When you’re dealing with multiple sessions (meaning a website visitor came to the site more than once before converting) the conversion data can be displayed in multiple ways. This is called “attribution modeling.”
It's possible the impression share loss due to rank includes losing out on impressions that the campaign is eligible for but are very broad so you would not want to show for them anyway.
Shared budgets are not ideal but they can be useful.
Most agencies will count a call that lasts 45 seconds or longer, for example, as a conversion. But we try to minimize false leads by hiring a designated team to listens to the calls that come in from an ad and give it a score. Our call rater team also gives us the added bonus of helping us to uncover any issues that might be negatively affecting conversion rates during the calls (like an automated answering machine that is confusing people, for example, or low quality leads who are asking for something different than what your business sells, etc.).
Start new campaigns with a manual bidding strategy until the algorithm has enough data to switch to automatic. Then, see how high you can push the algorithm to automatically raise its goal (i.e. target CPC of $X could be the goal). You must have conversion tracking set up correctly before running any campaigns or else you will be feeding the wrong data into the machine learning algorithms, which will obstruct its ability to get results. You can check this by going to the Conversions report and checking the “Include in Conversions?” column and making sure that only the appropriate conversion actions are included.