I recently interviewed for a Data Scientist role at Airwallex and wanted to contribute my interview experience.
Overall, it was SQL plus a case study. The questions themselves were not especially hard, but the pace was fairly fast, and the case placed more weight on product sense and analytical thinking.
First part: SQL
They gave me three tables, roughly sessions, signup, and merchant.
I remember two questions:
Q1. What is the overall signup rate for the 2023 cohort?
The main task was to join sessions and signup and calculate signup sessions divided by total sessions. You needed to pay attention to the cohort and date filters, as well as how the denominator was defined.
Q2. Rank marketing campaigns according to their KYC submit rate.
You needed to join the three tables along the path sessions → signup → merchant, group by marketing_campaign_url, calculate each campaign's KYC submit rate, and then sort the results.
The SQL was not hard overall. It mainly tested joins, conditional aggregation, distinct counts, date filtering, and understanding the definition of a metric.
Second part: Case Study
The case was about KYC submit rate.
They showed me several time-series charts and asked me to analyze the patterns, possible causes, and how I would validate them next.
One typical chart showed the KYC submit rate staying fairly stable for a period, then experiencing a clear sudden drop or structural break at a particular point, followed by an ongoing downward trend. There was also a brief increase around the end of each year.
My approach was to explain the phenomena separately.
Sudden drop or step change
For a cliff-like decline with a clear point in time, I would first consider a KYC product or UX change, a risk or compliance policy change, a tracking or logging issue, a KYC vendor or provider change, or a sudden change in acquisition traffic at that point.
Next, I would check whether there had been a product release, policy update, or tracking change around that time. I would also segment by country, device, campaign, merchant type, and similar dimensions to see whether all users were affected at once.
Gradual decline
A continuing slow decline looks more like a change in the population or traffic composition, such as a shift in the acquisition-channel mix, a growing share of low-quality signups, a change in the merchant or country mix, or an increase in fraudulent or low-intent users.
I would then examine each segment's KYC rate and signup share over time to distinguish between a decline in conversion within segments and an overall decline caused by a mix shift.
Overall, the SQL felt basic to intermediate. The case's structured analysis was the part that really mattered.
I hope this helps people interviewing for Airwallex Data Scientist roles later.
Discussion
Loading comments…