OpenAI Embeddings for GTM

Use cases

1. ICP lookalike scoring: Embed descriptions of your best 20 customers. For each new lead, embed their Clay-enriched profile and compute cosine similarity. Score > 0.8 = strong match.
2. Content-prospect matching: Embed your blog posts and case studies. When a new lead enters, find the most relevant content to send them.
3. Competitive positioning: Embed competitor messaging alongside yours. Identify gaps in your positioning.

Implementation

1. Use text-embedding-3-small (1536 dimensions, $0.02/1M tokens)
2. Store embeddings in a simple JSON file or Supabase pgvector column
3. Compute cosine similarity in n8n using a Code node
4. Cache embeddings — recompute only when source data changes

Tips

• Batch embed (up to 2048 inputs per request) to minimize API calls
• Normalize text before embedding: lowercase, remove special chars, trim to 500 words
• Refresh ICP embeddings monthly as your customer base evolves