Objective

• Build a repeatable, AI-powered system that produces on-brand PayPlan creative across formats, ad creative, one-pagers, retailer materials, without a design bottleneck
• Encode PayPlan's existing brand so that AI output starts on-brand by default, instead of being corrected after the fact
• Prove that a non-designer could generate consistent, on-brand assets from a plain-language prompt

What I Did

• Trained Claude on the PayPlan brand by structuring the brand guide into knowledge files, hex codes, typography, logo usage, spacing, and voice, so every output references the real brand rules
• Hosted PayPlan's actual brand assets (logos, product imagery, fonts) where Claude could reference the real files, not approximations
• Engineered the prompt structure and output specs, the creative brief, exact dimensions per placement, and safe zones — so the system produces coherent, consistent creative on demand
• Generated the creative as HTML and CSS rather than flat images, giving precise layouts, real editable type, and true brand color instead of blurry, uneditable AI output
• Converted the output into editable Figma designs using the html.to.design plugin, then applied the final layer of craft and judgment before delivery
• Configured the full PayPlan brand inside Claude Design and generated a retailer one-pager from a plain-language prompt as an end-to-end proof of the system

Outcome

• Delivered a repeatable AI design system, not a one-off asset, the brand is encoded once and referenced every time
• Produced on-brand creative in a fraction of the usual time, with consistency held across every format
• Built a foundation a non-designer could prompt, freeing the brand from the design queue
• Preserved craft by keeping a human in the loop on the work that matters, AI handles the first 80 percent, judgment handles the 20 percent that makes it good

‍