Back to Coschool Hub
Guided Practice Revamp Hero
Coschool · Guided Practice

Guided Practice Revamp

Giving students control over how they learn with AI

UX Strategy AI Learning Gamification EdTech

My responsibilities: UX Strategy · Product Direction · Information Architecture · Interaction Design · Gamification · Design System · Team Delegation

Research & validation: Student behaviour data · Heatmaps · Oakridge student workshop · ~12,000 Guided Practice interactions

Role UX Design Manager
Timeline 2 Months
Team Manager, Sr Researcher, Jr Designer
Impact 65% gibberish drop
12K Interactions Analysed
7K Gibberish Messages
↓ 65% Gibberish Reduction
01

The Old Experience

The old experience

Guided Practice was designed to help students learn through practice with Vin, Coschool's AI Tutor. The original idea was well-intentioned: don't just tell students whether an answer is right or wrong, help them understand the concept.

But the way this was implemented created an unexpected conflict. Vin controlled almost every part of the experience. The student answered a question. Vin responded. The student had to continue the conversation. Vin decided when enough learning had happened. And Vin controlled when the student could move to the next question.

The Original Flow

Teacher assigns GP → Student starts homework → Vin presents a question → Student answers → Vin explains / asks follow-up questions → Student responds again → Vin decides whether the student can move forward.

02

The Unexpected Behaviour

The problem wasn't that students didn't understand. A teacher might assign a 10-question Guided Practice expected to take 15–20 minutes. But if a student got several questions wrong, Vin could introduce two or three additional similar questions for each mistake. What looked like a 10-question assignment could feel like a 30-question assignment.

7,000 Gibberish Messages

We analysed ~12,000 Guided Practice interactions and found around 7,000 gibberish messages. Even focused students sometimes typed random text.

A Signal of Lost Agency

Students were trying to regain agency. They had discovered Vin was an obstacle. Typing gibberish became a way of trying to escape the interaction.

03

The Core Insight

To understand whether this was a genuine experience problem, we conducted a workshop with students from Oakridge School. The workshop reinforced what we were seeing: Students didn't necessarily dislike getting help. They disliked being forced into help when they didn't want it.

The Design Principle

The student leads. The system supports. Vin helps when needed.

04

The New Experience

Attempt

Starting screen Answering MQ1

Ask Vin

Clarifying question loop

Understand

Concept flagged Incorrectly answered hint

Similar question

Asking PQ choice PQ of MQ1

Continue Flow

After video Later Incorrectly answered Answering

We moved away from an AI-led conversation and introduced a more structured model. Instead of entering a conversation with Vin, the student sees a focused question and answer options. The primary task is clear: attempt the question. Vin is available, but no longer dominates the screen or controls progression.

System Controls Structure

Questions, attempts, progression, concept tracking, and learning recommendations.

Student Controls Agency

Attempt independently, ask Vin before answering, ask for clarification, discuss mistakes, and choose how to use support.

05

The Learning Loop & Gamification

We introduced a new path for students who want to learn first. Instead of "Don't know → Attempt → Get stuck → Forced learning", students could choose "Don't know → Learn first → Attempt with more confidence". If they answer incorrectly, they can discuss the mistake with Vin and reinforce it with exactly one similar question.

We also introduced an economy around student choices using coins earned through performance. Coins introduced an economy around student choices. Performance created rewards. Rewards created options. And options created more control.

Solution Checking

Get the Answer

Students can spend coins to get the answer with a clear explanation—useful when short on time or facing deadlines.

06

Before vs After

Before

  • AI controls progression
  • Unlimited / 2–3 similar questions per mistake
  • Forced AI guidance

After

  • Student controls progression
  • Maximum 1 similar question
  • Contextual AI guidance
07

Leadership & Results

Product leadership was largely focused on new initiatives. Rather than waiting for a formal mandate, our team (Manager, 1 Sr Researcher, 1 Jr Designer) continued developing and demonstrating the redesigned direction in weekly organisation-wide meetings within a 2-month timeline.

When you don't have the authority to change a decision, evidence and a better experience can still create influence. The redesigned experience eventually received recognition from the founder and delivered significant behavioral impact.

65% Drop in Gibberish

Students were no longer trying to fight the AI to move forward. The experience gave them clearer choices and more control.

More Predictable Learning

Assignments no longer expanded uncontrollably. Struggling students were directed toward learning concepts instead of endless questioning.

← Previous Case Study Coschool EDGE Next Case Study → Admin Portal Revamp