How to build quantitative, data-driven user personas for less than $30
Today I was lucky enough to watch a presentation by Eric Taylor Ph.D., one of Rackspace’s data scientists. He gave a presentation on building fast and rigorous user personas, one which he also gave at SXSW. It was nice to have someone with an academic background break down user personas and be so pragmatic.
In my past work I’ve done qualitative user personas – using interviews to build an understanding of users and to build user personas. These can be expensive and time-consuming. I’ve always viewed surveys as much less DIY – until Dr. Taylor proved me wrong. His methodology allows researchers to move quickly to create quantitative user personas. This gives me a data-driven confidence in my work, and I’m excited to share his system here.
Dr. Taylor’s user persona broke down into two main areas:
- Get qualitative data using a fundamentally sound method for gather data (a survey)
- Use statistical clustering to arrange your data into user personas
Writing an effective survey
Dr. Taylor said in his talk that surveys should follow a few basic data science 101 rules
- Use a standardized scale. He recommended 1-7, because this scale is granular without being overwhelming, and it has a neutral value, which is good.
- Ask questions that are:
- Broad – not too many questions on the same topic
As for the software to build the survey, he recommended 4 but I’m only going to recommend 1 – Google Surveys. It’s free, it’s easy to use, and the results get automatically populated into a spreadsheet.
How to build a survey and find participants on the cheap
The next question is, how to find participants in an efficient way? Again, Eric had a few recommendations, but I’m only going to pass on 2: Hotjar and Mechanical Turk. I’ve used Mechanical Turk with this method, and found great success. I used a free Google Forms to build the actual survey. The advantage of Forms is that you can easily transfer the survey data into Google Sheets – awesome for fast data analysis. For less than $30 I was able to assemble 100 people in a given demographic to take my survey. This completely revolutionized my understanding of fast and rigorous personas.
Using Amazon’s Mechanical Turk
Mechanical Turk, if you aren’t familiar, is an easy way to recruit and pay people to do online tasks. Eric mentioned that you should calculate how much time the survey will take, and pay a fair hourly wage based on that calculation. For instance, if the survey will take 6 minutes to complete (10% of an hour) and a fair wage is $10 an hour, then you should pay $1 per survey (10% of $10/hour is $1). 100 Responses will cost around $100 – that’s incredible value, given that typical surveys take days or weeks and cost into the thousands to complete.
Ratio of questions to participants
To the response rate, Eric mentioned that typical response to question ratio should be 10:1. That is, if you have 10 questions, then you should be looking to get 100 responses, to be sure that a single outlier doesn’t wreck your survey results.
How to organize the spreadsheet
Once the survey is complete, we need to a little housekeeping in our spreadsheet:
- Participants in rows
- Questions in columns
Building the persona
Our first step is to organize the questions into like ‘buckets.’ 10 questions can typically be reduced into 3-4 factors that will influence behavior. Columns can be re-arranged so that like factors are next to each other. The next step is to come up with an equation that gives each factor a composite score. Maybe one question is weighted more heavily than another to the factor at hand.
Each person can then be rated on 3 different clusters – and the averages of those clusters become the backbone of your user personas. Where there are large numbers of people that fall into the same range of scores, that can now be said to be a ‘persona.’ Individuals and outliers need to be placed aside and looked at later.
And there you go! When I tried this exercise, it took one day to write the survey, less than 24 hours to run it with 100 responses, and about a day to analyze the results. User personas took another few days to complete. Given that research like this can take months, I was happy with a 5-day timeline.