This is the deliverable the first three initiatives build toward. Every documented barrier becomes a candidate AI fix, and before we pilot anything, we rank candidates by one question first: how many students would this actually reach? A sophisticated tool that touches 200 students matters less than a simple one that reaches 20,000. Each pilot is co-designed with the staff and students who identified the need, and answers the simple, factual questions itself while routing the ones that need a person to the right human.
The temptation in any AI initiative is to build the most interesting tool. We resist it deliberately. The ranking starts with reach: how many students each fix would actually touch, drawn from the Initiative 1 baseline. Then each barrier from Initiative 2 carries a severity, and we weight toward the populations the data indicates are most at risk of leaving. We pilot from the top of that list down. The result is a defensible order, not a guess, and a clear before-number to measure each pilot against.
| Ranking factor | Source | Why it counts |
|---|---|---|
| Student reach | Initiative 1 baseline (students served / eligible) | The single biggest filter. Prioritize fixes that touch the most students. |
| Barrier severity | Initiative 2 usability study (0–4 scale) | How badly the barrier hurts when a student hits it. |
| Equity weight | The persistence evidence | Extra weight where the barrier falls hardest on at-risk students. |
| Feasibility | Initiative 1 inventory + IT | Can it run on tools we license now, and integrate with PeopleSoft? |
| Frees human time | Domain guardrail | Does it answer simple questions on its own and route the hard ones to a person, so staff have more time for the students who need them? |
The line we hold is not that every interaction needs a human. It is that a tool should free staff time and protect human connection where it matters, never quietly displace the relationships that keep students enrolled.
We are not assuming that connection-focused, AI-assisted support moves persistence. Institutions have already demonstrated it, with rigorous evaluation. These models set the standard Initiative 3’s pilots aim at, and the pattern across all of them is the same: technology that surfaces the right student at the right time, paired with a human who follows through.
Summer melt down 22%. Withdrawals from unpaid balances down 50%. Persistence up ~3%, about 1,300 more students enrolled.
An AI chatbot answered enrollment, financial-aid, and deadline questions by text, any hour, without judgment. First-generation and Pell-eligible students used it most, exactly the populations community colleges most want to reach. The most rigorously evaluated example in the literature, through multiple randomized controlled trials. [1]
Three-year graduation 53% vs 24.6% for a matched comparison group, more than double.
ASAP addresses every common barrier at once, advising, money, transit, tutoring, career planning, rather than leaving students to navigate supports alone. Replicated at 60+ institutions with the same result. The lesson for Initiative 3: students struggle with an accumulation of barriers, so coordinated connection beats optimizing any single service in isolation. [2]
Match between students’ education plans and actual course-taking rose from 43% to 60%.
An Aspen Prize winner that links faculty, staff, courses, technology, and services into one developmental advising system following students from entry to graduation. The conceptual ancestor of the connected student-support vision Domain 5 is working toward. [3]
120+ colleges implementing; positive retention findings from Columbia’s CCRC.
Not a technology program at all, a behavioral model that spreads specific student-connecting behaviors among faculty and staff. It is the human infrastructure that makes AI-driven outreach meaningful: a tool can surface a student who needs help, but a person has to follow through. Initiative 3 pairs AI identification with this kind of human response. [4]
Students matched with a coach retained at higher rates than those who were not.
We do not have to look outside Maricopa. GCC’s own program, a 2023–24 Maricopa Innovation of the Year finalist, proved locally what the national research shows: connection to a person keeps students enrolled. The open question it raises is the Initiative 3 question, if it works, what limits its scale? If the answer is staff capacity, that is precisely where AI can help a coordinator support more students. [5]
For a factual question, where is my class, when is the FAFSA deadline, what is my balance, a tool can answer on its own, instantly. That frees staff for the students who need a real person. The metric is more human time where it counts, not a human in every interaction.
Pilots are built with the staff and students who flagged the need, not handed to them. Training comes before deployment, because staff ranked training as their top AI priority.
A friendly, peer-to-peer adoption effort distributed at participation moments, so AI tools arrive as something students and staff helped shape, not something imposed on them. A small, friendly object or character, modeled on the stress ball that eased people into the accessibility work they had been resisting, can lower the anxiety and signal that trying AI is safe and optional.
Each pilot has a before-number from Initiative 1 and a target. A year in, we can say whether more students reached the service and whether they persisted, not just whether the tool shipped.
A pilot does not have to be a new app. For each top-ranked barrier we pick the lightest thing that works: adopt a tool that already exists, configure something we already license (for example a Salesforce plug-in, or a feature inside a platform we already run), or build an AI agent or workflow into one of the systems we audited in Initiative 1. Across all three the shape is the same: AI surfaces the right student or the right resource, and a human follows through. CopaMigo, the bilingual routing prototype below, is one example of the build option, but an existing vendor tool or a configured Salesforce workflow could solve the same barrier equally well. And we sequence by reach: we start in the systems the most students already use, so the first pilots help the most students.
The final ranked list comes from the Initiative 2 study and the persistence evidence, but two high-reach directions are already clear, because the barrier data and the persistence research both point straight at them.
Stop requiring overwhelmed students to come find help. Surface it at every touchpoint.
Students do not stop out by failing to try; they disengage and withdraw. They are working two jobs, caring for family, and they often do not know what to ask for. The fix is not one more website to visit, it is the right resource appearing where they already are: in Canvas, at enrollment, in First Year Experience, in the financial-aid flow, in a Discord channel. The more places the help appears, the less a student has to know the name of the office that provides it. CopaMigo, my bilingual routing prototype, is a first attempt at this, a student describes their situation in plain language and receives a direct answer when the question is simple, or a route to the right human service when it is not. It is an early proof of concept, a small first slice of a much larger idea. The direction is to make plain-language help ambient across the whole journey, answering the simple questions and routing the rest to a person, not a single tool a student has to discover.
Faculty notice first. Many still drop students after three absences. That forces non-completion.
Faculty are one of the most trusted sources of referral, and they are usually the first to see a student struggling. But a student missing class after three absences is often a student in crisis, working a double shift, or caring for a sick parent, and dropping them is the opposite of what the data says they need. This is student-support work, not curriculum work, so it sits squarely within our scope. The pilot is a five-minute briefing built for first-year faculty and adjuncts: here is who your students actually are, the basic-needs and work and caregiving pressures they carry; here is why an absence is a signal to connect, not to cut; and here are the few specific things you can do in thirty seconds to point a student to help. It is modeled on Caring Campus, a behavioral approach, considerate of faculty who already carry a heavy load, that gives them ways to help rather than one more mandate.
The five-minute faculty briefing can be produced as a short slide deck for First Year Experience and adjunct onboarding. It is a candidate deliverable for this initiative.
Beyond those two, the data points to a few more places the highest-reach fixes will likely sit. A plain-language needs-intake-and-routing tool, the ambient version of CopaMigo, would touch nearly every student and directly addresses the awareness gap. Proactive, channel-flexible outreach where students already are, text and Discord rather than email, reaches the disengaged students least likely to come find the institution. And surfacing AI tools the district already licenses but underuses, such as BigInterview, is a near-zero-cost win Initiative 1 will help locate. Each is a hypothesis until the ranked list confirms it.
Honest acknowledgment, because a plan that ignores the resistance will stall. We already know the headwinds, and the design answers each one rather than dismissing it.
Buy-in to learn AI is low, because employees worry the real goal is to eliminate their jobs. Our answer is built in: every pilot keeps people in control and is meant to remove routine work from staff so they spend more time with students, not so there are fewer of them. Training comes before any tool, pilots are co-designed with the front line, and a friendly, low-pressure buy-in effort (the stress-ball or character idea above) signals that trying AI is safe and optional.
Many students have been told by faculty that using AI is cheating, so they avoid it entirely. Our tools connect students to people and services; they do not do coursework, which sidesteps the academic-integrity concern. We also work with the student AI training group (our cross-domain liaison) so students learn what responsible, permitted use actually looks like.
AI’s energy and water footprint is real and it matters to our community. We favor tools we already license and right-sized, efficient models over standing up heavy new compute, and we select the lightest workflow that solves the barrier. Not every fix needs a large model.
We name these now, before we build, so the recommendations and the rollout plan account for them from the start.