AI is everywhere in higher education data circles these days. Predictive models, natural language processing, even generative tools are being pitched as the future of insight and efficiency. And while there are genuinely exciting possibilities here, it’s worth pausing to ask a simple question: do we always need AI?
Often, we don’t. Sometimes, a clean dataset and a pivot table get you just as far, faster, and with fewer moving parts to go wrong. Especially when you’re under time pressure or presenting to people who just want a clear answer, the simplest approach is often the best one.
The Allure (and Illusion) of Complexity
AI can sound impressive. And sometimes, it genuinely is. But there’s a growing tendency to reach for complex solutions without first thinking about what the problem really requires. If you want to understand why students are dropping out of a course, a simple comparison between completion status and entry qualifications might tell you more than a black-box model. If you’re trying to spot departments with worrying NSS scores, you probably don’t need a machine learning algorithm, you might just need a basic bar chart.
If you can’t explain the question clearly, an AI model won’t help you find the answer.
Good Data First
One thing AI won’t do is rescue a messy dataset. In fact, it can make the problem worse, producing convincing-looking results that don’t stand up to scrutiny. If the definitions are fuzzy, if fields are inconsistently used, or if no one agrees on what ‘active student’ means, no model in the world will give you trustworthy insight.
Before you start thinking about AI, ask yourself: have I explored this data properly? Can I trust it? Does everyone involved understand what we’re trying to achieve? You’d be surprised how many times a quick logic check or some manual exploration clears things up without needing anything more sophisticated.
Time, Effort, and Maintenance
There’s also the practical side. AI takes time. Developing a model, testing it, explaining it to stakeholders, and integrating it into existing systems is not a light-touch task. It demands expertise, time, and often a level of technical maintenance that’s hard to sustain if your team is already stretched.
Take the example of predicting student withdrawal. Yes, a machine learning model might give you a small bump in accuracy over a logistic regression. But if the difference is marginal and the new model is harder to explain to your leadership team, is it really the better option? Especially if the person who built it leaves and no one else knows how it works.
When AI Really Helps
That said, there are cases where AI is genuinely useful. If you’re working with large volumes of free-text responses, natural language processing can help you spot patterns you wouldn’t pick up manually. If you’re dealing with complex interactions between variables or building systems that need to adapt over time, like a chatbot or a personalised recommender, then AI earns its place.
But even then, it’s worth starting simple. See what your existing tools can tell you. Build up gradually. Make sure the added complexity is worth the effort.
Final Thoughts
In data work, judgment matters. The real skill isn’t being able to run an AI model, it’s knowing when you need one, and when you don’t. Data teams in higher education should feel confident pushing back when something feels overengineered. There’s no shame in doing something straightforward, especially if it delivers what your institution actually needs.
A spreadsheet in the hand really is worth more than an AI in the cloud – if it gets the job done.
