Training decision models from scratch is expensive and time-consuming. A library of open-source base models would let teams start with something already capable.
Create a GitHub-style platform where researchers share their trained models. Charge for hosting compute-heavy models or offering fine-tuning as a service.
Start with basic models for common business decisions like pricing or routing. The risk is that proprietary models may outperform open ones for competitive advantage.