Service DesignService BlueprintingAI2025

Exploring how AI and flipped classroom models can personalize higher education.

This project explores how a personal AI tutor, combined with the flipped classroom model, could meet students' diverse learning needs. It builds on Bloom's 1984 finding that one-on-one tutoring outperforms traditional lectures for 98% of students, and asks what we'd actually design if AI loosened the capacity constraint that has kept tutoring out of reach.

Students with one-on-one tutoring outperformed
98%
of their peers in lecture-based classrooms
Bloom, 1984. The 2 Sigma Problem
Role
Service Designer, Researcher
Context
Individual project, NTNU Master's programme
Methods
Literature review, service blueprint, value proposition canvas, customer profile, technology assessment
Impact
Opens critical questions about who gets access to personalized learning, how the teacher's role evolves, and what guardrails we need before AI reshapes the system.

The spark

When one-on-one tutoring beats 98% of lectures, why are lectures still the default?

Norwegian students spend around 33 hours a week on their studies, and the lecture remains the most widespread form of teaching (Studiebarometeret, NOKUT). The research has been clear for forty years: lectures are one of the least effective ways to actually learn. Bloom's 1984 study showed that students with personal tutors outperformed 98% of their peers in lecture-based classrooms.

The constraint is not the method, it's capacity. There simply aren't enough teachers. What changes if AI can carry part of that load, not as a replacement for teachers, but as a personal companion for self-directed study?

The 2 Sigma effectThree overlapping bell curves showing student outcomes under conventional learning, mastery learning, and one-on-one tutoring. The mean of the one-on-one tutoring distribution is two standard deviations to the right of the conventional learning distribution.ConventionalLearningMasteryLearning1:1 Tutoring
Difference in learning results between a student taught one-on-one by a tutor versus a student in a normal classroom.

The problem

Four reasons the dominant teaching format struggles.

Academic depth

Surface coverage, shallow learning

Lectures cover surface but rarely go deep. Active learning and structured self-study consistently produce stronger outcomes for the same time invested.

Cognitive load

Passive listening, drifting attention

Reduced attention, dropping attendance, especially when lectures don't engage students actively. The format works against how the brain learns.

Motivation

No clear value beyond the textbook

When students can't see what a lecture gives them beyond reading alone, motivation falls. Feedback is rare; the relationship to the teacher and the material weakens.

Structure

Strategic learning, short shelf life

Combined with exam formats that reward memorization, lectures often produce strategic, short-term learning instead of lasting understanding.


The concept

Two parts of one system.

Neither AI tutoring nor the flipped classroom is innovative on its own. What's interesting is what happens when you combine them. Each one fixes the other's weakness.

Part one

Flipped classroom

Theory moves out of the lecture hall and into self-study. Class time is freed up for practice, problem-solving, and collaboration. The things students actually need a teacher and peers for.

Pitfall on its own: without good preparation materials, students arrive unprepared and frustrated.

Part two

AI tutor

A personalized companion for self-study. Adapts to each student's background, learning preferences, and pace. Gives feedback in real time and flags students who get stuck.

Pitfall on its own: too much screen time, too little human interaction.

Together, they handle each other's blind spots: the AI does the patient, individualized work that doesn't scale for a teacher; the classroom keeps the social, practical, human side of learning intact.


The service

How it actually works, end to end.

The full course flow runs across eight phases from the teacher planning the curriculum, through student onboarding (where the AI maps background, prior knowledge, and learning preferences), through AI-guided self-study with continuous check-ins, into classroom activities and final assessment. The teacher remains the pedagogical owner. The AI handles personalization at scale.

Service blueprint showing the full course flow across eight phases: curriculum planning, student onboarding, AI-guided self-study, classroom activities, and final assessment
Service blueprint — end-to-end course flow

The value

Mapping student needs to design choices.

Every part of the concept traces back to a specific student pain or opportunity. The canvas below maps it: students' jobs, pains, and gains on the right; the corresponding pain relievers and gain creators on the left.

Value proposition canvas mapping student jobs, pains, and gains to pain relievers and gain creators in the AI tutor concept
Value proposition canvas — Strategyzer template

The implications

A concept like this doesn't land in one place.

A speculative design that ignores how it ripples through the system isn't useful. Every domain below carries both an opportunity and a risk worth taking seriously.

Economic

Opportunity

Frees teacher capacity from lectures and grading toward proactive coaching.

Risk

High licensing and operating costs may exclude smaller institutions and create new inequalities.

Social

Opportunity

Democratizes access to high-quality, personalized tutoring.

Risk

If classroom attendance declines, students may graduate with strong theory but weak collaboration skills.

Environmental

Opportunity

Less commuting and less physical infrastructure for pure lectures.

Risk

AI inference is energy-intensive. Every interaction has a carbon cost.

Ethical

Opportunity

Earlier, more honest feedback for students who would otherwise fall through the cracks.

Risk

Personalization needs data. Privacy, GDPR, and AI hallucinations all become first-order concerns.

Organizational

Opportunity

Teachers shift from lecturers to facilitators and curators, a more strategic role.

Risk

Without training, time, and cultural support, the change feels like a threat instead of a tool.


Guardrails

Responsible design, not technological inevitability.

A speculative design isn't useful if it pretends nothing can go wrong. Three principles would shape this concept's implementation.

01

Human in the loop

The teacher stays accountable for learning outcomes. Dashboards expose where students get stuck and where the AI drifts. Both can intervene.

02

Guardrails on content

The AI is constrained to a curated curriculum through RAG. When asked something outside its source material, it answers "I don't know" instead of generating.

03

Accessibility by default

Universal design isn't an afterthought. Readable language, screen-reader support, low-threshold interfaces, built in from the start, per Norwegian regulations.


Reflection

What kind of education do we want?

The technology behind this concept already exists in fragments. Some students are already opting out of universities entirely and building their own AI-led curricula. The interesting question isn't whether something like this will arrive. It's what kind of institutions, regulations, and culture we want to put around it before it does.

What talent, infrastructure, laws, and norms must we put in place today if we want to acheive this the future (or avoid it)?
← Back to portfolio

Other work

Continue exploring

Turning Nav's struggling budget tool into “a dream tool” for 300 users
A 20% rise in user satisfaction at the Norwegian Labour and Welfare Administration.
Using service design to improve mental health services through the eyes of healthcare specialists
An advisory engagement with Helse Sør-Øst: from clinician insight to a co-created improvement roadmap.