Data shows that 88% of university administrators agree their institution must invest in analytics to stay competitive. Yet only 40% say they can effectively act on their data. The gap isn’t about technology or talent. It’s about who can actually get to the data when they need it.
So, how to fix that?
This article breaks down the most common reasons universities struggle to act on their own data, outlines four practices that actually work, and shares what we learned building an AI analytics system for a US university. Read on to find out how questions that used to take days can be answered in minutes.
Why Most Universities Struggle with Data-Driven Decision Making?
Picture a typical Monday morning at a mid-sized university. A VP of Enrollment is preparing for a leadership meeting. They need to know how many students registered for next semester compared to the same point last year. Simple question, but not a quick one.
Instead of pulling the number themselves, they send a message to the analytics team. Not because the data doesn’t exist, but because accessing it requires SQL, system knowledge, and time that most staff simply don’t have.
By the time the answer arrives, the meeting is already over. Leadership didn’t ignore the data. They just couldn’t wait a few days to make a decision that needed to happen today.
The issue isn’t that institutions lack data. Data shows that 95% of institutions are actively investing in data and analytics projects. Student information systems, learning management platforms, financial aid databases, enrollment records — data is being collected constantly, from dozens of sources, across every department.
But collecting data and using it are two very different things. Here’s why most universities get stuck:
- Data lives in silos: Student records are in one system, financial aid in another, and course performance in a third. No single place gives a complete picture, and connecting the dots requires technical knowledge most staff don’t have.
- Only one team can access it: In most institutions, querying data requires SQL or deep familiarity with internal systems. That means every question has to go through the analytics team. They become the gatekeepers by default, not by choice.
- The queue never empties: Only 32% of higher education institutions report having sufficient analytics staffing. As reporting demands grow across departments, analytics teams are often left juggling competing priorities and slow turnaround times.
- Dashboards don’t answer follow-up questions: Many universities invest in reporting dashboards that show standard metrics. But when a director needs something specific (for example: how many nursing students are at risk of losing financial aid this semester) the dashboard doesn’t help. Another ticket gets submitted.
4 Ways to Make Data-Driven Decision Making Actually Work at Your University
1. Make data accessible through plain language questions
A VP of Enrollment shouldn’t need to know SQL to find out how many students are registered for next semester. An academic advisor shouldn’t need to submit a ticket to check whether a student’s attendance has dropped in the last three weeks. In most universities, even simple questions require going through someone who speaks the technical language of the data systems. Questions like:
- How many students enrolled this week, but only those in online programs?
- What’s the retention rate for first-year students who also received financial aid?
- Can you pull the nursing program numbers, but excluding transfer students?
The obvious solution seems to be dashboards. Build a view for enrollment, one for student success, one for finance, and let people self-serve. In theory, that works. In practice, it doesn’t.
The problem with dashboards is that they answer the questions someone thought to ask when they were built. The moment a user needs something slightly different (a different time period, a different program, a different filter), they’re back to submitting a request. Dashboards are static. Real questions aren’t.
Conversational AI changes that dynamic. Staff types a question the same way they’d ask a colleague. The system interprets it, queries the data, and returns an answer in plain language.
An example of a plain language query answered by an AI assistant — a custom solution built on top of Microsoft Copilot
Follow-up questions work the same way (“show me only nursing students,” “what about last semester?”) without starting over each time.
The analytics team isn’t removed from the picture. They’re still responsible for complex modeling and strategic analysis. But routine retrieval stops landing on their desk entirely.
If you want to go further, it’s worth considering Microsoft Copilot — a conversational AI that connects to your institutional data and integrates with tools your staff already use, like Teams or Excel, without requiring any new systems or logins. If you want to understand how it works and what ROI to expect, we’ve written about that here.
2. Build a data layer that’s designed for questions, not just storage
Most university data systems were built by IT teams whose priority was security and compliance, not speed of access. The result is infrastructure that stores data well, but:
- student records live in one system,
- financial aid in another,
- course performance in a third.
Getting anything useful out requires knowing which system holds what, how the tables relate to each other, and how to get the right result.
The practical solution isn’t replacing those systems, but organizing what’s already there into something people can actually use. Think of it this way: your data is a library, but right now it has no catalog, no shelf system, and no index. The books exist, but finding the right one without help is nearly impossible.
The solution is a data layer — the organization behind the library: the catalogs, the shelves, the indexing system, the rules about what lives where. What does it do?
- Takes data from multiple source systems, copies it into one structured environment, and applies consistent definitions across the board.
- Unifies terms, so, for example, “active student” means the same thing whether you’re looking at enrollment records or attendance data.
- Ensures that financial aid status is pulled from the same source, regardless of who’s asking.
It doesn’t produce a visible output that stakeholders can see in a demo. But it’s the foundation that determines whether everything built on top of it (reports, dashboards, or anything else) actually delivers consistent, trustworthy answers.
3. Start with one use case, not a platform
If you decide to modernize data access at your institution, the temptation is to think big: one system, all departments, full integration from day one. That ambition is understandable, but it’s also one of the most common reasons these projects stall. The scope becomes overwhelming, stakeholders can’t agree on priorities, and the project either gets delayed indefinitely or launches with so many compromises that nobody ends up using it.
The universities that succeed start small. One team with a specific, painful problem. One workflow that currently takes days and shouldn’t.
In practice, this might look like starting with a single department (say, student success), where advisors spend hours each week manually pulling the same data.
Fix that one workflow first. Show the time savings. Let people see the difference between submitting a ticket on Monday and getting an answer on Wednesday, and between asking a question in a meeting and getting an answer before the next agenda item.
That visible, tangible change does something no business case document can: it creates advocates. A director who suddenly has answers to her questions in real time becomes the strongest advocate for rolling the approach out to the rest of the institution. Because they experienced it themselves.
Start small. Prove it works. Then scale.
4. If you use AI, make its reasoning visible
If you decide to implement a conversational AI tool (and the previous points explain why that’s worth considering), the biggest barrier to adoption won’t be the technology itself, but trust.
When a system returns a number, the natural first reaction from experienced staff is skepticism:
- Where did this come from?
- Is this the same definition we use in our official reports?
- Why does this look different from what I saw last month?
These are good questions that deserve good answers.
The problem is that most AI tools return results without explanation. A number appears, but the reasoning behind it is invisible. For staff who have spent years working with institutional data, a result without a source is a result they can’t trust.
This is where explainable AI makes the difference. When an AI returns an answer, it also shows which data it pulled, what logic it applied, and how it arrived at the result. In practice, this might mean displaying the underlying query, flagging when a definition differs from an official metric, or simply explaining in plain language why a number looks the way it does. That transparency turns skeptics into daily users.
Real-Life Example: How a US University Replaced a Days-Long Data Queue With a Conversational AI
Let’s look at how one US university we worked with solved the challenges we discussed earlier.
The problem
A mid-sized private university in the United States was dealing with a problem that will sound familiar. They had data like:
- student records,
- enrollment history,
- course performance,
- financial aid status.
All of it was sitting in a central data warehouse. The infrastructure was solid. The problem was that almost nobody outside the analytics team could use it.
Querying the warehouse required SQL. Most staff (advisors, directors, and business-side analysts) didn’t write SQL. So every time someone needed an answer, they went to the analytics team. The analytics team pulled the data, formatted the report, and sent it back. Days later.
For a university managing thousands of online students, that lag had real consequences. Meetings happened without current data. Decisions were made on instinct or outdated reports.
The solution
We built a conversational AI assistant that connected directly to the university’s data warehouse and let staff ask questions in plain English. No SQL. No tickets. No waiting.
The system worked on three levels:
- Data foundation: a structured model on top of the existing warehouse, with consistent definitions across all key metrics.
- Conversational interface: staff typed questions the way they’d ask a colleague. The system interpreted them, queried the data, and returned answers in plain language. Follow-ups worked the same way, without starting over.
- Transparency: every answer came with an explanation: which data was pulled, what logic was applied. For staff who knew the data’s quirks, that visibility was what made them trust it.
What changed in practice
The shift became clear quickly. One of the first power users was a data analyst on the business side who started using the tool to investigate discrepancies in daily registration reports, the kind of work that previously required back-and-forth with the analytics team over several days.
One case stood out. There was a persistent gap in registration numbers that nobody could explain. Using the AI assistant, she traced the discrepancy herself and found the answer: a specific course code was being excluded from registration counts. The whole investigation, which would have previously taken days of coordination, took minutes.
She also used the tool to:
- generate data files for academic advisors,
- identify which students needed intervention,
- cross-check figures across different university systems.
Tasks that used to take an hour were taking 8 to 15 minutes.
Results
At the time of writing:
- 87% of questions identified in the project requirements are covered by the data model,
- 42% have been confirmed as returning accurate responses,
- Early adopters are showing consistent usage, with the team expanding coverage week by week.
Better Data-Driven Decision Making in Higher Education Doesn’t Require More Data
Most universities aren’t failing at data-driven decision making because they lack data or analytical talent. They’re failing because the gap between a question and an answer is too wide for data to influence decisions in the moments that matter.
The good news is that the fix doesn’t require replacing existing systems or hiring more analysts. It requires rethinking who has access to the data that’s already there, and how quickly they can get it.
If you’re working on data access at your institution and want to talk through what this could look like in practice, we’d be happy to help. Get in touch and let’s talk about your specific business case.
