Here are some figures. Not sure what to make of them, or how they compare with your own.
Impressions Clicks % of clicks to imp sales % of sales to clicks. Average Bid cost ACOS Keywords
Book one: 905,299 801 0.09% 87 units 10.86% $0.08 25.83% 404
Book two: 765,997 355 0.05% 20 units 5.63% $0.06 38.36% 687
Book three: 749,453 337 0.04% 21 units 6.23% $0.07 36.84% 476
Book four: 529,227 279 0.05% 21 units 7.53% $0.06 24.88% 471
Book five 351,449 281 0.08% 14 units 4.98% $0.03 22.15% 354
TOTALS 3,301,425 2,053 0.06% 184 units 8.96%
Note: I haven't ever stopped any of my ads and started them again as new ads, I have only paused them if I had a promo for say 1 day, and impressions have kicked off again when released back on sale.I have however, increased/reduced bids, and added or removed keywords, but basically I just now let it run and have done for some time. I understand the more data they have, the more likely they are to keep on giving impressions (to a point)
The only thing I have noticed is that if I run a free day promo and end up with sales giving me a better rank, I also seem to pick up sales on my AMS for that book, but once the rank falls back, sales are few and far between on AMS.
Also my book that has had most sales organically without ads, has picked up the most sales on AMS. So I'm thinking rank has something to do with clicks converting to sales in the reader's mind. At the moment it doesn't have a good rank and sales are slow for it on AMS, yet impressions soldier on.
I don't hold with the idea that they stop impressions forever if the don't produce clicks. Okay, they put a temp hold on it a week or so into a new ad and intermittenly after that while they assess the data, I've experienced that. I have one with over 5,000 clicks on a Stephen King book and no clicks, but I still keep getting impressions. I have many more examples like this.
The other thing I note is that book 2 & 3 aren't performing as well as book one, yet the impressions which were well down on book 1 (as much as 50% down) are fast catching up with that book. So that debunks the idea that they stop poorer performing ads, unless they only stop those with a rank bad ACOS that makes a loss to save us from ourselves.