O'Reilly recently released one of their free Radar Reports titled Data Jujitsu: The Art of Turning Data into Product by DJ Patil. This short (24 pages) report is packed with interesting insights and actionable information that will change the way you look at your next project. It's very much worth the minimal investment in time to download and read.
Contents: Data Jujitsu; Use Product Design; When In Doubt, Use Humans; Be Opportunistic For Wins; Ground Your Product In The Real World; Give Data Back To The User To Create Additional Value; No Data Vomit; Expect Unforeseen Side Effects; Improving Precision And Recall; Subjectivity; Enlisting Other Users; Ask And You Shall Receive; Anticipate Failure; Putting Data Jujitsu Into Practice; About The Author
The author thoroughly knows his topic, as he is a data scientist with real-world experience working on well-known sites such as LinkedIn. That is an example of a site that is driven heavily by data and connections between data. Observing what the site is actually used for, as well as what people attempt to accomplish, is what adds the additional value over time.
Given that this is a report rather than a "book", I found that the ideas and actionable content could be found on nearly every page. I thought his approach to solving the data issue prior to solving the application problem both refreshing and insightful. Why spend a fortune (both time and money) to create a product only to find out that no one has the problem you want to solve? Or worse, you spend far too much time solving the wrong problem (usually the more complex and "interesting" one) rather than the one that people would spend money to have solved. A cheap and usually non-scalable approach to collecting the data is preferable, as it allows you to figure out what needs to be the focus and what can be ignored. An example of that approach is Amazon's Mechanical Turk. MT can be a quick way to use humans to solve a problem (such as visual recognition) instead of spending untold amounts of time building a computer-based visual recognition system that may be tangential to the actual problem you're trying to solve.
Another interesting point was to have a conversation with the user instead of just having them fill in data. If you can design the interface to help the user enter "clean" and relevant data, while at the same time giving them feedback and value, they become your partner in finding answers and relevant links between data points.
Finally, I can't let my favorite title for a chapter/section go unmentioned... Data Vomit. It's what happens when you decide to focus on the amount of data coming back instead of focusing on what people are doing with it. Focus on actions, not volume of data returned. This hits home as it's very common to have users ask if we can generate data in formats x, y, and z. When you ask *why* they want that feature, the answer is often "I just thought it might be nice to have." Instead, I need to focus on what they actually want to accomplish, not on how many different contortions we can twist the data into...
Go ahead and download the report from either Amazon or O'Reilly. It'll take less than an hour to read, and you'll get far more than that in return.