Nowadays, if you want to start an open data project, you should rather check the availability of information first, and then imagine something useful because so little useful raw data is available. That's why for the foreseeable future, at least in countries such as Germany, collecting documents is the way forward.
Although there are millions of documents available on the Internet, the most interesting ones are hidden in databases, protected as PDF files or only partially offered on websites. It often takes hours to get figures out of a PDF file to be used for analysis. For example, the budget of the city of Frankfurt is offered in a PDF page with more than one thousand pages.
How can you possibly draw different conclusions or see problems and even public misexpenditure? Even a local press paper does not have the resources (any more) to disclose the puzzle of large figure columns. So I am glad to see my friends at Tactic Tools started a project called Open Budget (Offener Haushalt) to shed some more light in the public budget of Germany's government. Other great example comes from David McCandless, who presented them in a great TED presentation called "the beauty of data visualization". However the remaining problem is that it takes a lot of time to extract data, not talking about how to present it.
What are the different ways to get data?
To download open available data
To copy and paste data from documents
To scrape content from websites through software
To collect automatically data from different sources
Or by crowdsourcing your data
Number one and two are the perfect case. Number three can lead to incredible work. Imagine you copy and paste a PDF document of one thousand pages, which is probably printed as an Excel sheet version. Number four is even possible with PDF files thanks to OCR, but scrapping can lead to a load of information.
Number 6 is a very different collective approach. For example, the widely cited Ushahidi is an instrument to offer new channels for data collection if one gets a critical base of contributors. The Guardian uses it these days to track the effects of budget cuts in the city of Leeds.
If you do not have the network or public relation budget to run such a crowd sourcing initiative, then you should think about collecting it from existing sources. For example, a lot of data is offered in RSS or XML format. An advantage is the way data is already referenced, such as date, key words and even locations if you are lucky. A nice tool in this regard is Drupal driven Managing News, which I use also for Create Frankfurt to geo-reference all incoming information.
Such aggregating tools allow automatic collection of data. So you can identify such information, subscribe to it and look at it from time to time. Two examples: Public transport congestion alerts or political municipality documents. A year of such data can give you some insights and might lead to interesting conclusions.