How we automate Meta ad creative testing with AI

Description and use

Automating Meta ad creative creation, publishing, and analysis with AI to ensure reproducibility, durability, and accurate attribution.

Problem

Automating the creation, publishing, approval, and reporting of Meta ad creatives while ensuring reproducibility, durability, and accurate attribution.

Outcome

The design enables reproducible, durable, and auditable Meta ad experiments, reducing manual errors, ensuring consistent variable control, and providing rich reporting context for accurate performance analysis.

Workflow

• Define experiment brief with hypothesis, variable, variants, fixed assets, and primary metric.
• Generate creatives using AI, rendering images or videos while ensuring shared templates for variables.
• Save a manifest for each creative containing variantId, experimentId, briefVersion, creativeVersion, assetHash, generationJobId, and metaAdId.
• Persist state transitions: planned → generating → awaiting_approval → publishing → running → reviewing → complete.
• Checkpoint each worker’s output and store durable job IDs and write operations with idempotency keys.
• Implement approval as an immutable proposal revision; validate proposal freshness before dispatch.
• Handle ambiguous writes by marking timed‑out operations and reconciling against remote objects.
• Collect reporting data with full context: date range, timezone, currency, attribution, purchase action, fetch timestamp, and raw response.
• Normalize metrics and account for conversion delay and traffic allocation bias.
• Generate dashboards and trigger new experiments based on performance insights.

Evidence

Save a manifest alongside each generated creative

Source

Persist a write operation before making the request

Source

The review screen should show the account, creatives, destination URLs, targeting, placements, schedule, and budget

Source