User persona research has primarily been done on Business-to-Consumer (B2C) products, generated hybridly with interviews or surveys or purely quantitatively through clickstream data, often to serve sales / marketing teams. However, the application of these approaches in different circumstances to this is still relatively novel. How can we segment users in a way that is both helpful and accurate for a Business-to-Business (B2B) Software-as-a-Service (SaaS) product and its product development stakeholders, for a product with "imperfect" / obfuscated product analytics data and disjointed user journeys? We performed an iterative, qualitative and mixed quantitative approach to user personas that utilized survey data, user interviews, and preprocessed / manipulated web analytics data to provide a complete picture of the user. Survey data provided user background and insight into who to recruit while interviews provided insight into user journeys, personalities, and pain points. Using dimensionality reduction and an iterative clustering approach informed by survey and interview insights, we generated clusters based on users' time per page and clicks per feature data from our web analytics. Diving into the data behind each cluster and triangulating with interview and survey data enabled us to identify six distinct user personas. We recommend our approach to user segmentation, an as yet unexplored combination of qualitative and mixed quantitative methods, for the research teams of other mid-sized B2B SaaS companies as a way to address domain-specific challenges and bridge the data and user journeys gaps to paint a full, accurate picture of the user.

User Experience (UX) research teams following a user centered design approach harness personas to better understand a user's workflow by examining that user's behavior, goals, needs, wants, and frustrations. To create target personas these researchers rely on workflow data from surveys, self-reports, interviews, and user observation. However, this data not directly related to user behavior, weakly reflects a user's actual workflow in the product, is costly to collect, is limited to a few hundred responses, and is outdated as soon as a persona's workflows evolve. To address these limitations we present a quantitative bottom-up data-driven approach to create personas. First, we directly incorporate user behavior via clicks gathered automatically from telemetry data related to the actual product use in the field; since the data collection is automatic it is also cost effective. Next, we aggregate 3.5 million clicks from 2400 users into 39,000 clickstreams and then structure them into 10 workflows via hierarchical clustering; we thus base our personas on a large data sample. Finally, we use mixed models, a statistical approach that incorporates these clustered workflows to create five representative personas; updating our mixed model ensures that these personas remain current. We also validated these personas with our product's user behavior experts to ensure that workflows and the persona goals represent actual product use.