Student Journey Barriers Study
AI Strategy · Adoption · Case study
We are not naming specific fixes yet. Those come only after the study is scored and its findings are confirmed. What follows is how a ranked gap becomes a piloted AI fix, who builds it, and how it is judged.
How a gap becomes a pilot
From a ranked gap to a pilot
- Group. When the study is scored, the domain will group barriers into themes.
- Rank on two axes. Reach is how many of the roughly 140,000 students would plausibly hit it. Severity is the Nielsen 0 to 4 rating, the mean of at least three independent raters.
- Plot, not multiply. The two are plotted on a matrix, not multiplied into one score, so a small friction that every student meets sits alongside a rare but dramatic failure.
- Pilot the top. The highest-reach, highest-severity gaps are where the first pilots go.
- Prove it before it scales. Each fix runs in staged pilots against the success measures, is revised, and only then rolled out.
Deciding what to fix stays staff-led; no one is replaced.
The three shapes an AI fix takes
- Build a tool. A small, purpose-built assistant, built with staff and, where it fits, by student teams. Human in the loop.
- Buy or plug in a tool. An existing, vetted tool configured into a system the colleges already use, when that is the lightest thing that helps.
- Create a workflow. An AI-assisted workflow or agent built with the department that owns the service, to take routine, repetitive work off staff.
Students build the tools
The tools that come out of this study are built by three multidisciplinary student teams, drawn from across the ten colleges and led by ARC members. They work the way a design studio works: a brief, a client, a deadline, and a deliverable someone is waiting for.
The line is firm. Students build tools. When a fix turns out to be a change in how a department runs its own process, that redesign is led by the domain, working directly with the staff who own the service.
The intent is to pay the build teams. This is skilled, sustained work with a real deliverable, and it should be compensated as an internship, work-study placement, or part-time position rather than absorbed as volunteer time.
How a team will run: the sprint model
| Element | How it works here |
|---|---|
| The backlog | Not invented by the team. It is the ranked barrier list the study produces, translated into build items. Reach and severity set the order, so the team never argues about what matters most. |
| The client | The office that owns the service, represented by a named staff member from the domain. They accept or reject the increment. A student team building a financial aid tool answers to financial aid. |
| Product direction | Held by the domain, not the team. The domain decides what gets built and in what order; the team decides how to build it. |
| Sprint length | Two weeks, sized to an academic term. A semester yields roughly six sprints, which is enough for a discovery sprint, three build sprints, and two for accessibility, testing, and handoff. |
| Sprint planning | The team pulls from the top of the backlog only. Anything not in the sprint is not being worked on, which is how twelve students across three teams avoid quietly duplicating each other. |
| Daily check-in | Asynchronous and written, not a standing meeting. Students carry class schedules and jobs, so the cadence has to survive people never being free at the same hour. |
| Sprint review | A working demonstration to the staff client, not a status report. The rule is the same one the study applies to itself: show the thing, do not describe it. |
| Retrospective | Fifteen minutes, three questions, one change carried into the next sprint. Kept short because a long retrospective is the first ceremony students stop attending. |
| Definition of done | Accessible to WCAG 2.1 AA, tested with at least one student who is not on the team, and documented well enough that the owning office can maintain it. A tool that fails any of these is not done, regardless of how it looks. |
Timeline
- Summer 2026 · Build and start. All 52 agents are built, and Phase 1 (no login) has run twice, 1a and 1b, with 1c planned.
- Summer to fall 2026 · Approvals and deeper access. District and Domain 1 sign-off, the sanctioned test account, then Parts 2 and 3 as they clear.
- Fall 2026 · Rank the gaps and decide where AI fits first, buy versus build.
- January to May 2027 · Student teams build the chosen fixes.
- Spring 2027 to January 2028 · Pilot in three rounds, then move cleared tools into monitored production while newer fixes keep piloting.
How success is judged
Success is the student’s need met, not an office kept busy. Every fix is measured four ways: whether the need was met, how easy it felt (the Single Ease Question), time to resolution, and persistence over time. These read against a baseline taken before any tool ships. A busy office is not a working journey, and a quiet one is not a failure.
This is proven elsewhere
Connection-focused, AI-assisted support has moved persistence at other colleges. These are the models the design learns from, not findings of this study.
Georgia State University
An AI chatbot answering enrollment, financial-aid, and deadline questions cut summer melt by 22 percent and lifted persistence, used most by first-generation and Pell students.
CUNY ASAP
Wraparound support, advising, money, transit, and tutoring, more than doubled three-year graduation, 53 percent versus 24.6 percent, and has been replicated at more than 60 colleges.
Valencia College
Connected advising raised the match between students’ plans and the courses they actually took from 43 percent to 60 percent, an Aspen Prize model.
Caring Campus
Connection as a staff practice, now at more than 120 colleges, the human infrastructure that makes AI-assisted outreach meaningful.
Sources: Georgia State University’s Pounce chatbot and its summer-melt reduction (success.gsu.edu); CUNY ASAP’s doubling of three-year graduation, from the independent MDRC evaluation (mdrc.org); Valencia College’s advising results, an Aspen Prize institution (Aspen Institute College Excellence Program); and Caring Campus from the Institute for Evidence-Based Change (iebcnow.org).