Foundation Models vs. Domain AI
Foundation models are large, general-purpose AI models (GPT, Claude, Gemini) trained on broad internet data. Domain AI is artificial intelligence specifically built, trained, and optimized for a specific industry or operational domain using proprietary data, specialized architectures, and domain-specific evaluation criteria.
Why It Matters
The build-vs-buy decision often reduces to this question: can a foundation model with good prompting match a domain-specific system? For supply chain, the answer is no, because the critical intelligence (carrier behavior by lane and season, facility dwell benchmarks, exception resolution precedents) does not exist in foundation model training data. It exists in operational systems like the FourKites Graph.
The FourKites Perspective
FourKites uses foundation models where they excel (natural language understanding for FourSight queries, document parsing for Sam) and domain-specific AI where operational precision matters (ML ETAs trained on 116 million actual loads, carrier reliability models trained on 16,842 SCACs, exception resolution models trained on 2.5 million monthly decision traces). The combination is more powerful than either alone.