How Our AI Matching Engine Scores Your Partners

When you see a match score of 87/100 on Baltmodus, that number isn't arbitrary. It's the output of a hybrid scoring system that combines five distinct signals:

1. Category Overlap (0-15 points) Do your industries align? A food manufacturer matches better with a food distributor than with an IT consultancy. We compare the category arrays on both company profiles.

2. Geographic Overlap (0-15 points) Are you operating in the same markets? If Company A lists "Germany, Lithuania" as operating markets and Company B lists "Germany, Poland," the shared market (Germany) earns full points.

3. Offer ↔ Need Alignment (0-20 points) This is the directional match. Your needs are compared against the other company's offers, and vice versa. Token-level matching catches semantic overlap — "dairy supplier" matches "organic milk products."

4. Partnership Compatibility (0-10 points) Does the partner's business type match what you're looking for? If you're seeking a distributor and they ARE a distributor, that's a direct hit.

5. AI Embedding Similarity (0-40 points) The heavy hitter. We use OpenAI's text-embedding-3-small to create vector representations of each company's full profile text. Cosine similarity between these vectors captures nuances that rule-based matching misses — like understanding that "cosmetics" and "beauty products" are related.

The cold-start solution: Rules-based scoring works from day one, even before embeddings exist. New companies get meaningful matches immediately based on their structured profile data. As they post opportunities and the system generates embeddings, match quality improves automatically.