As UK universities grapple with severe financial challenges, the need for them to optimise their processes has become more critical than ever. Many in the sector may naturally look to AI as a means of generating efficiencies and reducing costs where they can. However, this appetite for AI exploration brings with it greater expectations. Any technology investments universities make absolutely must deliver value.
The student recruitment journey is ripe for AI deployment, with numerous opportunities for the technology to have a huge impact. But knowing how and where to use it can be challenging. It requires insights into what students really value, and an understanding that robust datasets – on which AI models are built – are hard to come by in higher education.
Student-facing vs back-end applications
Often, we see universities drawn to AI for student-facing solutions, namely chatbots. This can be partly attributed to not wanting to appear behind the curve, but it can also seem like a viable solution to a real pain point. Many institutions face high website traffic and enquiry volumes, and a chatbot can appear to immediately address some of these incoming questions.
So far, so good, in theory. In reality, though, ensuring chatbots are actually effective requires serious dedication to updating the information you’re feeding it. To train a chatbot on machine learning models, typically a university’s website is scraped to inform the model. Failure to regularly maintain each webpage on the site (something we know university marketing teams struggle with) means that a chatbot will be regurgitating outdated information – or worse, hallucinating and providing inaccurate information to fill gaps.
We know from research with prospective students that this is often a source of frustration. In almost all cases, they prefer speaking to human members of staff. For universities it’s a better use of resources to retain human interaction where they can and deploy AI for the behind-the-scenes, high-volume processes that students don’t necessarily see, but certainly feel.
A good example is using AI to identify what’s working well to inform future practices. Analysing the engagement and impact of outbound communications to extract performance patterns would be a lengthy project for a human team, but AI tools can do it in seconds.
Take email marketing as an example. AI technology can rank how different email components contribute to overall engagement, factoring in everything including subject lines, tone of voice used, and the emotion it is likely to evoke. It can even analyse differences in domains, looking at how well an email performed on Apple mail versus Gmail.
This builds a really clear picture of what drives positive engagement for students, and, because the AI is learning on a real-time basis, it can adapt to student behaviour and preferences in-cycle. Over time, the AI can provide feedback on draft email campaigns, helping to finesse the content and subject line to deliver maximum impact.
It also allows send time optimisation to be honed at an individual level to deliver personalised interactions at scale. If open/read data tells us that a prospective UK student is a night owl, it might render traditional 9am-5pm parameters meaningless for this individual.
This type of insight can shift thinking away from “everyone needs to receive this deadline email at the exact same moment” to a hyper-personalised approach – one that recognises that engagement might be higher if an individual receives it six hours, or even two days, later. Where we’ve deployed this approach with our partners, it’s delivered on average a 40% increase in open rates without tweaking the content of the email itself.
Machine learning predictions
Emerging areas for AI in higher education enable this optimisation to go even further. Propensity modelling is one example, and is something we have recently introduced to our partners.
Trained on machine learning via millions of data points, propensity modelling tools can identify offer-holders who are most likely to accept their offer, and the specific interventions that will have the greatest impact on that decision. The model learns distinguishing features between those who are likely to accept or decline their offer.
The implications for institutions here are huge, particularly in the current financial environment. This type of AI enables effective in-cycle decision making and forecasting, meaning human interventions can be hyper-focused on where they matter most.
However, the nature of higher education means it is only possible to build this type of model based on the availability of broad, deep data. It requires a large, aggregated data-set to identify what is a genuine pattern as opposed to random coincidences, or just the consequences of an individual institution’s own actions. Those institutions that can tap into these data-sets will see the pay-off when it comes to student conversion.
Retaining human oversight
In other parts of the student recruitment journey, we need to draw an important distinction between partial automation and true independent intelligence, and consider how far we want AI to go. Admissions is the perfect example.
Where institutions struggle to process the volume of applications they’re receiving, automation tools can streamline these applications based on whether a candidate meets a set of pre-defined criteria – automatically filtering out those who don’t and reducing the need for human intervention on every single application.
It’s vital, however, that a Human-In-The-Loop (HITL) layer is always retained here to allow for nuanced decision-making, reducing the risk of any machine-learnt bias creeping into admissions functions. Unlike other sectors such as e-commerce which are designed to be as slick as possible, higher education needs some ‘friction’ to maintain its integrity.
The big picture
AI’s potential in the student recruitment journey is nothing short of transformative, deployed in the right way. Its most effective uses lie in back-end, analytical tasks rather than those which seek to imitate human interactions, and its ability to identify patterns at scale can’t by matched by human teams.
When machine learning algorithms come into play, it’s almost impossible for universities to build anything meaningful here with just their own data – so, if they’re working with external partners, institutions should scrutinise the robustness of the data-set.
Not only do intelligent AI deployments reduce inefficiency, they can also generate a really solid return on investment by contributing to a higher yield of enrolled students in an ever-competitive landscape.
Increasingly, students hold institutions to the same standards as other brands they engage with. They expect universities to be more responsive than ever, across more channels. They want a higher degree of personalisation. And they want to be served the right information at exactly the right moment in the journey. Institutions that get this right will reap the rewards.
