AI Marketing Creative Engine

Problem & Operational Bottleneck

Listing Profit's manual campaign production required copywriting, formatting, and multi-channel asset delivery. Each campaign took about four hours and cost about $300 to produce, creating an operational bottleneck as volume grew.

Constraints & Requirements

• Consistent brand and campaign formatting requirements
• Structured outputs had to remain parseable across generation steps
• Transient API timeouts from downstream LLM and image providers
• Operators needed visible checkpoints before multi-format delivery

System Architecture & Ownership

Built a production n8n and OpenAI pipeline that accepts campaign inputs, generates schema-constrained content, applies QA checkpoints, and prepares multiple delivery formats. The surrounding integration layer connected campaign and lead workflows to Follow Up Boss, GoHighLevel, and webhook endpoints.

Engineering Decisions & Rationale

• Used n8n-style orchestration for rapid iteration while moving reliability-critical logic into Python services as volume grew.
• Required structured outputs at generative steps so downstream automation received predictable fields.
• Separated deterministic data transformations from generative LLM nodes to reduce token spend and increase testability.

Reliability, Retries & Safeguards

• Idempotency keys protected webhook-triggered workflows from duplicate processing.
• Exponential backoff handled transient failures from external services.
• Dead-letter handling preserved failed operations for auditable recovery.
• QA checkpoints validated generated content before multi-format delivery.

Reflection & Engineering Learnings

The project reinforced a practical rule: generative steps need deterministic validation and explicit recovery paths before they can support a repeatable production workflow.