Does your data answer real customer questions, or the ones your data model supports?

I worked on a project with two sociologists from the Future of Music Coalition a couple years ago. They conducted the first large-scale formal survey on musicians' revenue streams to help answer the question "how do working musicians make a living?" It was a really fantastic project funded in part by the Doris Duke Charitable Foundation and YouTube, and I worked with them to take all the survey data and turn it into an exploratory portal that musicians could use to learn about their peers and income streams.

Today, I'm going to share one key finding from this that had a lot to do with the quality of the UI/UX we came up with in my opinion: the survey has to ask the right questions to enable customers to get the right answers. The design of the survey itself (conducted years earlier) had a direct impact on the portal's ability to answer the questions we expected portal users to ask.

When we went through a process to define some personas (archetypal users), we found out that many of the questions musicians would want to ask couldn't be answered by the data that had been collected. As such, the portal UX was heavily constrained by the choice and design of the original survey questions.

If you've worked with engineers before, then this may not seem very foreign as you know that your data is a limiting factor. You can't show data you don't have.

The lesson today is to get the design thinking happening as early as possible, well before you do system architecture, or in this case, designing the stimuli (survey questions).  Had we gone through the process of understanding what questions the target users might want to ask at the beginning, we could have potentially adjusted the survey's design to account for these future needs. Additionally, it's much harder to fix this type of problem with a survey–if you hope to run it over multiple time periods. While technically the changes aren't hard, the response data will potentially be biased if you change the survey stimuli (questions) between solicitations and want to show historical trends.

But, survey design is a topic for another mailing 😉

Nonetheless, the team was happy with what we came up with given the project constraints, and as with all designed products, quality is usually not a binary measurement. At some point you have ship!

View the portal and learn about musicians' revenue streams at .

Want More Insights on Designing for Analytics?