Fast and rigorous personas are a data-driven method of my invention that leverages statistical cluster analysis to create rich user and customer personas that provide product teams with insightful recommendations in a short amount of time
How I leveraged my initial weakness in qualitative research upon starting my career to start a quantitative UX research powerhouse.
When I first started as a UX researcher, I had strong statistical skills, not a ton of experience interviewing users, and no time for creating personas. So, I did what I knew how to do. I ran a survey on a website that had some traffic and came up with my own way to create user persona. That was at CarStory (acquired by Vroom $120M) over 10 years ago.
This method uses unsupervised statistical approach to build cluster-based segments from the bottom up. Specifically, I perform hierarchical cluster analysis to identify groups of respondents who give similar answers to questions to a survey, specific categories of interest, or patterns of usage of a a digital product.
A published paper (Mereu at al., 2017) suggests that this clustering analysis offers a different and more refined lens for interpreting user data and is able to reveal sizeable differences that are often surprising, yet sensible after a deeper analysis. Most importantly, the differences across clusters tend to suggest actionable insights to the product teams precisely because they are emerging from the facors that matter the most in the data as shown in this picture (the darker the color, the more relevant the question in the survey was for the factor, which is then used to conduct the cluster analysis).
The methodology offered me the opportunity to present in front of a full crowd at SXSW, give workshops at Google, and create dozen of personas in a fraction of the time I would if I had to rely on interviews alone. Don’t be fouled! I am not denying the importance of qualitative user research. I think that each project might benefit from one or another, or even both, depending on the circumstances.
See the published paper below.
