Afriex

6,000.00 USD

243.00 NGN

0.00 GHS

USD

-₦282

+₦7,000

+$500

Afriex Inc.

Product Design

Driving transaction volume by increasing clarity

Impact

Increased send success by +250%. Design framework I shipped is still used today and is scaled for crypto.

Overview

Afriex enables immigrants to send money to Africa in seconds without any markup. While Afriex was in their seed round at YC, I joined as their first designer, leading the iOS/Android experiences. Customer support proposed an interactive walkthrough as a quick fix for customers having issues with sending. I rallied the team to put engineering time into improving the Send UI instead of adding more pop-ups between users and their money.

RESPONSIBILITIES

Design

Prototyping

Timeline

2021-2022

Team

Design (1)

Engineering (2)

Back

Problem | 01

Send experience wasn’t intuitive: +80% of support tickets were about "how to send money"

Send is the core screen customers land on when they open Afriex and every transfer starts there. That makes it the screen that drives volume and the company’s goal was $25M+ by the end of 2021. Despite Afriex highly competitive exchange rate, most customers never return after their first app install because they simply weren't able to send.

Indeed gives 500M+ job seekers free access to 30M+ jobs from thousands of company websites and job boards, helping companies of all sizes hire smart. But with one-click applications, employers (EMPs) are flooded with candidates, and when they can’t find strong hires quickly, they’re more likely to churn.

Problem | 02

As Afriex grew its user base, errors grew 2x faster: card failures rose 16.7% week over week

Current error communication is destructive and lack context. From April to June 2021, failed card adds grew 16.7% a week, more than twice as fast as successful ones (7.7%). Each week users dropped off for these core reasons: (1) add cards - failed (2) verify bank - failed . Add cards and verify bank are also a part of 'Send.'

Job seekers use AI too. Indeed's AI tells them why they fit a job and rewrites their resume in that job post's language. So when an EMP's AI summarizes each applicant, it's summarizing resumes that all echo the same post and every summary sounds alike. The words match but we don't know who has actually done the work. That's when employers stop trusting the summary and go back to the resume itself for more truth.

GOALS

Help users send money faster with an intuitive Send

Send is the rocket screen that people land on and we had limited weekly sprints effort. Changing it too much would also make our regular users relearn and retest our system. I prioritized on simplifying and increasing usability for the send interface because of its high impact area. I scoped the work around one question: what is the smallest, most impactful change that gets a new user to their first send?

Two directions I scoped

Out of scope

Redesign the whole Send flow

Fix the Send screen, card payments and bank payments in one project.

Why not now

Too many sprints, and we couldn't tell which change moved the numbers.

Chosen

Simplify the main Send screen

Test with customers, then ship with the PM, support agents and engineers.

Why we chose it

Small enough for our sprints, focused enough to measure.

Failed card and bank cases will be improved through lightweight design system

What does design success look like?

Send is the first screen every user lands on, which made it the highest-impact place to test. With the PM, I agreed on three goals we would measure after launch.

Goal 1

More completed sends

More new users finish a transaction

Goal 2

Fewer support tickets

Fewer people asking support how to send

Goal 3

New users come back

A second transaction, toward the company's 70% retention goal

SOLUTION · 01

Currency Switcher for Send & Swap

You Send, They Receive

Instead of a dialpad Send design with no context, you send and they receive brings how remittance works to the surface: people can decide real-time what their end recipient will receive instead of guessing.

You Send is also constrained by the users' main corridor and this never changes since one user can only be verified in one currency. This makes it adaptable for any starting corridor.

Send is adapted for Swap in Wallet

There are few variations for Send design i.e. adding the live exchange rate to help Afriex standout among its competitor like another African-focused remittance: Sendwave. Users liquidate their dollars into Nigerian Naira in the app to increase asset choices when performing transactions.

Buy Crypto

Make clear why the system did what it did.

Trace candidate cutoff to show quality in process

Sell Crypto

SOLUTION · 02

Put Error and Recovery Text to Reduce Key Dropoff

Real-time inline validation to reduce payment error

Provide global controls. Let teams decide how much AI does.

Card failures were the 2nd biggest dropoff in Send. Rejected cards is irrecoverable. I worked with engineering to build inline validation. Typos and incorrect data were caught while users typed.

Add Bank, Verify SSN Error

Provide global controls. Let teams decide how much AI does.

Linking a bank runs through Plaid and asks for an SSN, which we pass to internal tools to meet anti-money laundering (AML) rules. I added error and recovery text at this step, where people were getting stuck.

Results

After Currency Switcher was launched on Send,

I worked with the PM to bring these data,

Completed transactions went up 150% by the end of the quarter, and send success rate went up 16%. The chart below shows weekly successful sends.
Send-related support tickets dropped 32%.
Send page abandonment dropped 40%.

Role Calibration and SmartScreen became a reference point for Indeed's AI work. Indeed designers mentioned they will adapt our design that pulls rejection feedback loop for its training model.

4w8w12w24w

Since Sep 27, 2021

Current usersResurrected usersNew users

Research

Identifying current Send friction

I ran usability testing on the app with our customers on remote call and usability evaluation. I worked with customer support team to send out SMS and in-app notifications for compensated testing.

Role Calibration and SmartScreen became a reference point for Indeed's AI work. Indeed designers mentioned they will adapt our design that pulls rejection feedback loop for its training model.

A tutorial would have added UX debt, not fixed Send

As the first designer on a team without a design culture yet, I had to make the case for redesigning Send rather than covering it with a tutorial. My hypothesis was that new users didn’t understand how the two switchable numbers on Send relate to each other. For a remittance app, that is the core of the product.

Hypothesis: new users didn’t understand how the two switchable numbers relate.

Validated: in remote tests with 9 users, ~78% guessed incorrectly what the numbers meant.
Decision: a tutorial would add engineering cost and UX debt so we redesigned Send and tested V1 against V2.

Hypothesis: new users didn’t understand how the two switchable numbers relate.

Validated: in remote tests with 9 users, ~78% guessed incorrectly what the numbers meant.
Decision: a tutorial would add engineering cost and UX debt so we redesigned Send and tested V1 against V2.

Role Calibration and SmartScreen became a reference point for Indeed's AI work. Indeed designers mentioned they will adapt our design that pulls rejection feedback loop for its training model.

Through research, we also learned that users felt confused on where to start. This helped guide my design directions towards being clear and labelling numbers being shown while pushing for simplicity to avoid heavy-text use.

Role Calibration and SmartScreen became a reference point for Indeed's AI work. Indeed designers mentioned they will adapt our design that pulls rejection feedback loop for its training model.

IDEATION

Exploring how to simplify send that would work for all corridors: US, EU & Africa

I explored 10+ designs for Send and reviewed them with engineering and the CEO. Each one had to pass one test: could a first-time user tell what every number means without help? Some designs have their own tradeoff below:

Role Calibration and SmartScreen became a reference point for Indeed's AI work. Indeed designers mentioned they will adapt our design that pulls rejection feedback loop for its training model.

Validation

Validating the currency switch through A/B test

Based on team feedback, I selected 2 options above to A/B Test with 10 non-Afriex users. 9/10 successfully completed Send flow in an unmoderated user tests. They recommend having “Send” heading at the top since. 90% of non-Afriex users were still confused of the dropdown switcher in both design.

Part of testing screenshot

MVP

Send MVP was adapted for crypto: Buy, Sell and Swap

Distrust comes from the interface, not the model’s accuracy.

This framework was adopted to reduce our user learning curve with new features. It also bought engineering more time to work on Crypto backend and Binance launch. We launched “Send Crypto” before developing the full product suite that was planned in order to appeal to our younger demograhic.

Why I assumed it: candidate quality was poor, yet nearly 60% of applicants met the job post’s qualifications, and Indeed has about 150 job seekers for every employer. The behavior we saw pointed to distrust: under 20% of EMPs engaged with AI summaries, and most reread every resume. One-click applications and AI-polished resumes make applicants look alike, and creating doubt on automated systems, increasing the risk of abadonment.

That assumption set our scope. We didn’t try to bring in more applicants. We helped EMPs judge the resources they had. We believe the success of the designs will be reflected through metrics like resumes reread per hire, instead of applicant volume. If the pool turned out to be weak, these tools would surface that faster.

Reflection

Utilizing legible data to get buyins for roadmap

Indeed Design suggests we think more about…

I actually proposed the send redesign back in Q2 2020 but my project got scrapped. This project almost didn't see light of day and it affected my design confidence for a long while. I didn’t have enough data or user research to back up “why” it’s worth the redesign for 19K users back then. We didn’t have a product manager so this insight was born from my own frustration - which clearly wasn't enough for a collaborative culture.

I learnt through this process that I shouldn’t be afraid but also should have vetted my “hypotheses” thru testing my assumption first. In that way, I can convince and leverage these data to get buyins from my team while persuading them to value user insights!

More work…

See all work

ENTERPRISE

Indeed

2025

VISUAL DESIGN

Mayv

2020

All work

Proud Cholsaipan

Résume for Hire

Resume for Hire

All work

Résume for Hire

Resume for Hire