I'm not a developer but as far as I can tell the thing is.
1. No bot is as good as a decent human player,(edit) at least not with all types of decks.
2. There are both specific cards as well as deck archetypes that are exponentially harder for any bot to play.
3. Some cards are only good if you put them in certain deck archetypes, a small time buccaneer is broken in warrior and shaman but for a non weapon class it's useless.
4. This bot is not made for arena, it's made to farm gold, hero portraits and, for some decks, reach high ranks.
With that in mind you have to understand, the developers, already hard pressed as they are to maintain and optimize the bot for high ranking decks, can not give too much of their time to optimize a form of the game where you have no real knowledge of what type of deck you are going to end up with.
The bot can play an aggro or tempo warrior on arena, but if you draft a control oriented warrior, you are gonna have a bad time.
It makes much more sense to optimize the few archetypes the bot can play well in constructed mode.
The only way to make a bot capable of playing arena as good as a really good player, like an average of 6 wins is to make a behemoth A.I. that would possibly rival IBM's Watson.
Honestly arena is one of the few actually fun aspects of this game now that constructed is either aggro shaman/warrior or Reno decks.
Play it yourself if you want to gain extra gold cause as I said in point 1, no bot is as good as a decent human player.
And in the end, arena simply is not the focus of this bot.