Case study · Innovation · AI for Development

TARA — closing the migrant-student transition gap

An AI-powered, WhatsApp-native chatbot supporting Indonesian student migrants in the UK. Geneva Challenge 2025 Global Finalist.

Partner

Imperial College London × UCL × University of Hamburg

Role

Co-lead · Cost model + multi-level intervention architecture

Year

2025

01

< US$0.50

cost per interaction

02

~£5

cost per student reached

03

US$96k

national investment protected / scholar-year

04

90.9%

WhatsApp penetration leveraged

Context

A ~US$962M investment at risk every year

Indonesia invests heavily in sending its brightest students abroad — ~US$962M annually flows to UK universities through scholarships like LPDP, on the explicit expectation that scholars return and contribute to national development.

But a preventable transition gap — language, NHS, housing, visas, food, wellbeing — quietly erodes that investment. Students disengage, drop out or stay disconnected, producing 'brain waste' instead of returnable human capital.

The challenge

A multi-level problem with no single owner

The gap sits across three failing layers at once: institutional (universities don't know who needs what), community (scholar associations are stretched thin), and individual (students don't know what they don't know until it's too late).

Any solution had to be culturally attuned, bilingual, 24/7, low-friction, and cheap enough to scale to thousands of students without grant-funded gatekeeping.

The approach

WhatsApp-native, bilingual, multi-level

We chose to meet students where they already are. 90.9% of Indonesians use WhatsApp daily. No app installs, no onboarding tax, no learning curve.

Phase 1

Pilot — 2–3 UK campuses

Co-creation with PPI UK, university international offices, LPDP

Step-by-step guidance across academics, NHS, housing, visas, food and wellbeing — in Bahasa Indonesia and English.

Phase 2

UK-wide rollout

Layered partnerships with host universities

Verified-info loop with continuous learning from every interaction; ethics & data-protection officers embedded.

Phase 3

Replication toolkit

Other nationalities · other host countries · labour migrants

TARA's architecture is transferable. The same multi-level model maps to Indonesian workers abroad and to migrants of other origins.

The result

Decision-grade economics

Cost per interaction under US$0.50 vs. US$15–25 for equivalent human support. ~£5 per student reached protects ~US$96k of national investment per scholar-year.

  • Selected as a Geneva Challenge 2025 Global Finalist (top from 100+ international teams).
  • Aligned with SDGs 3 · 4 · 10 · 17 (health & wellbeing, quality education, reduced inequalities, partnerships).
  • Built a 5-person team across Imperial, UCL and Hamburg with a co-creation governance stack.
“If you spend US$962M sending people somewhere, the rational thing to do is spend the next US$5 helping them land. TARA is that landing.”

— Geneva Challenge 2025 — submission rationale

Methodology

How the work held up.

  • Demographic + economic data: HESA, ONS, DataReportal, LPDP open reports.
  • Cost modelling against published human-support unit economics and AI inference benchmarks.
  • Stakeholder co-creation across students, scholar associations, universities, AI developers and government bodies.
TARA strategy deck (PDF)
Next case