Diagnose Weekly Session Conversion Anomalies

You work as a Data Scientist for an e-commerce platform. You have a weekly time-series dashboard that spans the last three years. At first the dashboard displays weekly session counts, and in the most recent year there is a pronounced spike around May. Next, the interviewer adds weekly conversion rate plus one other engagement or quality measure—for example average session duration, bounce rate, or orders per session. At that same May spike, conversion rate falls sharply. You are then provided with week-level data and asked to look into it. The dataset includes at least these columns:

• year_week: a year-week identifier that may be stored as text rather than an actual date.
• week_start_date: the first day of the week, when present.
• sessions: the count of sessions that week.
• orders: the number of sessions that converted, or the number of orders.
• conversion_rate: orders divided by sessions.
• shop_type: the merchant or shop category.
• session_duration_bucket: duration grouping such as 0, 0-30 seconds, 30-60 seconds, and 60+ seconds.

Tasks:

1. Offer hypotheses for why weekly sessions spiked around May.
2. Explain why conversion rate might drop at the same time that sessions spike.
3. Describe how you would validate the issue with week-level data, including how you would address the year_week data quality problem.
4. Suppose your analysis shows the session spike occurs only for particular shop types and only for sessions with duration 0 or 0-30 seconds. Interpret that finding and recommend next steps.

Overview: This question tests skill in time-series anomaly detection, conversion-rate analysis, data quality handling (particularly parsing and validating week labels), segmentation by shop type and session-duration buckets, and causal reasoning about traffic composition effects.