The Portable Loop: Import, Enrich, Patterns, Runbook, Export

Owned creative intelligence is the headline. The operating system underneath is intentionally boring—on purpose. If you are building for teams who are already on fire, the workflow has to read like a checklist, not a manifesto:

Import → Enrich → Find patterns → Generate runbook → Export

That is the spine of what we are shipping at Creative Patterns: pull campaign data in, add the creative context the raw export forgot, surface what actually won or lost, turn it into a structured brief you can hand to any model (or a human) as a runbook, and export prompts, copy, and files that are ready to go back into the ad platform or the next channel. The point is not to replace ChatGPT, Claude, Midjourney, or the native gen boxes inside Reddit, Meta, or Google. The point is to give the team a serious place for memory and method on top of all of them.

Why the CSV is not enough

Exports are honest about numbers. They are almost always lying by omission about what the creative was trying to prove: the hook type, the persona, the offer shape, the visual pattern, the claim the brand was testing against last quarter. The platform optimizes. It does not owe you a durable, diffable, versionable story of the creative hypothesis. "Enrich" is the step where the team re-attaches that missing semantics—so when you look back in six months, you are not guessing why a line of copy was on the ad.

Patterns, not vibes

"Find patterns" means: score what is winning and losing against the baselines the engine can actually compute, then narrate the numbers—never the other way around. That is the same posture as a compiler: the model can help you write the sentence, but the invariants and thresholds belong in the system. I wrote about the discipline of tight payloads in high-density agentic payloads: if you are paying in tokens, pay for entropy, not decoration. A runbook that drops straight into a notebook or an assistant is only useful if it is dense, structured, and grounded in the evidence you own.

Same bet as machine-readable design

Google's DESIGN.md direction with Stitch is a push to make design intent legible to tools. Creative Patterns is the adjacent layer for the buy side of performance work: a machine-usable, human-readable record of what the spend proved about creative—portable, reviewable, and not trapped in a single product's UI. The brand spec and the learnings from live traffic should be able to sit in the same culture: version control, diffs, and a cold no when something is untrue.

Where to look next

The vision piece is here. The piece on taste, memory, and feedback loops is the essay on why this discipline wins when more variants are free. The surface area for the product itself is creativepatterns.app.