The core diagnosis. Three research personas live the full journey, from application to the goal they came for, graduating, transferring, entering the workforce, or building a specific skill, at all 10 Maricopa colleges. Run at every college, that is 30 distinct journeys, not three, and each one covers every service the inventory identified, financial aid, advising, registration, and the rest, college by college. Personas are given problems, never destinations, and we observe whether they self-serve or disengage. Every stop is logged: the path taken, the time, the ease, the severity, and the AI opportunity. The walkthroughs are run as AI-agent-assisted sessions, a researcher authenticates and directs, an AI agent drives and records the journey at a scale a handful of testers could not, and each service is scored twice: for human usability, and for whether a student’s own AI assistant could navigate it, the emerging practice of Agent Experience (AX). The result is a ranked map of where students hit barriers, and, just as important, where the support that already exists is going unfound.
A participant takes on one of three research personas and works real tasks across the whole student journey while thinking aloud. We never tell them where to go; they receive a goal and must navigate independently through the live systems. The route is 60-plus touchpoints across 10 stages, plus a population track and a swirl track, run over several sittings. Each persona is run at every one of the 10 colleges, so the study produces at least 30 distinct journey results, one persona at one college at a time, not a single pass.
The study is run and analyzed in Notion (research repository, tagging, and synthesis), FigJam (journey maps and service blueprints), and Miro (collaborative workshops and affinity sessions). Findability checks (card sorting, first-click, and tree-test-style exercises) are built and run in FigJam and Miro.
These stages mirror the Initiative 1 service inventory: every one of the 50-plus services in the crosswalk has a stop here, so nothing we identified goes untested.
A session has two pieces. The run sheet gives the tester their assignment for one run: the persona they are, the test login created at their college, the college, and the tasks to work in order. The capture form is where they log each task as they go. The tester reads the run sheet, works the tasks in character, and records one entry per task.
One page that hands the tester everything for a single run: which persona they are, the test login created at their college, the college, and the full task list in order, with what counts as done.
ReadyThe instrument testers use at each task: persona and college, the task attempted, the path they took, time-on-task, ease, and severity. One record per task, 60-plus per run, across the 30 runs (3 personas × 10 colleges).
Live · pilot-readyWe never tell a persona where to go. They are given a real felt need (“I am failing math,” “I need to register”) and must find the path themselves. The gap between self-serve and abandoning the process to call a human is the barrier data.
Two personas enroll at a second college, a second application, residency re-proof, consortium aid, making credits count back home. Maricopa does not operate as “one Maricopa.”
Primary research across all 10 college sites, the GCC Student Frustrations Survey (83 barriers, 4.5/5), and Rio Salado Student Senate notes, all cross-walked into the tasks. See the evidence for the full data.
Every documented barrier maps to a proposed, human-centered AI fix in the AI Opportunity Map, ranked by impact on student success. That hand-off is Initiative 3.
Built with realistic detail so the study fails where students actually encounter barriers. Drafts here are pending validation with real students (design-studio students and a Student AI Group via student government).