Go back

EdAbroad: designing AI recommendations students can trust

A mobile app that matches Indian creative students to US programmes and explains every match, built against the one thing the research kept returning to: these students had been misled before.

31.1%sign-up on the fake-door test, 14 of 45 visitors, against a 25% target
30 of 35usability tasks completed across 5 sessions
Role
Product strategy, UX research, product design
Team
Solo
Timeline
8 months
Context
UX master's thesis, Thomas Jefferson University
EdAbroad cover: the app logo and its tagline, unlock your creative future

EdAbroad is a mobile app that matches Indian creative students to US programmes and explains every match. It was my UX master’s thesis at Thomas Jefferson University. Before designing a single screen, I ran three pretotype tests to check that students wanted it, would trust it and would finish onboarding. In usability testing, students completed 30 of 35 tasks, and both failures changed where features sit.

In short

Problem
Indian students in design, art and architecture rely on consultants paid by commission. 11 of the 12 students I interviewed had been steered toward programmes they didn’t want.
Opportunity
Students didn’t lack information. They lacked advice they could check. An app that shows its reasoning could earn the trust consultants had lost.
What I did
Researched the problem, validated it with three pretotype tests, and designed a mobile app for programme matching, alumni mentorship and application tracking.
Outcome
A 31.1% sign-up rate on a fake-door test (target 25%). All three students in a simulated-AI test trusted the explained matches more than their consultant’s advice. 30 of 35 usability tasks completed.
Proudest decision
Making the AI explain every recommendation, because the problem students described was trust, not accuracy.

01 / Context and constraints

Advice that bends toward whoever is paying.

Consultancies dominate study-abroad advice in India. They earn commissions, so their advice bends toward what pays them rather than what fits the student. Creative students get the worst of it: consultants know the engineering pipeline and push students into it.

  • Solo and unfunded. No engineers, no live AI model, no university data partnerships. Every “AI” behaviour had to be simulated to test it.
  • Trust starts at zero. These students had already paid someone who misled them. A new app asking for their grades and goals looks like another consultant.
  • High stakes, long cycle. Applications run for months and cost real money. A wrong recommendation can mean a lost year or a visa rejection.

02 / What research told me

Twelve students, one pattern.

I interviewed 12 Indian students from different academic backgrounds.

11 / 12students were steered toward programmes they had no interest in.

11/12

Got guidance unrelated to what they wanted to study

“They were trying to hook me up to technical courses… which I had zero interest pursuing.”

09/12

Experienced stress, regret or mental-health strain

“They demotivated me… I had to raise my voice for them to apply.”

07/12

Lost money to a consultancy

“I ended up paying them ₹7000 and I got my certificates back.”
Venn diagram of three overlapping problems: broken guidance systems, the neglect of creative students’ needs, and no tools for independent planning. Where all three meet, creative students face confusion and misdirection.
Three problems overlap on one group: creative students planning a move abroad alone.
Students didn’t lack information. They lacked advice they could check.

That changed the brief from “recommend programmes” to “recommend programmes in a way a burned student will believe”.

how might we

help Indian students misguided by traditional consultancies make better decisions about studying in the U.S.?

Three pretotype tests, before any screens

Fake front door test7 days

31.1%signed up, 14 of 45 visitors

A one-page site described EdAbroad as if it existed. 45 unique visitors, 14 sign-ups, against my 25% success threshold.

Mechanical Turk test30 days

3 / 3trusted the explained matches

I hand-picked programmes for 3 students and presented them as AI output, each with a “why this is a good match for you” note. Two had already shortlisted some of my picks. After 30 days, all three said they trusted these matches more than their consultant’s advice.

Pinocchio testform vs. app

1 of 4finished the long form

Participants abandoned a long Google Form partway through, calling it “just another boring form”. Shown the same questions one per screen, they said they would finish it.

Test 2 made explanation a core feature. Test 3 killed the long onboarding form.

03 / Key design decisions

Four decisions that shaped the app.

01

Start with creative students going to the US, and nothing else

Context
Every student group has the consultancy problem, but creative students had it worst.
Options
  • A — all students, all countries
  • B — all students, US only
  • C — creative students, US only
Choice
C.
Why
A narrow group with acute pain gives the product immediate relevance and keeps matching tractable for one person to design and test.
Trade-off
A smaller launch market. The quiz and the match model treat fields and countries as data, so expanding to computer science, medicine, the UK or Canada adds content, not a redesign.
02

Explain every match instead of just ranking

Context
Students distrusted advice they couldn’t check. A black-box ranking would feel like one more consultant.
Options
  • A — a ranked list
  • B — a match percentage
  • C — a percentage plus plain-language reasons tied to the student’s own answers
Choice
C. Partner universities are labelled, and partnership never changes a match score.
Why
Every pick in the Mechanical Turk test came with a reason, and all three students trusted those picks more than their consultant’s. Reasons also let a student disagree: “the fit is right but the cost isn’t” is a useful signal.
Trade-off
Taller cards, so fewer fit on screen. A precise percentage can also suggest false certainty, which is why the reasons sit directly under the score.
03

A quiz instead of a profile form

Context
Matching needs a lot of data: interests, goals, academic background, and IELTS and GRE scores.
Options
  • A — one long profile form
  • B — a transcript and CV upload
  • C — a short quiz, one question per screen, with details added later
Choice
C.
Why
The form test showed students abandon long forms. A quiz shows progress, feels like it is working for the student, and gets to a first set of matches fast.
Trade-off
Early matches use less data, so they’re rougher. The app asks for test scores and budget when a student shortlists a programme, so matches sharpen as the student commits.
04

Keep the core free, and charge for people, not matches

Context
The product needed a business model that didn’t recreate the consultancy problem.
Options
  • A — a subscription for everything
  • B — commissions from universities
  • C — free matching and tracking, with a one-time upgrade for mentorship and application support
Choice
C. A ₹15,000 one-time upgrade adds alumni mentorship and full application support. Other revenue comes from per-application service fees, white-label licensing and clearly labelled university partnerships.
Why
Charging for matches, or taking commissions on them, would bring back the incentive the product exists to remove.
Trade-off
If mentorship is paid, EdAbroad can start to look like a consultancy. Where the paywall sits is the first thing I’d test.

04 / Testing and what changed

Both failures were about placement.

I ran 5 moderated sessions of 7 tasks each, averaging 40 minutes. Students completed 30 of 35 tasks (85.7%).

User testing summary: 5 participants, 40 minutes on average, an 85.7% success rate, and four findings including users missing the shortlisting feature and struggling to find mentorship.
Five sessions, seven tasks each.
Finding 1

5 of 5 users missed the AI programme matching feature.

Finding 2

4 of 5 users didn’t notice mentorship unless guided, because it was hidden in the side menu.

What changed

What failedWhyWhat I changed
AI programme matchingThe entry point was a home-screen banner everyone scrolled pastAdded a shortlisting prompt to the Explore screen, where students already look for programmes
MentorshipIt was hidden in the side menuMoved it onto the programme page: “Talk to an alum from this programme”
OrientationUsers landed on Home with no sense of what the app could doA four-step walkthrough after onboarding, repeatable from the menu
Task analysis, before and after: the programme shortlisting banner on the home screen was skipped by every user, so a shortlisting prompt was added to the Explore screen.
Matching: the banner everyone scrolled past, and where the prompt went instead.
Task analysis, before and after: mentorship was hidden in the side menu, so a prompt promoting mentor support was introduced.
Mentorship: out of the side menu, onto the programme page.
Task analysis, before and after: users landed on the home screen with no orientation, so a four-step walkthrough was added after onboarding.
Orientation: a four-step walkthrough after onboarding.

I know walkthroughs are often skipped. I chose one because both failures were about features people didn’t know existed, and a walkthrough was the cheapest fix to test. Next, I’d test it against prompts shown at the moment of need.

05 / Outcomes

A validated concept, not a shipped product.

EdAbroad is a validated concept, not a shipped product. It proved:

31.1%of visitors signed up for a product that didn’t exist yet
3 / 3test students trusted explained AI matches over human consultants
30 / 35usability tasks completed, both failures traced to placement, not concept
8 mofrom first interview to tested prototype, solo

What the app does

01

Smart shortlisting

Matches come from the quiz: preferred programmes, goals, academic background and test scores, each with a transparent score and the reasons behind it.

02

Mentor matching

Students are paired with alumni who studied the same field, which answers the isolation and the generic advice at the same time.

03

Application support and tracker

Personalised task breakdowns and reminders, so deadlines stop being the thing that costs a student a year.

Three phone screens of the application process: the application fee and document checklist, reviewing and confirming details, and an application status tracker.
The tracker organises, reminds and links out: US programmes apply through their own portals.

If it launched, I’d measure time from sign-up to a saved shortlist, the share of students who act on a match, and deadline misses among tracked applications.

06 / What I would do differently

What I would do differently.

Test with more students, and with consultants

Five sessions found the big problems, but they can’t tell me which programme data students trust most.

Be clear about what the tracker does

US graduate programmes use their own portals, so the tracker organises, reminds and links out to apply. I’d say that in onboarding, so nobody expects to apply inside the app.

Prototype explanations with real programme data,

to see whether the reasons hold up when they’re generated rather than hand-written.

Use match bands instead of exact percentages

Strong, Good and Reach would be more honest about how precise a match can be.