We present our work on the introduction of expert knowledge in a Bandit-Based Monte-Carlo Planning algorithm applied to the game of Go. The contributions include (i) opening books (ii) bias in the tree part (iii) improvement of the Monte-Carlo playouts and point out general elements around Bandit-Based Monte-Carlo Planning, namely the risk of diversity loss in random playouts, the efficiency of counter-examples for Monte-Carlo design or for the introduction of bias. The resulting program has recently won a non-blitz game against a professional player in 9x9 Go.