Harmattan AI
Harmattan AI's pitch faces the most common challenge for vertical AI startups: how to convince an investor that a specialised model is defensible when foundation models improve every six months?
Harmattan's answer was architected in three layers. Layer 1 (the problem): general LLMs fail in financial markets because they hallucinate, are not traceable, and are not trained on proprietary sector data. Layer 2 (the data): Harmattan had built partnerships with European financial institutions to access historical datasets that do not exist in public internet crawls. Layer 3 (the defence): every new data agreement strengthens the model, which improves results, which brings new data agreements — a flywheel that general competitors cannot replicate.
The pitch narrative structure was deliberately non-technical in the early slides: "Financial markets move trillions of euros per day based on decisions where a 2% error costs millions. The AI models available today are not suited to this context. Here is why." Only after this premise did the technical solution appear.
For founders: the most effective vertical AI pitch always starts from why general models are not enough, not from why your model is superior. It forces the investor to agree with the problem before evaluating the solution.
The first slide of the most effective vertical AI pitch does not show your solution — it shows why existing solutions fail unacceptably for that specific market.