Well, I've come firmly into the camp that says evaluation of keywords is totally broken.
And it didn't take a campaign with thousands of keywords or hundreds of dollars a day. At the point I joined this conclusion, the campaign had six keywords and a budget of $5/day, which it wasn't filling.
After getting the "slowing" message last week (hey, a $5 fee is not small change for this size campaign and it isn't prorated), I had a short exchange with a Google rep (I suppose this is now my account rep). Then Monday morning I deleted all the "underperforming" keywords and added some others to try out. I also lowered the max CPC for all the keywords as another experiment.
And then just twelve hours later I received yet another "slowing" email. Huh? Two keywords got nearly 1000 impressions with no clicks. OK, I could understand disabling those keywords, though I'd kind of like a little time to experiment with a higher CPC first (they'd been running in positions > 14). But slowing the entire account? And of course they won't tell me how soon I have to fix "underperformers" or how often I am expected to check the account -- though this latest episode implies I'm expected to check in every two or three hours. This isn't compatible with Google's emphasis on serving the small advertiser, to say the least.
Oh well. Google was only one thing I was going to try, and the results so far haven't been promising. Looks like it's about time to go back to more traditional methods -- less comfortable for a technologist but apparently much better use of my time. Oh, there is that other campaign ... few impressions, fewer clicks, low cost, no conversions yet ... but Google isn't complaining about the fact that it's underperforming. Might leave it alone for a while.
Does anyone at Google care? When your users perceive your actions as random, they hate it. Long ago research established that users much preferred a consistent 5-second response time to a varying, unpredictable 1-to-5 second response time with a 3-second average. Google is giving us unpredictability without even any benefit. Perhaps all those innovative employment ads in CACM etc are attracting a lot of highly innovative software engineers who haven't a clue about human factors. Well, not knowing isn't stopping me from guessing.