Increasing Course Discoverability at Entri
3 steps and a broken taxonomy between users and courses
A Funnel Untouched for a Year
The onboarding funnel had been untouched for over a year. Rather than redesigning based on intuition, I started by pulling Mixpanel data to understand where users were actually dropping off — and more importantly, why.
The first step wasn't the problem — 83.59% of users selected a goal. The collapse happened immediately after: navigating categories to reach actual courses. Only 22.24% made it. That 61-point drop between "goal selected" and "course started" was the signal. Users weren't disengaged — they were getting lost in the taxonomy between their intent and the content.
The Taxonomy Itself Was Broken
The 2-option goal screen looked clean — "Prepare for an Exam" or "Learn or Improve a Skill." But this binary broke down with real courses. As Entri's catalog grew to 600+ courses, cross-category courses became more common, not less. The goal screen wasn't just adding friction — it was actively hiding relevant content based on an arbitrary first choice.
This shifted my framing of the problem entirely. It wasn't just "how do we reduce onboarding steps" — it was how do we surface courses without forcing users into a taxonomy that doesn't reflect how they think about their goals.
Montessori Teaching
IELTS Preparation
Everything We Ruled Out First
The real tension wasn't "how many steps" — it was when the user first sees actual courses, and whether the system's structure gets in the way of finding what's relevant.
Flattened Categories with Inline Previews
Search as a Complement to the Funnel
Chat-Based Conversational Onboarding
Search-First with Browse as Default
What if users could search instantly or browse popular categories and start a course right away?
Testing the Idea Before Designing It
Instead of starting in Figma, I began by rapidly prototyping using Claude AI and quick code to simulate the search interface, filter chips, and result timing — grounded in real API response data to make it feel as close to production as possible.
This allowed fast feedback cycles with PMs, Engineers, and stakeholders before investing in high-fidelity design. Once the experience was validated, I moved to the full design phase.
Why real data mattered
Why Every Detail on the Search Screen Exists
The search-first direction was clear, but the details required a lot of deliberate tradeoffs — especially for Entri's audience of first-time smartphone users navigating in regional languages. Here's what I worked through:
The Design Decisions Behind Each Element
Search bar — always visible
Placed at top of screen with no tap required. "Start with a course, you can change later" reduces commitment anxiety — critical for first-time smartphone users where irreversible-feeling decisions cause hesitation.
Popular chips — 4-5, never more
Dynamically weighted by regional search volume. No horizontal scroll (hides options) or second-line wrap (pushes content). Chips like "Teaching" cut across the old taxonomy — surfacing both skilling and exam prep.
Course cards — content as navigation
Thumbnails on the left (not top) for vertical density — 4-5 courses visible without scrolling. Category label as secondary text teaches taxonomy passively. No price at this stage — discovery, not conversion.
Dynamic filters from results
Typing "Teaching" generates "Teaching Soft Skills" and "Teaching Exams" as chips — from search results, not pre-defined categories. Both Montessori and NET/SET appear. Users see everything, then narrow.
The Activation Gap After Discovery
Once search solved the discoverability problem, I noticed a second drop-off: users landing on course pages but not starting a lesson. The original course page buried video content below module listings — too many scrolls between "interested" and "watching."
I explored auto-play, but rejected it for reasons specific to our audience. Instead, I landed on a sticky mini player — a persistent, low-commitment invitation to start.
Why a Mini Player — Not Auto-Play?
Auto-play was considered and rejected for three reasons specific to Entri's audience:
Data costs
Many Entri users are on metered mobile data. Auto-playing video without consent would erode trust with the audience we most need to retain.
Context setting
Users need a moment to read the course title and see the content structure before committing attention. The mini player lets them orient first.
Agency over automation
The mini player says "ready when you are." Auto-play says "we've decided for you." For an audience navigating a new app, preserving control matters.
Visual hierarchy detail
What the Test Confirmed
We ran a controlled A/B test using GrowthBook over 14 days with approximately 10,000 users in each variant. The results validated the search-first approach decisively.
What the numbers tell us
The real problem was twofold — too many steps before content, and a taxonomy that didn't match how users think.
Stop making users navigate abstractions. Start showing them what's here.
What Comes Next
The search-first redesign validated the hypothesis, but there's more to explore.
Broaden measurement
Extend tests to 14- and 30-day retention, try multiple discovery variants, and monitor course purchase conversion.
Improve ranking signals
Add engagement metrics like completion rates, ratings, and recency to boost search relevance.
Seasonal awareness
Elevate time-sensitive courses around exam calendars and hiring cycles automatically.
Richer course previews
Add short descriptions, duration, and "why this course" hints to improve click confidence.