I wrote a few words on how AI forces higher education to reconsider what counts as evidence of learning. Once evidence moves from outputs to processes, universities require new infrastructures, analytics, and governance to render those processes observable and evaluable: substack.com/home/post/p-...

What does process evidence actually demonstrate, and how do you prevent process assessment from becoming a new form of performance gaming in higher education? A thread on assessment culture and redesign 🧵 (1/6)

What needs to happen in higher ed. for process-oriented learning to be more than a rhetorical commitment? Assessment redesign is one thing, but what about existing policy frameworks and behavioral and cultural change in students and staff? Thinking about this for my next article. Ideas welcome!

AI competes for the most valuable currency of our time: human attention. And while chasing engagement, it flattens the very qualities that make relationships human (patience, empathy, presence, friction). I'm wondering what our collective empathy and patience will come to look like in the future?

What counts as evidence of learning, and how is it evaluated? Since my recent thread on process-based learning, I'd like to connect with people who know more about assessment redesign and how we can prevent process assessment from becoming a new form of performance gaming

Universities are quietly accumulating a new kind of debt, but not in financial sense. Vibe coding is creating a generation of dead-end tools, data products and analyses. Here's a thread with my thoughts on tech debt, architectural literacy, and what data teams should actually do about it. 🧵

Four pieces on educational data infrastructure now up on Substack. If you work in institutional research, educational policy, or data governance in higher education (in the Netherlands or elsewhere) I'd be genuinely curious what resonates and what doesn't! substack.com/@thelearning...

I wrote a four-part series on the progression from event logs and dimensional models to process mining, causal inference, and real-time signaling in educational analytics. To summarize the argument across all four pieces: it's all about developing shared definitions. substack.com/@thelearning...

Fourth and final piece in the series is up! On what happens when AI enters the analysis layer before the infrastructure is ready, and how the higher ed. data infrastructure questions scale to institutional policy for universities. thelearningobservatory.substack.com/p/when-the-a...

The central data team that positions itself as an approval gate in a world where AI is lowering the analytical barrier is solving the wrong problem. The team that builds the semantic layer (shared definitions that make decentralized analysis consistent) is solving the right one. A thread 🧵(1/7)

As a woman operating in (mostly) male-dominated fields ranging from elite level powerlifting to data science and tech, I thought this was so cool worth a share! The AI girl gang womanifesto by Lauren Ingram open.substack.com/pub/aigirlga...

When any analist on campus can query education data with AI assistance, the quality of the underlying data model stops being a data team concern and becomes an institutional accuracy concern. Outputs that look authoritative are not, by that fact alone, reliable ones.

A phrase from a 2026 EDUCAUSE case study that keeps coming back to me: "AI does not solve governance, it exposes it." Universities deploying AI on institutional data systems are finding this out in real time. Very interesting read: shorturl.at/tHD1d

Third piece in the series is up! We've made the case for getting the data infrastructure right. We now ask the harder question: once it's in place, what can you do with it? Short answer: quite a lot, but only if you know what each analytical layer can/cannot tell you. Long answer: shorturl.at/foz6G

New article coming tomorrow! On what becomes analytically possible once the foundation is in place. Process mining, causality, near-real-time signals and where the honest limits of each layer lie. Written for people who care about the analytical questions, not just the data infrastructure.

Important distinction: finding that students who resit a first-year course have (s)lower graduation rates is a diagnostic observation. Concluding that the course is causing the problem is a causal claim. Those require different data, different analyses methods, and different epistemic care.

"AI is transforming higher education from output-oriented to process-oriented learning." It's an interesting claim that might even be right. But it deserves a bit of stress-testing because the implications for institutions are significant. Here's a thread with some my thoughts. 🧵 (1/10)

Apparently elementary school kids are saying "Everything he says is AI" when calling their peers out on BS. And far from the stereotype of young people looking for shortcuts, Gen Zers have had some of the loudest and most detailed objections to generative AI use.

A well-built event log with shared definitions tells you what paths students actually took through a programme; not what the curriculum assumed they would take. Those are often meaningfully different. The gap between them is where the interesting policy questions live.

Second piece in the series is up! A technical reference for data and institutional research people. On event logs, dimensional modeling as semantic governance, and why process mining hasn't scaled in Dutch higher education. Read the article here: thelearningobservatory.substack.com/p/what-proce...

Technical detour tomorrow! 📝 For data professionals and institutional researchers specifically. On what process-oriented analysis actually requires at the data level, why event logs are the foundation, and why a governance layer above them is a precondition for anything reliable to happen on top.

The raw material for process-oriented learning analysis already exists in most universities: every LMS click, every assessment attempt, every grade entry produces a timestamped record. The thing is that maybe only a few institutions connect and retain that data in a form that's analytically usable.

There's a precise reason why most university data systems can't quickly answer process questions about learning: the data modeling pattern was designed for output accounting and is never updated for a different question category. This is important as higher ed will likely become less output driven.

Inspiration is still lingering one month after attending the AI in Higher Education Summit hosted by @escp.bsky.social in Paris. I'm still somewhat buzzing and can't wait to tell you more abt what was discussed on how AI is fundamentally altering academic institutions. dx.doi.org/10.2139/ssrn...

Curious whether this is recognizable outside the Netherlands: Do other European university systems face the same fragmentation between standardized external reporting and internally divergent data definitions? Does it affect the external accountability process? Genuine question.

Hi! I'm Dominique, a data scientist based in the Netherlands. Interested in how data and AI can meaningfully improve education, I write about learning analytics and scaling data-informed innovations in institutional systems. I'm a mother, thriving on strong espresso and flaky croissants. Talk soon!

What happens when AI reaches the operational heart of the institution? Read my first article on thelearningobservatory.substack.com/p/before-ai-...

Most Dutch universities' data structures were built to answer one kind of question: how many students enrolled, how many credits obtained, how many degrees conferred. From a funding perspective, those were (and still are) the main questions that mattered. (1/2)

"They built a wife and called it AI" My husband tête-à-têtes with Claude about cars 'to save me from technical drivel'. Now it's explained: LLMs really just are 'Lovingly Listening Mechanisms' or 'Life Logistics Managers', aka WIVES! A very amusing read [email protected] shorturl.at/C4a6u

As a data scientist, it is my daily mission to provide nuanced perspectives on innovation. Schools that improve do so through unglamorous means: adopting coherent curricula and maintaining focus. Explaining that none of it is flashy is the core statement of most discussions I’m in. shorturl.at/5mY6T

If your institution asked: "How do students move through Bachelor programmes and which courses are being retaken more often or cause delays?", how long would that take to answer? I'm curious what solving challenges like this looks like beyond Dutch higher education.

Dutch universities report to OCW using the same templates. Yet producing those numbers takes weeks of reconciliation work at every institution. Same data, same question, but different definitions. I think this is a surface symptom at the reporting layer of institutional autonomy on national level.

Today, I asked Claude to be my critical mentor, rather than to agree with and build on what I say. I prompted it to a) check what I could improve on my idea and how, and b) whether there's anything I missed. I received the best feedback yet. I'm wondering how others are optimizing this?

Will AI push traditional universities into ‘retro’ obsolescence, OR… will people come to see academia as more reliable and more trustworthy than AI? A few scenarios for a potential return to human-led instruction if AI would lose public trust. open.substack.com/pub/aiedusim...