The 72% Problem: Why Most Students Never Get the Help They Need

Here is a number that should keep university administrators up at night: 72% of students with identified academic problems never actively seek help. Not because they don't want it. Not because they don't know it exists. Because the systems designed to provide it were built for a world that no longer exists.

That figure comes from Zhao et al. (2025), a study spanning 435,768 students across multiple institutions. Only 28% of students who were struggling actually reached out to available support resources. The rest sat in silence, fell behind, and in many cases left.

Three stakeholders, one broken system

The student

You're stuck on a concept at 11 PM. Your professor's next office hours are Thursday. You email, and join a queue of 200 students asking variations of the same question. The average student gets 72 seconds of individual faculty attention per week (derived from NCES contact-hour data). That is 1.2 minutes. For context, it takes longer than that to explain what net present value means.

22.3% of students drop out within the first year (NCES, 2023). Among EU students who leave, 50.2% cite program difficulty as the primary reason (Eurostat, 2025). Not financial hardship. Not personal circumstances. They found the material too hard, and the help came too late or not at all.

The professor

Faculty at research universities work an average of 53+ hours per week (Iowa State survey, 2023; N=1,355). They teach 200 students across two sections, supervise graduate research, serve on committees, and publish. Then they answer the same question about whether Chapter 7 is on the midterm. A hundred and twenty times per semester.

64% of professors report burnout (Healthy Minds Study, 2023; N=1,003). 53% have considered leaving their jobs entirely (TimelyCare, 2023). Email and meetings consume roughly 30% of the work week. The time that could be spent on mentorship, research, and the conversations that actually change a student's trajectory is instead spent being a human FAQ.

The university

Every student who drops out costs the institution roughly $21,000 in lost tuition revenue. Replacing them costs another $2,795 in recruitment spending (Ruffalo Noel Levitz, 2022). Meanwhile, universities spend $2,933 per student on academic support services that most struggling students never use.

One college per week is now closing or merging in the United States (Deloitte, 2025). The annual revenue lost to dropouts across U.S. higher education totals an estimated $16 billion. These are not abstract numbers. They are departments being cut, programs being shuttered, and institutions that existed for a century disappearing.

The cost of inaction is not zero. It compounds every semester.

Why existing solutions fail

Office hoursoffer 2 to 3 hours per week for 200 students. 76% of faculty aren't even present during their posted hours (Griffin et al., 2013). The format assumes students can rearrange their schedules to match a professor's availability. It scales to zero.

Teaching assistants are, in most cases, graduate students with no formal pedagogical training. 50% never engage with education research. 60% meet the clinical criteria for burnout. They provide coverage, not expertise.

Online forumslike Piazza and Ed Discussion look promising on paper. Best case: 96% of questions answered. In practice, questions average 4 duplicates each, participation collapses without active instructor presence, and the format fails outside STEM disciplines where answers aren't clearly right or wrong.

MOOCswere supposed to democratize education. Completion rates sit below 7%. MIT and Harvard's edX saw completion rates drop to 3.13% by 2018. 39% of users never performed a single learning action after signing up.

University chatbotshandle administrative tasks reasonably well. Georgia State's Pounce bot improved summer enrollment. But only one randomized controlled trial exists for academic content chatbots. Nobody has solved course-specific, professor-aligned academic support at scale.

LMS platformslike Canvas (39% market share) deliver content effectively. But 81% of instructors report that online courses take more time to manage, not less. An LMS delivers content. It doesn't teach.

The gap we know how to close

Benjamin Bloom's landmark 1984 research demonstrated that one-on-one tutoring lifts the average student above 98% of a control group receiving conventional instruction. This is known as the “2 sigma problem” because the effect size is two standard deviations.

Modern meta-analyses confirm the finding. Kraft et al. (2024), reviewing 265 randomized controlled trials, found an average effect size of 0.42 standard deviations for tutoring interventions. That translates to roughly a 16 percentile-point gain. The evidence is not ambiguous: personalized academic support works.

The problem has never been whether personalized support produces results. The problem has been delivering it at the scale of 264 million university students worldwide.

We know what works. We have always known what works. The constraint was never the pedagogy. It was the ratio.

What changes when help is always available

Early implementations of course-specific AI tutoring show results that are hard to ignore. Praxis AI, which builds professor-specific Digital Twins at universities like Clemson and Notre Dame, reported engagement rates of 75% compared to 14% for generic chatbots. Students using the system showed a 35% improvement in academic performance. 90% of all questions were asked outside office hours (EdScoop, 2025).

Georgia Tech's Jill Watson, running since 2016, found that 66% of students earned A's in sections with AI support versus 62% without. The system now runs across 17 classes in graduate, undergraduate, online, and residential formats.

The pattern across these implementations is consistent. When students have access to course-specific help around the clock, three things happen: help-seeking behavior increases because the barriers (scheduling, social anxiety, email delays) are removed; professors gain visibility into where students are struggling, sometimes revealing gaps they hadn't noticed; and the support is pedagogically aligned with the course because it draws from the professor's own materials rather than the open internet.

The 72% is not a statistic. It is a population.

Behind every percentage point are real students sitting in dorm rooms at midnight, staring at a problem set, deciding whether to ask for help or give up. Most of them choose silence. Not because they lack motivation. Because the system gives them no good options at the moment they need one.

Closing this gap requires meeting students where they are: inside the LMS they already use, at the hours they actually study, with answers grounded in the course they are actually taking. Not a generic chatbot. Not a search engine. A tool that knows the syllabus, references the right lecture, and asks the follow-up question the professor would ask.

The technology to do this now exists. The economics work. The evidence is clear. The question is whether institutions will act on it before the next 72% gives up in silence.

Sources

• Zhao et al. (2025). Help-seeking behavior in higher education. N=435,768.
• NCES (2023). First-year retention and dropout rates in U.S. higher education.
• Iowa State University Faculty Work-Life Survey (2023). N=1,355.
• Healthy Minds Study (2023). Faculty mental health and burnout. N=1,003.
• TimelyCare (2023). Faculty career satisfaction survey.
• Ruffalo Noel Levitz (2022). Cost of student recruitment in higher education.
• Deloitte (2025). Higher education institution closures and mergers.
• Eurostat (2025). Reasons for early leaving from tertiary education in the EU.
• Griffin et al. (2013). Faculty office hour attendance and availability.
• Bloom, B.S. (1984). The 2 Sigma Problem.
• Kraft et al. (2024). Meta-analysis of tutoring interventions. 265 RCTs.
• Praxis AI / EdScoop (2025). Engagement and performance data from university pilots.
• Georgia Tech College of Computing. Jill Watson performance data.