Latest / Machine Dreams / Can AI Predict When Students Will Drop Out?
Transcript
- speaker-0: you speaker-1: Hi, everyone, and welcome back to Machine Dreams, where we come through the AI hype, talking to the people doing the actual work, building AI infrastructure and AI products that make sense to understand how that's working, what's breaking, and what's next. My name is Colaweli Samuel Adebayi. I'm a technology analyst and Forbes contributor. And as always, I'm joined by my co-host. speaker-2: Hi Sam, I'm Leah Stern, Global Communications Strategist and Venture Capitalist. How's everyone doing today? speaker-1: I believe everyone is doing fine. Great. speaker-2: Today we're talking with Raz Dar, CEO of Elvie. Here's a number that should alarm anyone involved in higher education. Nearly one third of college students drop out every year. And according to Elvie's 26 higher education retention report, 42 % of those students say they fell close to dropping out at some point. Raz's company uses AI to identify students at risk of leaving, giving a heads up before they actually do. giving universities a window to intervene before it's too late. It's also used across other industries, helping them identify risk before it becomes a crisis. Roz, welcome to Machine Dream. speaker-0: Thank you, Leah and Sam. Nice to be here and bringing this really important and critical topic to light, affecting millions of students and actually families around the world. So really appreciate being here and talk to you guys. speaker-1: Thank you. Once again, it's great to have you. I believe there's a lot to unpack, not just from the report, but generally from what you're doing with AI, not just for higher education, but also in contact centers, which we spoke about the last time we had a conversation. But I think this is where I'd like us to begin. Your report says that 42 % of students consider dropping out, or have considered dropping out. And that was really a very interesting statistic for me. that because that's almost half of the entire population and so I'm curious to know what's actually driving that number, what is beneath that statistic. speaker-0: Yeah, actually, to be honest, it surprised us as well. ⁓ It's a very high number. And I actually experienced it myself as well with one of my college kids as well. ⁓ But think about the pressure that these kids are having. It's not just academic. It's a pressure that they have in their entire life. Some of them work. Some of them, of course, academic. Some of them have mental. Over 71 % in the report of the kids that we surveyed indicated that they have mental issues as well. So think about all the pressure that they're having, the frustration that they're having, the struggles, the impact to their self-worth and confidence as a result of them not succeeding academically or wanting to drop out. It's also about letting their parents, who oftentimes pay for their academics, letting them down. and investing for many of the kids, it's first generation in their families. So it's actually a very big problem. And again, 42 % is an astonishing number. speaker-2: It's quite scary to understand more about the stress, the burnout, the financial pressure. And LV can identify at-risk students in advance. So what is the AI picking up on that humans are actually missing? speaker-0: So we identify over, we actually look at the entire student journey. ⁓ nowadays, most colleges and higher ed institutions, they supplement their in-class education and academics with online tools, like LMS learning management systems, tools like Canvas, Blackboard, Moodle. And we can identify based on the student behavior in these tools, we can identify almost everything the student is doing, how many times they log in, how much time it takes them to submit an assignment, how much time they spent in a class or not spent. And ⁓ of course, the grades, of course, if they submitted or did not submit an assignment or submitted late, we look at, as we said, over a hundred of different signals across the board. There's not one sign or signal as we call it that can indicate a sign of disengagement. If I'm submitting, if a student submits an assignment late, does it mean that he wants to drop out? He may have a certain issue. It might even be his pattern. He always submits an assignment late. But when you correlate all these together and you see that it accompanies with not enough time spent in a class or starting a class and dropping in the middle and all these other signals that I talked about can indicate a sign of disengagement, as we call it. which can potentially lead to dropout. And the idea here is to identify it early on when there is enough time to intervene and course correct the situation. So we're not waiting for the grades to fall or to see an attendance miss. We look at those signs of disengagement and now is the time to intervene and help and before it's too late. speaker-1: Great. Raj, you know, the last time we had a conversation, you spoke to me a lot about what you're doing in contact centers with your AI solution, helping to predict employee behavior and helping businesses manage their employees better. And so now you've sort of moved that technology into higher education. And so I'm wondering, what is the correlation between contact centers? And is there correlation there? Absolutely. did you choose higher education? speaker-0: So we actually expanded. We didn't pivot. We didn't change. We expanded. We're still supporting the contact center and the workforce retention and attrition. I'd love to find a way to connect the two, by the way. If you can monitor a student from enrollment to graduation to his workforce and see his entire journey and help him succeed in both of these worlds, that would be amazing. But we actually see, to your question, we see a lot of correlation. In terms of the personality, what causes people to a trip, to leave the work, their employment, is sometimes burnout, it's lack of motivation, they're not meeting the performance or the expectations of their managers, which oftentimes you find similar signals in education. burnout, need to spend a lot of time, you know, learning for submitting assignments, learning for the final exams. Sometimes they fail, which is fine. I failed the course when I went to school. have an engineering degree and I failed the course and had to repeat it. It doesn't mean that I wanted to drop out, but it happens. So there's burnout. It's a long period. You go to school for four years. sometimes. ⁓ so there's a lot of similarities in terms of the human behavior, which we can identify and dissect and analyze using, nowadays, using AI. ⁓ speaker-2: And that's just so interesting because you have access to a large amount of data. And when you talk about burnout, which is a huge problem globally, one in five in the UK workers take off due to stress and 77 % of US employees have reported feeling burnt out. So it's a common problem, not just in the US or in the UK, but globally. And I wonder based on the data that you're looking at, are certain companies or universities doing something better or different where you see lower retention rates or lower reduction rates due to maybe the culture in a company or university? What are some of the insights that maybe companies and universities can take away? speaker-0: Yeah, great question. So if we focus on universities, we see that there's a new role in a university that has evolved in the last few years called student success. And not all colleges adopt this mentality. So student success, and there are colleges and universities that we work with that have dozens of advisors in their student success department, that their entire job in life is to identify those students that struggle. Now they have an enormous amount of tools in their portfolio to help a student once they identify him. They can provide mental support, they can provide tutoring, they can provide financial aid, can provide ⁓ just, ⁓ you know, counseling. These tools, there is a lot of capabilities that they can provide. The idea is to identify these students early enough when there's still enough time to make a change. So, and what they do and where we help them is to not only identify, so identifying early on is a big part of it. It's really important because you still have time to intervene, but also monitor the intervention because oftentimes they work with Excel. And they identify a student based on grades and attendance, which is sometimes the student is halfway out the door already. He already opened a huge gap in his academics and he feels that it's almost impossible to ⁓ fill. But even if they caught him early and they provided the support, they provided the tutoring, they don't have the tools today to monitor the student activities from that point onwards and make sure that he's re-engaging back. I call it from disengagement to re-engagement. And based on the activities, based on the monitoring, that we use now AI and the tools that we develop, we can detect those signs. So we can report back to those student success teams, your intervention worked, or if it didn't work, let's do something else. Let's maybe be more aggressive. Let's try to contact the student and see what else can be done in order to retain it. speaker-1: Great. know, Raz, you talk about, generally when we talk about predictive systems, AI systems that are able to predict behavior or a certain pattern for a certain group of people, bias always comes to the fore, right? So I'm curious about, you know, if there was a situation where your data showed that students from, you know, certain zip codes or income brackets are more likely to drop out than others. How do you separate actual signal risks from demographic correlations? speaker-0: Yeah, we try to avoid looking at personal data and PII. We look at the student behavior. We don't correlate, we don't look at gender, we don't look at race, we don't look at these things. So we try to look objectively at the student activities and identify based on his activities, signs that he's disengaging. So we are very, very, we are GDPR compliant. We are SOC 2 type 2. in the company, we pay a lot of attention. Now we're working on our AI policy as well. pay a lot. It's very, very sensitive. Student information is very, very sensitive. So we do whatever it takes to secure that information. speaker-1: You seem to be doing great at handling the issue of bias with your AI technology. What would be your piece of advice for the general industry or people who are in your kind of space where they are predicting behavior? How can they get around the issue of bias? speaker-0: It's a good question. I think we insist on having a human in the loop. So when we provide our guidance or our recommendations that there's a student at risk and sometimes we even provide recommendations for an intervention, it's for the student success advisor. So he's taking the final decision. He's the one talking to the student. He used the data and the signals and the information that we provide to empower him or her ⁓ and help him make the right decision at the right time. But eventually we, and we, and we discussed this a lot, whether our tool, what we develop should communicate directly with the student and help the student, you know, with AI, we can do that. We can create a great avatar as well. I think that at the end of the day, there are those subject matter experts. So this is what they do. We are here to support them, make it visible, help them do their job better, and eventually retain those students. speaker-1: Bras, thank you so much for joining us. This was thoughtful and incredibly relevant. ⁓ speaker-0: Absolutely, happy to be here. speaker-2: And thank you to everyone listening. This is Machine Dreams. We'll see you next time.