Last week I attended a webinar on developing personas put together by User Interface Engineering. Jared Spool was a bit distractible but explained things in a very approachable way. He described IUE's research methodology for creating personas, which involves 6-12 site visits, extensive note taking, analysis, persona development and incorporation of the personas into the development process.
Here are a couple of take-aways I thought you might find interesting.
Personas Are Not Fiction
A while back I wrote a piece about personas where I talked it about why it's a bad idea to base your personas on a single real individual. Using a real person as a substitute persona might seem like a good idea since you know that everything about that user is for real. The problem, though, is that one individual comes with quirks that may not be representative of the market as a whole. If you interview enough people, you can eliminate their biographical peculiarities by focusing your personas on the characteristics common to the group.
Jared also cautioned against adding a lot of biographical detail to personas, but he came at it from the opposite direction. He warned us to watch out for the temptation to make up background info on your personas. He did recommend giving your persona a name (though not the name of a real research subject for the reasons above) and even finding a reasoanble-seeming photo of that person from stock photography. Both of these are to make the persona more real, easier to remember, and easier to talk about. He said, though, that the color of their eyes, the kind of snacks they like or other details that don't relate to the product you are trying to design are just distractions. He argued that you should eliminate any detail from the persona that you can't connect with a design decision you need to make.
I asked about his inclusion of age as a detail in an example persona they were showing for a medical application and he reasonably replied that age, if representative of the real people you interviewed, could tell you something about the level of computer-sophistication of the user. He emphasized the word "representative," saying that otherwise you were "just making stuff up," and that didn't make for useful design tools.
See a great response Jared posted on his blog to a rant by Jason Fried of 37Signals about personas being "artificial, abstract, and fictitious." Jason sure missed the mark here and I was glad to see so many responders taking him to task.
Clusters Lead to Personas
Another thing I took away from the seminar was a method of extracting data from interviews and turning it into personas. I usually take a lot of notes during customer interviews. When I am finished with them all, I go through and look for key data points like reports people want to see, goals they have, frustrations they have, tools they use, etc. Then I count up how many people use each tool, for example, to get a picture of what people as a whole are using. This is not a scientific, quantitative approach, but it's often directionally helpful.
The folks at UIE have developed a method of clustering these data points about their subjects (they call them "informants" but that's a little too CIA for me) that helps them develop multiple personas. They compare and contrast the individuals they interviewed and plot each on a series of sliders. They pick a subject they interviewed them on, say, computer-savviness, and they plot all the users relative to each other. They do this for all aspects they can then look for patterns in the charts where individuals seem to be close together on several sliders. Looking at what these clusters of neighbors have in common is how you then generate the descriptions of personas.
One of the key things this method does is to weed out irrelevant detail. It might be that all of the neighbors in a particular cluster are young while other clusters seem to skew older, so that's a good bet for a detail to be included in the persona. It might also be that the majority of your subjects overall drive Fords but if you find there is no consistency as to automotive brand choice within clusters or little contrast across clusters, then auto brand preference is probably not relevant to the personas you are going to write up.
I find I like the rigor this clustering concept seems to bring to the qualitative process of developing personas and I plan to try it out on my next project. Hope it's useful for others as well.