Each team of five picks the hero’s they will play the game with. This is like an extra complicated game of rock-paper-scissors. At the end of the picking, one teams composition may be be so setup to exploit the weakness of the other, that the actual 30-40 minute game is almost pointless.
Not mentioned was each team gets to ban out 5 heroes as well.
This is a bigger factor in my opinion. Each team alternates two bans, then two picks, then there is another round of two alternating bans followed by two picks and finally a single ban and pick round for the 5th hero.
Various in-meta heroes are usually "first pick/ban worthy" which means they tend to get picked or banned in the first phase and tend to shape the rest of the draft as teams will build the core of their strategy around the first phase heroes or around countering oppositions first phase heroes.
Another strategy is to avoid "showing your hand" during first phase by first picking strong but generic heroes that can fit into many potential lineups to keep opponent guessing. This leads to a lot of mind games where even commentators don't know what role the hero is going to be played in until the culminating 5th pick when the draft comes together.
Some teams are very good at specific strategies or have certain players exceptionally skilled at individual heroes which necessitate certain first phase bans against them lest they have an advantage.
For instance If a team is known for having a player good at the hero "Wisp" it will often force out a first phase Wisp ban from opponents because it is the kind of hero that when played well can be absolute nightmare to play against.
In some ways I find the draft mini game to be just as interesting as the main game especially in the longer tournaments where you can see new metagames emerging as captains adjust their pick strategies.
There are so many drafting combinations that it's not quite obvious it's a bad idea until you hit the 3 minute mark.
Take OpenAI game 3 as an example. The first two games, OpenAI wiped the floor with the humans and taunted them that they had a >90% chance of winning. The third game, OpenAI was saying the bots had an >80% chance of losing by 5min. The sole difference was the draft.