Diagnose a dip in approval/conversion rate
Company: Airwallex
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: medium
Interview Round: Onsite
You are a Staff Data Scientist analyzing a time series metric for a payments/fintech product.
**Data:** Daily (or weekly) time series from **Jan 2013 to Jan 2015**.
- Metric shown is a **rate** (e.g., *credit card approval rate* or *conversion rate*).
- There is a noticeable **dip around Feb 2013**, a **small surge around Nov 2013**, and then **sustained growth starting around Jan 2014**.
**Task:**
1) List the most plausible explanations for the **Feb 2013 dip**, **Nov 2013 surge**, and **Jan 2014 growth**, including both **product/business causes** and **data/measurement issues**.
2) Because the metric is a **ratio**, explain how you would decompose and diagnose changes by separately analyzing:
- **Numerator** (e.g., approved applications / conversions)
- **Denominator** (e.g., total applications / eligible sessions)
- Any changes in the **mix** of traffic/users/applications
3) Propose a concrete investigation plan:
- What slices/segments would you break down by (examples: channel, geo, device, risk tier, issuer/bank, new vs returning)?
- What supporting metrics would you pull?
- How would you distinguish **seasonality**, **one-off shocks**, and **true causal impact** from a launch or policy change?
State any assumptions you need to make (e.g., timezone, aggregation cadence, definition of approval, logging stability).
Quick Answer: This Analytics & Experimentation prompt evaluates time-series diagnostics, ratio decomposition (numerator/denominator and traffic mix), segmentation and data-quality triage, and causal inference skills at a staff-level applied analytics abstraction.