Mats Schmidt

  School of Business and Economics | Bachelor Business Engineering

Deep Learning as decision support in early-stage venture capital

Mats's elevator pitch
Venture Capital (VC) investment decisions are made under uncertainty, and traditional heuristics often fail to capture the complexity of startup performance. My thesis introduces a Multi-Task Learning Long Short-Term Memory model trained on 10,000+ European startups to forecast interdependent indicators of success: employee growth, social traction, follow-up funding, and exit events. By learning shared representations across these tasks, the model significantly outperforms single-task models in accuracy and stability. The results demonstrate how multi-dimensional prediction can improve the reliability of data-driven investment analysis. This research provides a more holistic framework for VC decision-making, helping investors identify high-potential startups with greater precision.