Key points
- Edith Cowan University is building an AI system called ‘AI Advantage’, using Databricks technology, to help process around 70,000 admissions applications a year.
- The AI pulls together course entry requirements, application information, uploaded documents and existing records to produce an explainable recommendation, with a human-in-the-loop reviewing uncertain or complex cases.
- The university now wants the system to make automatic offers without human involvement and to eventually predict eligibility for courses and scholarships, though no timeline has been set.
Image credit: Edith Cowan University.
Edith Cowan University is working on an AI system to help it process around 70,000 admissions a year, with stated goals to predict eligibility for applicants and to make offers automatically.
The project is being called ‘AI Advantage’ and uses technology from Databricks “to improve business processes with admissions”, director of student administration Peter Corbett told Databricks’ Data+AI World Tour event in Sydney.
“We receive about 70,000 admissions applications a year, and manage all domestic, international, high degree research and credit applications for the university,” he said.
Admissions is one of the university’s “most scrutinised” operations, since it is so closely tied to revenue. It is also highly time-sensitive, associate director of admissions Annette Thompson said.
“In the world of admissions, turnaround time is king,” she said.
“Everyone expects a decision yesterday, and you lose a competitive advantage if you are not meeting your service level requirements as fast as you can.”
However, admissions is also a complex process, not one that’s conducive to being driven wholly by data matched against a checklist of “strict criteria”.
“No one person takes the same path when they arrive at university,” Thompson said, noting that this added complexity to decision-making.
The AI system today
In designing an AI system to assist with the admissions process, Corbett said the university wanted something that could bring together “the relevant evidence that we need to assess an application to determine whether a student is eligible to enter the university” from disparate systems.
It also needed to produce an explainable recommendation – showing how the evidence it collected “relates to the course requirements” for what the prospective student has applied for – and be capable of directing “uncertain or complex applications to an experienced assessor”.
The admissions process is now AI-augmented.
“We’ve got the course entry requirements. We have all those published on our website and our future students website. So we’ve got those as an input [for the AI]
“We’ve [also] got structured application information where the student is filling out the application form in our student management system. We’ve got the documents that are uploaded by the students when they submit their application, and we’ve got the existing records that exist in our own database.
“We’ve passed that through Databricks into an AI-powered assessment. It’s mashing all those things together in the background and then making a recommendation.”
Thompson indicated that the initial capabilities of the AI-augmented admissions process are promising.
“We weren’t striving for perfection. We were striving for recommendations, goals and targets, and [for] a holistic view of an applicant and our admissions cycle – and that’s what the model gave us – but still with the capacity to use a human-in-the-loop to review the decision, use the recommendation in a meaningful way, and finalise an outcome in lightning-quick time,” she said.
Automation goals
While there are operational efficiencies to be had, the work is being pursued through a student experience improvement lens, and that continues to drive the direction of the project.
“Students are at the heart of what we do, and student experience is high on the list of importance for ECU as an institution,” Thompson said.
The university sees opportunities to build on the capabilities of the ‘AI Advantage’, notably evolving the system to be able to make more accurate and appropriate determinations itself.
“We wanted to see whether we needed a human-in-the-loop [initially], and now we’re reasonably confident we don’t need a human-in-the-loop every time,” Corbett said.
“We would like to see the system make automatic offers to applicants, so that we don’t have to have a human involved, and we get higher applicant satisfaction.”
That being said, there will still be some decisions that need to be referred to human assessors, and the intent is not to completely replace human judgement in the admissions process, Thompson said.
In addition to automated decisions, it is hoped the AI system will also be able to perform predictive eligibility for courses and/or for scholarships, although a timeline on this was not discussed.
