Start with 7 day look-back window
For each keyword, compare performance vs AdGroup performance
If statistically significant difference, adjust bid (by 10%?) + flag/unflag keyword so it is not processed in next iteration
Repeat with 14, 30, 60, etc... windows and only process statistically significant keywords that were not processed yet
Start with 7 day look-back window For each keyword, compare performance vs AdGroup performance If statistically significant difference, adjust bid (by 10%?) + flag/unflag keyword so it is not processed in next iteration Repeat with 14, 30, 60, etc... windows and only process statistically significant keywords that were not processed yet