* Always choose geographic and language settings properly.
The number of accounts we would review where some (eg.) UK-only business was just throwing away money advertising worldwide ...
Again, Google could provide better tools for targeting adverts, or make the location options clearer, but it is absolutely in their interests not to.
How do you go about doing that? Just decide on an amount upfront manually watch the balance? I didn't see any function to cap the total lifetime spend on a campaign or stop it once it hit a certain level.
Obviously this is not measurement by itself. At that company I wrote a suite of very detailed measurement tools, and we used commercial JS-based tools too (Nedstat, Urchin). My point though is you need to understand the logs first to even know what the tools are telling you.
* Start small, just like the author did. Try to spend as little as possible, either the 1p/click minimum or target the bottom of the page (position 9.9 so you're just "above the fold"). Cap your spend.
* Measure, measure, measure. It's not even worth starting a campaign unless your whole site is being measured in every way possible. Personally I'd recommend tailing the Apache logfiles in real time to start with, rather than relying on an analytics package. You can move to analytics later once you've got the deep understanding from the logfiles. Also, save those logfiles forever.
* Don't rely on Google's tools for choosing keywords etc. They seem designed to waste your money (no surprise there!). Instead, look at the natural and paid search terms people actually use -- by following those logfiles -- and generate your keywords and your negative keywords from there. Be careful with broad-match keywords. We tended to use exact match only.
* Don't use consultants, positioning tools etc. At least, not to start with. They're all expensive and won't be as good as doing it yourself. When you've got the deep understanding, then you can go with 3rd parties if you choose.