Creating personas from actual online user information is an advantage of the data-driven persona approach. However, modern online systems often provide big data from millions of users that display vastly different behaviors, resulting in possibly thousands of personas representing the entire user population. We present a technique for reducing the number of personas to a smaller number that efficiently represents the complete user population, while being more manageable for end users of personas. We first isolate the key user behaviors and demographical attributes, creating thin personas, and we then apply an algorithmic cost function to collapse the set to the minimum needed to represent the whole population. We evaluate our approach on 26 million user records of a major international airline, isolating 1593 personas. Applying our approach, we collapse this number to 493, a 69% decrease in the number of personas. Our research findings have implications for organizations that have a large user population and desire to employ personas.