Universities across the country have been financially strained due to COVID-19 operational disruptions. It is more important than ever to manage institutional financial aid budgets effectively and efficiently.
Developing an accurate annual financial aid projection for continuing undergraduate students is challenging. If your institution does not know which of your continuing undergraduates are returning next year, it is very difficult to estimate how much institutional aid will be paid out. Under pre-COVID conditions, financial aid budget errors could reach up to 10% annually. Universities create a budget cushion tying up real money that could be allocated more efficiently within the university.
This challenge is exacerbated during COVID-19 due to changing student behaviors driving enrollment and retention. In this article we address a data-based solution to this challenge and how the solution can be adjusted to account for the impact of COVID-19.
At a high level, we leverage existing student data to predict which continuing undergraduate students will be retained for a given time period. Machine learning and artificial intelligence models are trained for an institution's unique data to predict student retention on a long-term basis. These analytics are applied to develop financial aid projections or forecasts for students who are predicted to return. Our forecasts greatly reduce budget uncertainty, allowing Universities to plan more accurately and allocate funds more efficiently. This reduces the need for traditional budget cushions.
Continuing Undergraduate Student Retention:
Traditionally student retention modeling has been focused on first to second year retention. Similar data modeling frameworks can be applied to continuing undergraduate students as well. An obvious application of these models would be for targeted student interventions to increase student retention rates. In particular, universities can now tackle challenges for ‘late departure’ students, or those students who drop out despite nearing graduation (fig.1). Machine learning models predicting student retention accurately identify at-risk students with a two-semester lead time. In our validation studies, we have found that, absent an active intervention, 86% of students we identified as at risk of dropping out, actually did drop out within a year.
Institutional Financial Aid Budgeting:
A secondary, yet significant and strategic, application of this level of analytics is to develop institutional financial aid projections. These projections are based on which continuing undergraduate students are predicted to be retained over the next year.
These methods have been used for annual financial aid budgeting and planning at a medium sized Midwest university since 2016. A total of 15 machine learning based financial aid forecasts have been delivered and the forecast error has been reduced to less than 3% annually (fig. 2). This has freed up millions of dollars annually that had previously been set aside as a cushion or security fund.
Unlike most machine learning algorithms, we also use highly computer intensive resampling methods to estimate 95% confidence intervals on financial aid forecasts. This provides an upper and lower limit on what may actually be paid out by the institution, providing an added level of security when making operational decisions. With more traditional budget forecasting methods, financial aid projections can exceed the budget and leave a deficit in the annual budget. Depending on the magnitude, shortfalls may lead to many other complications at the University, particularly for those already in fragile financial situations. Understanding the range of possible outcomes and what may be paid out in the coming year is critical.
Our data-based forecasting methods are also highly automated, freeing up time that is usually spent on revising and reviewing spreadsheets.
Accounting for Additional Uncertainty During COVID:
It is indisputable that student decision making processes will be evolving as COVID disrupts normal schedules and educational plans. The good news is that once new enrollment and retention outcomes are known for Fall 2020, the impact of COVID on the student decision making process can be quantified. Retention and engagement factors may be changing due to newly remote or hybrid learning styles and fewer on-campus activities. Once preliminary data is available on student enrollment and retention outcomes for Fall 2020, comparisons can be made to the previous terms. New machine learning models can be developed with new training data and the new enrollment and retention factors can be quantified. We can prepare for the upcoming year, whether students are returning to campus in the Spring or hybrid/online courses will continue into next year.
Managing funds at your institution has become more critical than ever. Reactions to COVID-19 have included housing refunds, adjusting pricing and other monetary strategies in efforts to normalize enrollments for Fall of 2020. Universities are forced to optimize remaining resources leaving no rock unturned in the search for funds. The opportunity to re-invest remaining dollars towards their original goal of improving recruitment, enrollment, and retention could ultimately be a strategy setting institutions up for a competitive future. Positioning for 2021 will require careful planning, development of updated retention and enrollment models and collecting data needed to understand new behaviors that may represent a new normal. The methods proposed above may not only help your institution prepare for the coming year and mitigate risk in planning, but perhaps even stimulate newly available strategies for more resilient initiatives.