ($10 → $0.15)
Per-asset production cost fell from roughly $10 in designer and copywriter time to $0.15 in API costs. That margin improvement unlocked sustainable scalability without destroying unit economics.
70% First-Pass Approval Rate Maintained Brand Safety
Human review caught off-brand tone, inaccurate copy, and layout issues before ads went live. The 6 bucket rejection feedback system trained the knowledge base, improving quality over time while maintaining brand safety across 30+ client brands.
All Clients Saw CAC Reduction and ROAS Improvement
High-velocity testing enabled by the production system drove measurable performance improvements across the client base. While specific metrics varied by account, every client saw CAC decrease and ROAS improve as testing depth increased.
Reclaimed Margin Without Adding Headcount
The agency scaled creative output 25x without hiring additional designers or copywriters, solving the original margin bind. Time previously spent on manual asset production shifted to strategy, account management, and higher-value creative direction.
- 25x production volume increase
- Faster turnaround: same-day concept delivery
- Scaled to 30+ clients without adding headcount
- Human approval maintained brand and quality
- API architecture to upgrade as models improve
- 98% cost per asset reduction ($10 → $0.15)
- Margin: met client needs without scaling payroll
- Universal CAC reduction and ROAS improvement
- Faster testing cycles = faster optimization
- Competitive advantage in ad testing depth
Key Takeaways
Margin pressure forced innovation and unlocked scale
The agency could not hire their way out of the creative bottleneck. That constraint forced them to build a system that worked at the unit economics they needed. The result was a solution that scaled output 25x while cutting costs 98%.
Human approval is not a compromise, it's the quality gate
A 70% first-pass approval rate meant the system was not perfect, but it was good enough to deliver value at scale. The 30% rejection rate proved the quality control worked, feeding the knowledge base to improve future output.
Volume unlocks testing depth, testing depth unlocks performance
High-velocity creative testing is the competitive advantage in modern performance marketing. Agencies and brands that can test faster, test more variations, and iterate based on data will outperform competitors stuck in slow, manual production cycles.
Build to evolve
API based architecture means the system can adopt newer models and better image generators, without rebuilding the tool. That future-proofing mattered. AI capabilities improve fast, and systems that cannot adapt become obsolete.
Want to test faster and spend smarter?
We build custom AI creative systems that turn your high-performing ads into hundreds of on-brand variants, maintain quality with human approval, and connect directly to your ad platforms so you can scale testing without scaling payroll.