How it works

We learn how good heroes are against each other from recent Dota2 matches played in the high skill bracket.

The process is fully automatic, using machine learning, and does not rely on any particular player's personal experience or votes. The data speaks for itself and our suggestions reflect what usually works for the pros.


Our growing database includes over 1 million matches from the high skill bracket where players did not disconnect or leave. For 63% of these matches, our method predicted the winning team correctly based on picks alone.

Powered by Steam


Try using our tool with the in-game browser. Press Shift-Tab, then select "Web Browser" and go to

Pick and ban suggestions are sorted in order of importance. Follow the suggestions on the left first.

You can use the Arrow Keys, Delete and Backspace in the team counter lists.

The algorithm

We model hero picking as a zero-sum-game and learn the game matrix by logistic regression. When suggesting picks and bans we assume that teams are mini-max agents that take turns picking one hero at a time.

Our team's picks / Suggested picks:


Opposing team's picks / Suggested bans:


Bans / Do not suggest for our team:


Chance we win based on picks:   68%

(clear all selections)

Warlock Weak against:
Strong against: