Sharing the two interview rounds I recently did for the Airwallex Staff DS role. I was supposed to go on to the HM round afterwards, but the recruiter said the HC might be pushed to next year or moved to another region...
The first round was forty-five minutes. SQL first, then a case about KYC submission rate.
The SQL part gave three tables:
- sessions: session_id, cookie_id, start_timestamp, marketing_campaign_url. session_id is the unique key.
- signup: signup_session_id, signup_date, account_id. This table only contains records that have already signed up.
- merchant: account_id, signup_date, kyc_submit_date, lifetime_volume.
The four questions were:
- Calculate the overall signup rate for the 2023 cohort.
- Rank marketing campaigns by KYC submit rate, where rate = KYC submits / signups.
- For each monthly cohort, find the 5 campaigns with the lowest KYC submit rate.
- Calculate the MoM visitor growth rate for each campaign.
The case after that gave a time series where the KYC submit rate went up from 25% to 35%. They asked what the core hypothesis would be and what data I would pull to verify the increase.
Follow-up: assume I had looked at the data, the increase is real, there is no logging problem, and no obvious segmentation anomaly, and keep digging for the cause. There was a hint in the middle: one part is things the company might have changed internally to get more users to submit KYC, and the other part is external factors the company cannot control that could also make users more willing to submit.
Second round
This round was a single 45-minute business case.
The background they gave at the start was that the US put a 3% tariff on goods imported from Mexico five weeks ago. The question was how to measure the impact of this change on the Airwallex business.
They then followed up on direct and indirect impact. For example, would importers switch to importing from other countries, and how should I think about substitution effects and cannibalization. They wanted me to look at the impact on the whole Airwallex business, not stop at the customers directly affected by the tariff.
Next they changed the condition: assume the tariff stays for the long term and will not be removed soon, so how would I redo the revenue forecast.
Finally they kept pushing: the forecast is now built, but the error between actual and forecast has become 7%, when it is usually only 3%. How would I investigate this gap and find the reason the error got bigger.
At the end I was asked to report to the CEO, explaining how I would evaluate the impact of this tariff and how I would explain the forecast's performance.
Discussion
Loading comments…