Here's the AI funnel & split testing stack and process we run for 8-9 figure brands
1. Foundation
A custom Claude Code plugin with our own skills and commands.
Brand-agnostic and it carries every framework, every winning structure, every copy and design pattern we've validated across hundreds of funnels.
2. Brand-specific assets
The MCP pulls everything specific to the brand: winning ads, web and Reddit scrapes for the niche, competitor breakdowns, customer reviews, quiz answers, heatmap/clickmap/scrollmap data from existing pages.
Every brand has its own structured intelligence file.
The file is the brain. The plugin is the hands.
3. Research and angle building
Before anything gets drafted, our skills run research. It pulls the brand assets through the MCP and cross-references it against the agnostic patterns.
Output is a stack of validated angles with the evidence already attached
"This avatar struggles with X, here are 300 reviews mentioning it, here are 2 competitor ads using a similar angle, here's the framework that converts best on this awareness level"
4. Copy
The plugin's skills write the first draft, using all the wealth of data build in the previous 3 steps.
A human reads, reviews, and edits the output.
5. Design
Figma Claude Design with our MCP.
Every brand has its own design system and design tokens we set up for them.
We also keep a Figma repository of design patterns that have actually performed across multiple brands.
Claude iterates inside the brand's tokens and system so no generic AI slop is created.
Designer reviews and curates the outputs.
6. Development
Codex/Claude build the funnel using the design copy research from the previous steps, all pulled and managed cleanly through our MCP.
Output is clean, reusable Shopify templates and sections directly in the theme.
7. Deployment
Shopify's AI toolkit MCP to run GraphQL queries automatically that create new URLs with the correct templates already setup.
8. Split testing
We use Convertify AI, our own split testing app with its own MCP.
We spin up split tests with natural language straight inside Codex or Claude.
"Test this product's price against 2 variants, one with 5% and one with 10%" and it creates the test automatically with a preview link (DM if you'd like beta access)