Introduction: Data as a practical guide for student support

Higher education institutions now have access to far more timely and actionable student data than even a decade ago. The challenge is no longer whether information exists, but whether institutions can organize it into a support system that is ethical, coordinated, and useful in practice. Predictive analytics can help institutions identify students who may benefit from additional attention, but analytics alone do not improve outcomes. Improvement occurs when Institutional Research (IR), student success leadership, and frontline advisors translate information into purposeful action (Bird, 2023; Hussain, Hammad, & Al Qahtani, 2025; Ngulube & Ncube, 2026; Sghir, Adadi, & Lahmer, 2022).

At Indiana University Indianapolis, we have been building that translation layer through a partnership among IR, student success leadership, and academic advising. In an earlier article (Fiorini et al., 2026) we described the analytic development of our approach. Here we focus on implementation: how data-informed strategy becomes day-to-day practice on campus. Our central argument is simple. Predictive models are most valuable when they are embedded in a human support process that is growth-oriented, coordinated across units, and continuously monitored for both effectiveness and unintended consequences.

This partnership has become a cornerstone of our student success strategy. IR helps identify where support is most needed and evaluates whether interventions are working. Student success leaders design the broader strategy, set expectations, and connect outreach to institutional priorities. Advisors create the direct student relationships that make support credible and actionable. None of these functions is sufficient on its own. Together, however, they create a workable model for acting on data without reducing students to data points.

Two principles have guided our work from the beginning. First, we frame the model as a call for institutional responsibility, not as a label for deficient students. We do not use the model to declare who is “at risk”; we use it to determine where the institution must step up earlier and more intentionally. Second, we insist that predictive information must be used with care. Student support can quickly become stigmatizing if outreach is poorly communicated or if staff treat model outputs as destiny (Acosta, 2020; Almalawi, Soh, Li, & Samra, 2024; Ekowo & Palmer, 2016; Imundo et al., 2025; Shafiq, Marjani, Habeeb, & Asirvatham, 2022). For that reason, our process combines targeted monitoring with humane outreach, proactive advising, and careful limits on what information is shared operationally.

Who we are stepping up for, and why

Our work centers on a Priority Population (PP): new first-time, full-time beginner students identified by our predictive model as likely to need additional support during the first semester. The model was designed to predict whether students would finish the term with a GPA of 2.00 or above. At census, the strongest predictors of lower first-semester performance were incoming high school GPA, unmet financial need after institutional and gift aid, later entry into the orientation and enrollment cycle, a more difficult first-term course schedule, and lower levels of pre-college credit. Just as important, our testing suggested these factors were stable across student groups and did not introduce obvious predictive bias by race, income, or gender.

This point mattered greatly on our broad-access urban campus. Any predictive strategy intended to guide student support must be seen as fair, intelligible, and connected to conditions institutions can actually address. We also found that no simple demographic profile explained student outcomes on its own. That result reinforced a practical lesson for our campus: disadvantage is often produced through the interaction of prior preparation, financial pressure, course-taking patterns, and timing, rather than through any single student characteristic.

The census model gave us an initial map of where concentrated support might matter most, but it did not tell the whole story. As the semester unfolded, within-term indicators became even more informative. By week 2, performance on assignments submitted in Canvas had already emerged as the second-strongest predictor of academic difficulty by end of term. By week 4, Canvas performance became the strongest predictor of first-semester success, while negative indicators from our Student Engagement Roster (SER), institutional faculty-generated early alerts, also became highly informative. These findings shifted our approach from static prediction to dynamic monitoring. In practical terms, that meant we could combine an early list of students for proactive attention with ongoing signals that showed when any beginner student, especially a PP student, might need a timely intervention.

Relationship-building, monitoring, and outreach

Our implementation strategy in fall 2024 and fall 2025 had to accomplish two goals at once. We wanted to direct additional attention to students most likely to benefit from it, while avoiding language or practices that might stigmatize them. We were especially concerned about creating a self-fulfilling prophecy in which institutional attention might inadvertently signal low expectations. To manage this tension, we formally tagged PP students in our student information system for internal coordination, but used a neutral group label: “First Year Beginner Outreach.” The tag made tracking possible without broadcasting a deficit-based status to students.

Academic advisors served as the primary operational arm of the strategy, so advisor preparation was essential. We provided training materials, guidance on appropriate outreach, and clear expectations for engagement with beginner students. Advisors were asked to respond carefully to relevant alerts for all students, with particular attention to those in the PP group. At the same time, we deliberately limited the information shared with advisors about why a student had been identified by the predictive model. Advisors did not receive each student’s detailed predictive profile. This decision reflected our belief that the model should guide institutional attention, not shape staff assumptions about a student’s capacity or character. Once the term began, current academic signals such as Canvas Activity Scores (a composite measure of students’ learning management system engagement relative to peers, our best live proxy for performance; Rust & Willey, 2023) and negative SER indicators were more actionable than the original model inputs anyway.

The first element of our strategy was proactive advising. Each fall, we set goals for the number of new beginner students who would complete an early advising appointment during weeks 2 through 7. These appointments were designed not only to answer immediate questions, but also to establish a relationship that could support future intervention if challenges emerged. PP students received especially persistent outreach if they had not yet scheduled or attended an appointment. We viewed this relationship-building as foundational (American Institutes for Research, 2022). A later nudge or referral is more likely to succeed when a student already recognizes the outreach as coming from a known and trustworthy person.

The second element was the use of low Canvas Activity Scores (CAS) alerts. Indiana University already had an automated outreach process that sent students comparative information when their course engagement fell below expected levels. Our campus built on that structure by making advisors and academic support staff more aware of these alerts for beginner students and by connecting students to additional support when needed. Our evaluation suggested that engagement in alerted courses improved, on average, relative to non-alerted courses, so this was an area where analytics and practice were already reinforcing one another.

The third element was faculty-generated early alert information from the SER system. When faculty submitted negative feedback, students automatically received a notification. We then layered follow-up onto that process, especially when students had not viewed the feedback or when multiple indicators suggested a downward trajectory. In effect, our approach linked three forms of support: an early relational touchpoint through advising, ongoing monitoring through LMS-based signals, and targeted follow-up using faculty feedback. The aim was not to create more alerts for their own sake, but to build a coordinated response system around the moments when support might matter most.

Early signs of positive impact

Because the work was intended to be adaptive, IR staff examined evidence of impact as soon as first-semester grades became available for the fall 2024 cohort. Using the census model, they simulated comparable PP cohorts from fall 2021, 2022, and 2023 and compared those groups with the actual first intervention cohort. This matching analysis suggested that fall 2024 PP students generally performed better than similarly situated students from the prior three entering classes.

Across most of the outcomes we tracked, the fall 2024 PP cohort showed encouraging results. Relative to matched prior cohorts, these students were less likely to finish the semester below a 2.00 GPA, less likely to earn D or F grades, and more likely to be retained. One notable exception involved withdrawals and incompletes. Those increased not only among PP students but more broadly. We interpreted this cautiously. On one hand, withdrawals and incompletes can indicate academic difficulty. On the other hand, they may also reflect more intentional decision-making supported by advisors, such as withdrawing from a course before an F occurs. The broader context made the second explanation plausible.

At fall 2025 census, the overall retention rate for the first-year class increased by 2.2 percentage points over the prior year. The gain was much larger among pre-major students than among students admitted directly into academic schools (+4.4% vs. +0.6%). Because roughly 70% of our PP cohort is concentrated among pre-majors, this pattern suggests that the strategy may be helping where support needs are most concentrated. While these results do not establish causality on their own, they offer enough signal to justify continued refinement and investment.

Results for the fall 2025 cohort are so far more mixed, though still informative. Before that cycle began, we improved the model with stronger financial aid data, which reduced the share of the incoming class classified as PP from 42.0% to 34.5%. With a more precise model, average PP GPA and fall-to-spring persistence appeared slightly lower than in the prior year. We do not interpret this as evidence that the strategy became less effective. A more likely explanation is that the enhanced model did a better job isolating students with the greatest academic and financial vulnerability. In fact, the decline in persistence among non-PP students was larger than the decline among PP students, which may indicate that the intervention helped buffer at least some of the challenges faced by the students with the greatest need.

Even so, the gap between PP and non-PP students remains substantial. That is an important reminder that targeted outreach, while valuable, cannot by itself erase structural differences in preparation, financial strain, and course experience. The strength of the current model is not that it has “solved” retention, but that it has given the institution a more disciplined way to identify students for support, evaluate whether response efforts are reaching them, and revise practice based on evidence rather than intuition alone.

Lessons learned and the next mountain to climb

Our experience to date points to several lessons for institutions pursuing similar work. First, local model development matters. Many institutions will recognize familiar predictors such as prior academic performance, unmet need, and course difficulty, but the credibility of an analytic strategy increases when the model is built transparently on one’s own students and tested against one’s own context. That local grounding made it easier for our campus partners to trust the results and act on them.

Second, the most useful analytics strategy is not a one-time prediction exercise. Initial model outputs should be extended with time-sensitive indicators that show how student trajectories are developing during the term. On our campus, the growing importance of Canvas activity and SER indicators gave faculty and staff a clearer sense of where their participation mattered. It also helped campus leadership reinforce the value of timely grading practices, use of LMS assignments, and faculty participation in early alert systems.

Third, institutions should resist the temptation to confuse targeting with care. Identifying students for outreach is only the first step. The more difficult work is building a response system that is relational, coordinated, and respectful. That includes using growth-focused language, limiting unnecessary exposure of sensitive predictive information, and ensuring that outreach invites connection rather than shame. In our setting, proactive advising has been especially important because it turns future intervention into a continuation of an existing relationship instead of a sudden reaction to a problem.

Finally, the model has clarified where our next improvements must occur. If unmet financial need is a major predictor, then financial intervention must happen earlier and with greater precision. If course difficulty shapes first-term outcomes, then departments and campus leaders should examine course sequencing, gateway course design, and the alignment of tutoring and academic support. If within-semester engagement becomes the strongest predictor by week 4, then outreach systems must be ready to act before that point, not after the term is already slipping away.

The partnership among IR, student success leadership, and academic advising has given us a structure for doing exactly that. IR provides the evidence base, student success leaders convert that evidence into institutional action, and advisors make the strategy real for students. This three-part partnership does not replace the broader work of teaching, belonging, and financial support. But it does create a more coherent way to align those efforts. For campuses trying to move from analytics to action, that may be the most important lesson of all.

Christina Downey Ph.D. is the Assistant Director of Business Intelligence at the University of Alabama at Birmingham. She oversees data architecture and modeling, reporting and dashboard development in support of institutional research, reporting, and analytics, with a focus on translating complex data for varied audiences.

Tradara McLaurine, Senior Executive Director of Campus Career & Advising Services at IU Indianapolis, is a three-time alumna of Indiana State University with two bachelor’s degrees one in accounting, one in legal studies and her master’s degree in student Affairs and Higher Education and her certification in Women’s Entrepreneurship from eCornell. She has over 15 years’ experience working in higher education in the areas of career advising, academic advising, student conduct, conferences and events, programming, teaching, and residential life. Tradara currently holds certifications as a Certified Diversity Professional from the Institute of Diversity Certification, Certified Trainer for Understanding and Engaging Under-Resourced College Students from Bridges Out of Poverty, Certified Predictive Index Practitioner and Certified Wellness & Health Coach.

Rishika Samala has 9 years of experience working in Data Science and Research with a master's in data science. She currently serves as a data scientist with the Research and Analytics team, a subunit within Indiana University’s Institutional Analytics office. She has extensive work expertise in machine learning, deep learning, recommendations, and predictive analytics from various domains including automotive, health research and manufacturing.

Stefano Fiorini Ph.D. is a Social and Cultural Anthropologist with the Research and Analytics team, a subunit within Indiana University’s Institutional Analytics office. He has extensive applied research experience in the areas of institutional research and learning analytics. He has published in peer reviewed journals and conference proceedings, presented at national and international conferences (e.g. AIR Annual Forum, CSRDE, LAK, LAP), earning best paper awards from INAIR, AIR and SoLAR, as well as served as chair of projects’ working groups, editorial boards and conference organizations.

Gina Deom has more than a decade of experience working in higher education data and research. She currently serves as the Team Lead of the Institutional Reporting Development unit within the Office of Institutional Analytics. Gina has given several presentations at national and international conferences, including the SHEEO Higher Education Policy Conference, the NCES STATS-DC Data Conference, the Learning Analytics and Knowledge (LAK) Conference, and the AIR Forum. Gina has earned a best paper award from INAIR, AIR, and LAK.

Ram Marupudi has 7 years of experience working in the field of Data Science and Analysis with master's in data science degree. He currently serves as Data Scientist with the Research and Analytics team, a subunit within Indiana University’s Institutional Analytics office. He has extensive experience in Machine Learning Operations and Predictive analytics in verticals like ERI, FSI and Higher Education.