Early small-scale machine learning projects can help teams gain faith and learn about the technology. Starting small can also lead to successful collaborations with other teams and improve the accuracy of larger systems.