Diary study with 18 students
Study sessions were short, interrupted and mostly on shared phones. Anything requiring more than one decision to start was skipped.
02Edtech · Learning platform· 2022 — 2023
A learning home that tells students what to do next
Turning a dense diagnostic engine into one clear next action per student, per day.
- Role
- Product Designer — learning home, practice loop, design system
- Team
- 3 designers, 2 PMs, data science partners
- Platforms
- Android app, responsive web
- Timeline
- 2022 — 2023
Outcomes
What changed
+27%
session completion
1 action
per visit, by default
9 modules
unified in one system
Context
The situation
Embibe's diagnostic engine knows a great deal about each student: mastery per concept, attempt behaviour, predicted score movement. All of it was on screen at once.
The result was a dashboard that looked powerful and produced nothing. Students opened the app, scrolled, and left without starting a session.
The problem
Students landed on a wall of scores, topics and recommendations. Everything was visible, so nothing was prioritised — and practice sessions stalled after the first attempt.
Research
What I learned first
Score anxiety
Percentile framing made weak students avoid the exact topics they needed. Mastery framing kept them in the loop longer.
Drop-off after first attempt
Practice ended when feedback felt like judgement rather than direction — there was no obvious next thing to do after a bad score.
Process
How it came together
Ranked what a student needs per visit
Ran a card sort with the PM and data team to force a single primary action per session; everything else became progressively disclosed.
Designed the focus card
One recommended session above the fold with the reason it was chosen ('your weakest concept in Kinematics'), so the recommendation felt accountable.
Reframed scores as mastery
Replaced percentile-first reporting with concept mastery bars and a delta since last attempt, so progress was visible even when scores were low.
Unified nine modules into one system
Built shared components for cards, chapters, attempt states and empty states so nine feature teams stopped inventing their own patterns.
Key decisions
Calls worth defending
One action by default, more on demand
The full topic map still exists — it just sits behind an explicit 'choose my own' route rather than competing with the recommendation.
Every recommendation states its reason
Explaining the why turned an opaque algorithm into advice students were willing to follow.
Revision nudges tied to decay, not streaks
Nudges targeted concepts likely to be forgotten instead of rewarding daily opens, which kept the loop useful rather than compulsive.
Reflection
What I’d carry forward
Data-rich products fail on prioritisation, not visualisation.
A recommendation without a reason is treated as noise.
Design system work was the fastest way to make nine teams ship consistently.
- Information architecture
- Data-heavy UI
- Personalisation
- Design ops
Keep reading
