ROLE
TIMELINE
8 months
TEAM
1 PM, 3 engineers, 1 content lead
SCOPE
Figma, Voiceflow…
00 · ABOUT BRIDGE FOR BILLIONS
Bridge for Billions helps early-stage entrepreneurs, many from underserved communities, turn an idea into a business. Its programmes run with partner organisations and pair each founder with a volunteer mentor.
01 · THE PROBLEM
Before Parrobot, half the entrepreneurs dropped out. Most had little digital or educational experience, the people the programme was made for.
ACTIVE PARTICIPANTS BY WEEK
WHY THEY LEFT
01
The platform was the first barrier.
Logging in, menus and forms came before any learning.
02
One level for everyone.
Lessons assumed business vocabulary many had never heard.
03
Alone between sessions.
Stuck on a Sunday, with nobody to ask until next week.
02 · RESEARCH
We talked to past participants, half of them dropouts, surveyed many more and watched how people actually used the platform.
18 in-depth interviews,
9 with dropouts
64 survey responses
from past participants
40 platform session
recordings reviewed
FROM INSIGHT TO PRINCIPLE
“I spent the first week just trying to log in.”
DROPPED OUT · WEEK 3
Talk, don’t navigate.
No menus to learn. A conversation is the one interface everyone already uses.
“The lessons used words I had to look up every time.”
DROPPED OUT · WEEK 6
Meet them at their level.
Same goals for everyone, different ways in. Nobody left behind, nobody held back.
“When I got stuck on a Sunday, there was no one to ask.”
FINISHED THE PROGRAMME
Never leave them alone.
Help at any hour, and a human when the AI isn’t enough.
03 · WHO WE DESIGNED FOR
Research showed three very different starting points. Instead of one course for everyone, each one became a level: Parrobot asks three quick questions and adapts every lesson from there.
Cooks for her neighbourhood. Uses WhatsApp, has never used a laptop.
Runs a repair shop. Online, but never planned prices or finances.
Has a project and a basic plan. Wants to grow and find funding.
04 · CONVERSATION DESIGN
Before any UI, I mapped how Parrobot should talk: how it finds each person’s level, how it teaches, and when it steps aside.
05 · DESIGNING THE INTERFACE
The old platform had six ways to ask for help: a help drawer, tips, examples, deep dives, hints inside each field and the community. People didn’t know which one to open, so they opened none. Parrobot replaced them all with one conversation, always in the same corner.
THE REDESIGN · BEFORE → AFTER
With help in one place, the rest of the platform followed: one home, one question at a time, and visuals instead of tables.
06 · TESTING
10 past participants per round, from all three levels, on their own phones. What failed in round 1 shaped round 2.
07 · DESIGNING FOR TRUST
An AI guide for first-time founders can’t afford to sound sure when it isn’t. I designed three rules for the moments where Parrobot could fail people.
08 · Impact
0%
of entrepreneurs finished, up from 52%.
Parrobot worked alongside mentors and live sessions. This is the result of the whole programme, not the AI alone.
Parrobot
Participants after the programme
Sources: platform analytics · post-programme survey
09 · What I learned
Design the conversation before the screens.
The flows and scripts did more for usability than any visual decision.
An AI needs limits as much as answers.
Deciding what Parrobot shouldn’t say was as important as what it should.
Support is part of the product.
The AI worked because there was a human behind it.
What I’d do next
Voice notes, for people more comfortable speaking than typing.
It was the most requested feature after launch.