In principle we would want to collect as much data for each Keyword as possible to determine the right bid. The possibility to do this is, however, is limited by the costs. Therefore it is necessary to only do so if can be done for low costs or the keyword is promising.
How do I identify the keyword where I can collect data for low costs?
If a low bid is set for a keyword, clicks only occur if they can be purchased to low costs. Therefore you can set a permanent low bid for bad performing keywords (e.g. Longtail). The clicks that are being missed that way, would be to expensive and probably not amortize. The clicks that are created are obtained to a low price and help collect more data about the keyword behavior so that future bids can be adjusted.
How do I recognize if a keyword is promising although it recently did not generate any clicks
Higher keyword bids are justified if
- The CR has been above average in the past (we evaluate the last 120 days)
- Other keywords in the same ad group are exceptionally lucrative
- In principle high bids are profitable in that account (then only the keywords are bidden down where enough data is available to proof that these keywords are performing below average)
If for all keywords data would get collected the costs would not always be at a reasonable level. For this, a short calculation example:
Given is a Long-tail account with 1 million keywords. If the CR equals .1%, we would need 1000 clicks for each keyword in order to have a ~63% chance to observe one conversion. If we would only pay one cent for each click, the costs would be 10$/Keyword, a total of 10 million $. Which obviously is not a sensible approach.
Why are we only evaluating the past 120 days?
Older data often does more harm than help the evaluation. Mostly it makes more sense to use current data from the ad group, as old data for the keyword.