Home / Industries / Education
Industry — Education

AI for Education

Teachers are stretched and admin is endless. We build AI that takes routine work off educators, gives learners personalised support, and respects student privacy — so people can focus on teaching and learning.

Last updated: 11 June 2026

Why AI here

Support teaching and learning — protect students

Schools, universities and edtech all face the same squeeze: too much admin and a wish to support every learner individually. AI helps on both sides — automating the routine and offering students grounded, on-topic help — without replacing the educator's role.

Student data is sensitive, often involving minors, so we build privacy-first: GDPR-compliant, with private models and safeguards where the context demands it.

Where AI helps

Use cases across education

01

AI tutoring assistant for students

Grounded assistants that help learners with explanations and practice, on your curriculum.

02

Admin automation

Automate scheduling, reporting and routine correspondence to free up teacher time.

03

Content & assessment

Draft lesson materials, exercises and quizzes from your own content, for teachers to refine.

04

Student support chatbot

Answer common student and parent questions from official information, around the clock.

05

Learning analytics

Spot students at risk of falling behind so support reaches them earlier.

06

Accessibility & translation

Make materials accessible and multilingual so every learner can keep up.

How we work

From use case to the classroom

Step 1

Audit

We map the biggest time-sinks and your privacy constraints.

Step 2

Build an MVP

By the second call you get a working prototype on your use case — not a spec.

Step 3

Deploy privately

We integrate with your LMS and deploy with privacy and oversight built in.

Step 4

Monitor

We track quality so the tool keeps genuinely helping teachers and learners.

What you gain

Outcomes for educators and learners

What admin actually costs a Dutch school — and where AI claws it back

Walk into any vmbo, hbo or mbo back office in the Netherlands and the bottleneck is rarely the teaching. It is the paperwork around it: attendance reconciliation, ouderavond scheduling, leerlingvolgsysteem updates, exam logistics, subsidy and inspection reporting, and the steady drip of parent and student emails that all want answering today. Dutch teachers consistently report some of the highest administrative loads in Europe, and the workload (werkdruk) debate is now a permanent fixture of every CAO negotiation. When a docent spends an evening copying grades between a spreadsheet and Magister instead of preparing a lesson, that is not a software gap — it is senior, expensive, hard-to-replace human time leaking out of the system. That gap is the case for AI for education Netherlands institutions can defend to a board: not novelty, but hours handed back to the people who teach.

This is where AI in het onderwijs earns its keep, and it is the lens we apply to every onderwijs AI build: not "what can the model do" but "which recurring task is eating teacher hours, and what does the week look like once it stops." A well-scoped administrative automation layer does not need to be glamorous. It needs to read the messy inputs schools actually have — PDFs from the municipality, exports from the SIS, free-text emails — and turn them into the structured action a human would otherwise do by hand.

The outcomes that move the needle

  • Admin hours saved. Routine correspondence triage, schedule conflict resolution, and report drafting are the highest-volume, lowest-judgement tasks in a school. Automating the first draft and the routing — with a teacher approving, not authoring — is where the hours come back.
  • Early-risk detection. Student analytics that flag a learner slipping before the report card, not after, change the conversation from damage control to support. The value is the lead time, not the dashboard.
  • Teacher time back, returned to teaching. Every hour reclaimed from admin is an hour that can go to feedback, differentiation, or simply a less exhausted teacher. That is the whole point — the human stays at the centre.

Grading support that keeps the teacher holding the pen

Grading is the most emotionally and legally loaded place AI touches education, so it is also the place where careless tooling does the most harm. We do not build systems that hand out final marks. We build grading support: tools that take the mechanical weight off the docent while the judgement stays human.

In practice that means a model that reads a batch of open-answer responses or essays and produces a structured first pass — surfacing where each answer hits or misses the rubric your department already uses, flagging the borderline cases that genuinely need a human eye, and drafting comment text the teacher edits rather than writes from scratch. For short-answer and language work, this can turn an evening of grading into a focused review session where the teacher spends their attention on the 20% of scripts that are actually difficult, not the 80% that are routine. Marking benchmarks across the sector suggest the bulk of a grader's time goes to the easy and middle band; compressing that is where the realistic time saving lives.

Two guardrails are non-negotiable and we wire them in from day one. First, the teacher always sees and can override every suggestion — the tool proposes, the educator decides. Second, the model is grounded in your rubric and marking scheme, not a generic notion of "good," which is what keeps grading consistent across classes and defensible if a grade is ever challenged. For institutions building this into their own platform, our LLM optimisation work is about exactly this: making the model's output reliable, on-rubric and explainable rather than impressively fluent and quietly wrong.

Student analytics that flag risk early — without surveillance creep

Most schools already sit on the data that predicts who is about to struggle: attendance trends, submission patterns, grade trajectories, login activity in the LMS. The problem is that this data lives in four systems and nobody has time to read it as a whole until the damage shows up on a rapport. Student analytics done well joins those signals and answers one question early enough to act on it: which learners are drifting, and why.

The operational change is concrete. Instead of a mentor discovering at the ten-week mark that a student has quietly stopped handing in work, a weekly risk view surfaces the pattern in week three — while a phone call home and a small intervention still make a difference. The model does not decide anything about the student; it points the mentor at the right conversation sooner. Education-sector studies on early-warning systems consistently find that the value is measured in weeks of lead time, because earlier contact is dramatically more likely to keep a learner on track than a late one.

Because this touches minors and sensitive personal data, the design constraints are heavy and we treat them as a feature, not a footnote. Risk scoring is a decision-support signal for a human mentor, never an automated judgement about a person — which is also exactly the line the EU AI Act draws around AI used in education and the assessment of students. We build to keep these systems on the right side of that line: explainable signals, human-in-the-loop by design, data minimisation, and clear records of why a flag was raised. Getting the plumbing right — clean, consented, well-governed data flowing out of the SIS and LMS — is its own discipline, and it is where our data engineering practice does the unglamorous work that makes the analytics trustworthy in the first place.

EdTech teams: ship the AI feature, not the AI risk

A large share of the people reading this are not running a school — they are building the software schools and universities buy. The Dutch and wider European EdTech market is crowded, and "we added AI" is no longer a differentiator; a tutoring assistant that hallucinates curriculum or an essay tool that helps students cheat is now a liability your buyers' procurement teams will catch. For EdTech founders, the senior-led model is the point: you get an AI engineering partner who has shipped production systems, not a feature bolted on by a generalist.

Where we plug into an EdTech roadmap

  • Curriculum-grounded tutoring and content. A generative AI layer that answers from your course material and stays on-topic, so the assistant is genuinely useful and defensibly safe — not a thin wrapper around a public model that will confidently invent a fact in front of a fifteen-year-old.
  • Adaptive learning and recommendation. Machine learning models that sequence practice to each learner's actual gaps, the kind of personalisation that justifies a premium tier.
  • Automated assessment inside your product. The grading-support pattern above, delivered as a feature your teacher-users trust because the human stays in control.
  • The product around the model. Most AI features fail on the engineering, not the model — auth, multi-tenancy, LMS integrations, latency, cost control. Our application development team builds the production system, so you ship a feature, not a demo.

The commercial logic for an EdTech company is straightforward: an AI capability that is accurate, compliant and genuinely time-saving for teachers becomes a reason to choose your platform over a competitor's — and a reason your existing customers do not churn when the next better-marketed tool appears.

Compliance is the product in Dutch education

Nowhere is "compliance-first" less optional than in education. You are handling the personal data of children, and from 2025 onward AI systems used to assess students or determine access to education sit in the higher-risk tiers of the EU AI Act. Add the AVG/GDPR, a privacy-conscious Dutch sector, and parents who — rightly — ask hard questions, and the message is clear: a clever model that cannot survive a DPIA or an inspection is worthless.

We design for that reality from the first conversation rather than retrofitting it before launch. Concretely, that looks like data minimisation (the model sees only what it needs), private or self-hosted models where student data must not leave your control, clear human oversight on anything that affects a learner, audit trails that explain why the system did what it did, and documentation built to satisfy both an AVG assessment and the record-keeping the AI Act now expects. None of this slows a project down when it is the starting assumption — it only hurts when it is an afterthought. This is the same compliance backbone running through all of our AI consulting in the Netherlands, refined across thirteen delivered case studies in regulated and sensitive fields including healthcare and clinical NLP, where "the model is impressive but we cannot deploy it legally" is the most expensive failure of all.

Why a boutique partner fits Dutch education better

Schools, universities and EdTech SMEs do not need a fifty-person delivery org or a generic web shop that rebranded as an "AI agency" last quarter. They need senior people who understand both the model and the messy operational reality of a Dutch onderwijsinstelling — and who stay on the project rather than handing it to a junior after the kickoff. As a boutique, senior-led consultancy, that is structurally what Crux Digits is: the people who scope your build are the people who deliver it, and the goal is that your institution owns the result, not that you rent it from us indefinitely.

A transparent, low-risk way in

You do not have to commit to a big platform to find out whether AI helps. The path is deliberately stepped so you prove value before you spend at scale, with fixed pricing (excl. VAT) and no open-ended day-rate surprises:

  • AI Audit & Strategy — EUR 2,500. We map where teacher hours actually go, your privacy constraints, and the two or three use cases worth building. You leave with a plan, whether or not you build it with us. See AI Audit & Strategy.
  • Proof of Concept — EUR 20,000. A working tool on your real use case and your real data — a grading-support pass, an early-risk view, an admin-automation flow — so the decision to scale is based on something you have used, not a slide.
  • Production launch — from EUR 50,000. Full AI implementation: integrated with your LMS and student systems, deployed with privacy and oversight built in, monitored so it keeps helping.

If you are weighing whether AI belongs in your classroom, your faculty or your product, the honest first step is a conversation about where the time and the risk really sit. You can see how the pricing works, look through our case studies, or read more about the team — and then book a free consultation. We will tell you plainly where AI gives teachers their time back, where it flags risk early enough to matter, and where the honest answer is not yet. The teacher stays in control of teaching; we just take the routine work off their plate.

FAQ

Questions, answered

How is student data protected?

Privacy is the starting point — GDPR-compliant handling, data minimisation and, where minors are involved, extra safeguards and private models so student data stays protected.

Does AI replace teachers?

No. We build tools that support teachers and students — taking admin off teachers' plates and giving learners help — while teaching and assessment stay with educators.

Will it integrate with our LMS?

Yes — we connect to your learning management and student systems so AI works inside your existing tools.

What about academic integrity?

We design tools to support learning, not shortcut it — grounded in your material, with guardrails and a focus on understanding.

Who builds AI for schools and edtech companies in the Netherlands?

We do. Crux Digits is a boutique applied-AI firm in the Utrecht region, working with schools, universities and edtech companies across the Netherlands and the EU, remotely or on-site. We build tutoring assistants, admin automation and learning tools, with senior engineers doing the work in English or Dutch, privacy-first from day one.

What does an education AI project cost, and how do we get started?

First a short paid discovery scopes your use case, budget and privacy constraints; then we deliver at a fixed price and you own the code and IP. A proof-of-concept starts at a few thousand euros, a production build usually from around 20k. Because education data often involves minors, we build to the AVG (GDPR) and design under the EU AI Act.

Drowning in admin, or want to support learners better?

Tell us where the time goes — we'll map a safe, privacy-first path to value in a free consultation.

Book a free consultation →