Tactico employs a two-phase training approach: first using tens of thousands of human games for imitation learning, then using millions of self-play games for reinforcement learning to converge toward a Nash equilibrium.